AMD market share approaching 25%
AMD result is out. Units grow 18% sequentially. It's interesting to compare my earlier projection on AMD's unit output. It seems it gained some more market share., especially considering the fact that AMD's 2Q06 had an extra week. Revenue grew 32% year/year.
Gartner and IDC reported that PC shipment grew 7% in 3Q06. AMD's market share was 22% in 2Q06, therefore, AMD's CPU market share in 3Q06 is 22*1.18/1.07 = 24.3%. One should note that Intel's revenue growth was mainly from mobile and chipsets.
Given the massive price cuts, AMD did a very good job.
Unlike Intel, who had to sell investments to book profit, AMD had organic growth, its mobile units are up 50%. AMD's gross margin suffered a few points mainly because of transitioning into DDR2. Now that the transititon is done, its margin should improve.
Going forward, with the launch of 65nm parts, AMD's cost will be down substantially.
AMD also indicated that more OEMs will sign up in 2007, presumbaly those Japanese vendors who were excluded by Intel 5 years ago.
On units, AMD will grow 35% in 2006, and 35% in 2007. In other words, a growth of 80%.
posted by Sharikou, Ph. D. @ 10/18/2006 01:29:00 PM
108 Comments:
Psst - you need to quickly edit your prediction 2 blogs down about 50% YoY revenue gorwth or add in the eventual qualifying statement after the fact to make it true....
A quick exercise in sharimath on AMD's results:
A 15% increase in 65nm mix (from 10% in Q2 to 25% in Q3) has eroded AMD's gross margin by about 6%.
Extrapolate to 75% mix in Q4 and the gross margin goes from 50% in Q3 to 26% in Q4 for AMD.
Assume that Kentsfield is twice as effective at killing AMD margin and a 10-15% Kentsfield mix in Q4 shipment and you arrive at AMD operating with 18% gross margin. Not a pretty picture for the green team.
However if Intel can push the Kentsfield mix up to 25-30% in Q1, AMD will post operating losses then and until their K8L comes out with at least a 50% mix which probably won't happen until Q2 2008 (ramp from Q3 2007).
If K8L is a dog, AMD will BK in Q1 2008. If it is equal to Kentsfield at release AMD will post losses in 2007 (Q1, Q2, Q3) and will be able to fend off BK if AMD lays off 80% of it's employees and forces the remaining 20% to hold weekly bake sales.
Heard it here first.....
AMD shares down 12% in after-hours trading.
There is no way you can make this a positive quarter for AMD......look at the huge miss on margin!!!!
Basically, to hold market share, AMD had to give away product at cheap/free prices. A 5-6% shift in margin is huge. Either this is Dell getting product at cost or it is representative that AMD is dumping product. Another quarter like this and AMD will be a $8 stock again.
AMD grew inventory, but reduced days of sales by 3.5%. Not bad. The questions that should be of most concern are the margin reductions, and whether or not AMD Osborned itself with the old Opterons that they are now slashing prices on in an attempt to clear out. Sounds like a writedown risk...
The CC should be interesting, as the market is headed into with a very negative attitude on the report.
So good that they just tanked 14%
Wow!
AMD and INTC just had stock price cross-over. AMD is getting killed on their crap Q3 results.
All those AMD employees dumping stock last week knew this was in the wind. Look for the feds to get involved in this soon....AMD really dropped the ball.
Hector sounds like he is crying in the CC right now....
With AMD stock down by $3.45, looks like the market isn't seeing this quarter in the same positive light.
Factor in Intel's margin # includes flash and chipsets sales (both lower margin) and the AMD margin # is not very good - what will be interesting (but we'll probably never know) was how how much was due to Dell and how much just due to price cut war.
Overall though given the environment AMD and Intel's Q3 numbers were both pretty good. Q4 should pick up for both companies. I don't see a lot changing until 2007 (when AMD's capacity picks up vs Intel increasing Core2 product mix)
OMG:
Take a look at Apple. They are up 6% in after hours and the Apple Intel PC's are selling like mad. Looks like all the market share that Dell lost went straight to Apple where Apple now has about 25% of the MS that Dell has and it is increasing about 20-30% per quarter.
Apple is going to be buying the Christmas ham for a lot of Intel employees this year!!!!
Looking at Q2'06 - Q3'06 sequential net income #'s and gross margin #'s, it looks like price war hurt Intel less than AMD.
And Sharikou before you start blathering about business unit sales, also include the restructuring charge, the inventory writeoff and additional loss incurred in flash segment for Intel... (I don't have the #'s in front of me but I think overall this is close to a net of 0 despite your yapping about $400 Mil)
Oh and also might want to look at tax benefit which helped the AMD net income #'s (I think this was 21Mil) which represents ~15% of AMD's Q3 net income. Considering AMD's Q2 numbers also had an additional tax hit of -18mil, this would put Q3# at ~113Mil vs Q2# at $106mil - kind of scary when AMD claims they increased unit sales by 18%!
"Assume that Kentsfield is twice as effective at killing AMD margin and a 10-15% Kentsfield mix in Q4"
"However if Intel can push the Kentsfield mix up to 25-30% in Q1, AMD will post operating losses"
lol kentsfield will be 3% mix in 3Q 2007.
"Heard it here first....."
What? That you're retarded?
"With AMD stock down by $3.45, looks like the market isn't seeing this quarter in the same positive "light.
- AMD stock crashed after ATI aquisiction for example... the market is dumb.
"A 15% increase in 65nm mix (from 10% in Q2 to 25% in Q3) has eroded AMD's gross margin by about 6%."
How did you conclude that the gross margin decrease was due to 65nm mix increase? Even if that was true, it could never be extrapolated. You have no idea of semiconductor production or ramping, do you?
Looking at AMD's 3Q gross margin - it's still over 50%. Remember how the Inquirer was yelling Dell destroying AMD's ASP, and how some fanboy was estimating 30%-40% in comments here? They simply have no clue.
Also notice the record sales and increased ASP of AMD's mobile processors. That's exactly what I predicted a week or two ago. Intel fanboys were saying the opposite in the comments - again, they have no clue. Meron has better FP/SIMD performance, yes, but it's power & price were honestly disappointing - i.e., it is not for the mainstream.
There are people shouting about AMD's share down. That's because at it WAS too high at $25, not because it's having bad business. Good business performance do not always correlate to good short-term stock performance.
http://www.tgdaily.com/2006/10/18/amd_turion_sales_q3_2006/
"We are facing a bit more competition in the 2P category,"
Can someone tell me what the equivalent change in profit margin was for Intel? They just reported yesterday.
Sharky - HERE is some analysis thats been completely missed by the Wall Street Morons today ... Here it is ... intel is claiming "oh and we think we gained market share vs AMD" ... notice how "sly" this is. WHY? Here is the analysis ... intel's market share last year was X86 (did not include any GX share = apple) ... so now the overall "share" is expanded to include X86 + GX. Why is this important - Obviously ... because AMD DID increase share ... its a "sophisticated view" of market share (previously defined as X86) - so basically intel is doing a Shell game here. Think it thru. (I am Comtechnet and have been posting on yahoo stock message boards for over 10 years now).
15% of AMD's profit was from a tax adjustment - that is what you consider "organic growth"?
"AMD's gross margin suffered a few points mainly because of transitioning into DDR2."
WTF? you really need to share the coolaid with others.... 5.4% is due to DDR2 transition? (I guess this is a few points to you - but to others this is a relative 10% margin drop)
"Remember how the Inquirer was yelling Dell destroying AMD's ASP, and how some fanboy was estimating 30%-40% in comments here? They simply have no clue."
What percentage is Dell of AMD's overall business in Q3? It is not clear whether gross margin drop was due to price cuts or Dell pricing... If it was due to Dell pricing (I don't know if that's true) a 5.4% overall hit in margin means Dell is getting a sweet deal vs other AMD customers.
"AMD takes more makret share"
How do you know this wasn't just AMD stuffing Dell's coffers as part of their agreement. Might want to wait for Gartner #'s to come out.
Also while I'm not the spelling police - you may at least want to spell the title of your blog correctly.
Look at these vultures releasing so much odor spitting out nonsense. You moron hollow heads should realize that Intel cooked the books. 11000 laid off amounts to .6 billion operating cost that did not show. Add the selling of the junk businesses, Intel's real numbers should have been 10cent earning per share and negative growth. AMD did not have to sell any business and did not fire 1 employee to reduce operating cost. Furthermore, building a new fab, signing w/ charters, and buying ATI cost AMD over .5 billion $ operating cost (legal fees, business planning). Considering Intel’s artfully preparing the CC #'s and what AMD was up against (that included business investments), AMD has shown much better result for 3Q than Intel. It’s going to get worst for Intel comes next Q. No one else to fire, no other junk business to sell, and AMD is going to kill Conroe sells w/ 65nm's, 4x4, and AM2 chips more attractive because of K8L (too complicated for you hollow head to understand). 3/4 of server market will belong to AMD next year. Intel is standing on thin ICE on the lake. Pretty soon its own debts weight will crack and Intel sunk to the bottom. AMD’s Dell deal is just the beginning.
silly sharikou, confusing "global PC market" with "x86 microprocessor market". have some logic. do you think the "PC" market includes x86-servers?
LOL EVERONE has seen that the Phd Pretender has been caught with his pants down.
It was a horrible quarter for AMD. THe market has digested it. The fact INTEL recovered says already that the analysts have mulled in conjuction with AMD results.
AMD is in very bad shape going into their 65nm ramp. Margins are pressured, no demonstrated high margin high performance high volume part. K8L is quarters away from generating any revenue. Kentsfield is already shipping and will have mind share and shocket share. K8L is coming very late. If it is good, as it should be as they have had what 8 years to try and better the design. It better be good. If it isn't 40% faster then the current architecture then it will be AMD's version of Northwood to Prescott....
Sorry AMD's glory days are over..
We see what you got in your pants and it aint' anything to be proud of.
Hey Sharikou, did you even realize that the Gartner number is YoY, while AMD's number is sequential? What Gartner is telling is how the PC market grew from Q305 to Q306, and what AMD is telling is how they grew from Q206 to q306. As such, the two numbers cannot be combined (without additional data) to infer anything. To draw any meaningful conclusions, what we want to know is how the total PC market grew from Q2 06 to Q3 06. If it grew more than 18% (which is quite possible considering Q2 is the weakest quarter, and this year it was especially bad), then AMD lost market share. If it is less than 18%, then AMD gained market share?
Get it doctor? It's 5th grade math...
Just how bad is Sharikook's math...
Let's assume Q4 overall chip growth is 0% (for simplicity).
For AMD to go from current 24.3% (shari estimate) to 40% in Q4 would mean an increase in unit sales of nearly 70% in one quarter.
On top of that since "AMD is selling every chip it makes" that would mean it would have to increase capacity by that 70% in less than 1 quarter to hit the 40% market share #.
Yeah I think a 70% unit growth QoQ seems like a pretty reasonable expectation...
I guess the other possibility is the overall market could go down 40-50% by unit volume and it only hits Intel (as they make crap products and all AMD products are always sold first). I hear Q4 is generally a seasonably slow quarter, no?
I have just one question...
Could someone post a link to this $4 Billion in Netburst inventory?
Thanks
Hahahaha. Exactly as I told you, miserable AMD results. I told you it would happen and I told you I'd be back to laugh at you.
Ahahahahaha. Rot in it.
Intel reminds me of GM , AMD of Toyota.
Enumae: "I have just one question...
Could someone post a link to this $4 Billion in Netburst inventory?"
Just look at the Intel financial link in one of Sharikou's earlier blog this week. I just posted a lenghty comment on how that inventory breaks out (only ~1.7Bil of the 4.5Bil is actually finished goods); it'll be interesting if Sharikou actually posts it.
But to answer your question Intel doesn't specifically breakout what % of the inventory is what, but if you look at the financial statement 0.54Bil of the inventory is raw materials which already would take potential P4 inventory below 4Bil.
If you factor in work in progress P4 inventory is much less unless you assume every wafer in progress at Intel right now is P4 and Intel has decided to stop producing all Core2, Core, flash, chipsets, embedded products, itanium, motherboards, wifi chips....and every finished good is a P4 as well.
Keep writing Sharikou...
One blog, and a million Intel fanboys are here to be the first to respond... talking trash.
Intel:
- $4.3 Billion (YES, BILLIONS) in inventory
- Inventory approaching 100 days
- YOY profit dropping like there's no tomorrow
- Selling business, laying off 10k people, morale down the toilet, Intel'ers are looking for jobs elsewhere for "greener" pasteurs (no pun intended)
- 70% is 32-bit junk until year end
- 100MB L2 cache soon to hide their FSB bottlenecks
- No scalibility in Servers
- List goes on and on...
enumae said...
I have just one question...
Could someone post a link to this $4 Billion in Netburst inventory?
Thanks
The link doesn't exist but in Sharikou and Edward's dream :)
I guess this is what you want, can get it from Sharikou previous post as well
http://www.sec.gov/Archives/edgar/data/50863/000115752306009975/a5251620ex99_1.txt
Inventories:
Raw materials 535 496 409
Work in process 2,265 2,331 1,662
Finished goods 1,677 1,505 1,055
------------ ------------ -----------
4,477 4,332 3,126
There are 1.x B is finished product, significant amount are working progress and raw material. Somehow Sharikou and edward dream that all the material and the work in progress are of almost all Netburst architecture product, none are for the coming Kentsfield or existing C2D range.
And i also wonder why they forgot there are other product in intel too, such as PM, Core, chipset, etc. besides that, Just imagine AMD produce all the mid to high end product ... they are going to lose the some sales which come from the low end.
intel has to keep produce a range of product as they are demands from various segment. I feel lucky for the intel share holders that they are not the one the decide the product mix in intel
How much $ did you lose Shariku? Next time use option to hedge dude.. blogs wont help with stock price
"The link doesn't exist but in Sharikou and Edward's dream :)"
What a BS, pointer. Where/when did I say $4B Netburst inventory? Do you understand that $3B~$4B means less than $4B?
In fact, the worth of Netburst inventory at the end of Q1 2007 will almost definitely be much less than the current level, but it might be due more to depreciation than reducted unit volume.
"There are 1.x B is finished product, significant amount are working progress and raw material. Somehow Sharikou and edward dream that all the material and the work in progress are of almost all Netburst architecture product, none are for the coming Kentsfield or existing C2D range."
Mr. Pointer, again, you are pure BS. I NEVER said work in progress are almost all Netburst. I've always said 75% of total worth.
I wonder how many of the $1.6B finished goods are Core 2, though? And how much those $1.6B would have worth before the 50+% Netburst price cut in late September?
"AMD is expected to post 50% year/year revenue growth tommorrow.
posted by Sharikou, Ph. D @ 10/17/2006 02:28:00 PM"
"Revenue grew 32% year/year.
posted by Sharikou, Ph. D @ 10/18/2006 01:29:00 PM"
Once again, you show us your 'genius' prediction skills.
Stop 'fragging' yourself with stupid and unrealistic predictions.
I'd love to see your excuse this time next year and Intel is nowhere near BK.
Here is some fun math:
http://www.theinquirer.net/default.aspx?article=35205
AMD revenue $1.3 bill, profit $134 mill
Intel revenue $8.7 bill, profit $1.3 bill
For every earnded $ AMD used $9.7 and Intel $6.7. Doesn't that mean Intel is more profitable than AMD?
The questions that should be of most concern are the margin reductions, and whether or not AMD Osborned itself with the old Opterons that they are now slashing prices on in an attempt to clear out.
And Intel, Intel fan boy?
The huge amounts of Celeron, P4, PD, Core Solo and Duo they have to dump. Billions of them...
Opteron 1xx Socket 939 is less and 0,5% of AMD production...
Q1 2007 we will see…
Thanks Pointer
I can't believe it but your math is actually right(I think)
:)
It's just a market boosting projection though.
Applying Sharikou-Math (as seen in the Intel BK post directly preceding this) to AMD's numbers and taking into account the cash expenditure that will soon be necessary for the ATi buy-out, one can't help but conclude that AMD wont have any money by the end of the month and is therefore practially bankrupt.
Acording to Intel, price war is over, they cannot lower prices anymore, this quarter, they had their last attempt to kill AMD because AMD is capacity constraint and had issues in ramping production. This next quarter is going to be funny because is possible that if 65nm products prove to be more competitive, specialy Turions and 4X4, before K8L products arrive, Intel is going to have a rought 2007. Comments?
Sharikou, maybe you could explain this, and please be patient as I am not an Economics major :)
Raw materials = $535 Million
Work in progress = 2.265 Billion
Finished goods = 1.677 Billion
1. How do you count/consider the raw materials as junk, its not even a processor yet, it could be Netburst or Conroe?
Shouldn't that be removed from your total $4.477 Billion?
2. If Intel is 83% of the mobile market, and we assume that the mobile market is 30% of all processors made, how does that equate...
30% of Work in progress = $679.5 Million.
30% of Finished Goods = $503.1 Million.
That totals $1.717 Billion (535 + 679.5 + 503.1) for mobile and the subtraction of raw materials.
At the same time we have not factored in chipsets, motherboards, graphics or Flash, I am adding graphics because not every chipset has Integrated graphics. If we assume that all of that equals about 30% of the remaining (4.477 - 1.717 = 2.76) $2.76 Billion = $828 Million.
3. This leaves 40% of the Work in progress and Finished goods to be Server and PC processors totaling (2.76 - .828 = 1.932) $1.932 Billion.
If you assume that Server is 10% that gives you (1.932 * 0.10 = 193.2) $193.2 Million, subtract that and now you have $1.738 Billion.
Now of that $1.738 Billion, your source said that 40% will be P4, 25% Pentium D (90nm), 10% Pentium D (65nm) and 25% Conroe, which equates to 75% being Netbusrt based (as you say junk)...
So if Intel carries a $4 Billion invevtory next quarter only $1.3 Billion would be Netburst (junk).
I understand that I did a lot of assuming, but I think it should be a good reflection of reality of the inventory, or I could be totally wrong... :)
4x4 up and running...
-[Link]-
Revenue grows 9% and margins drop by 10%. There are red bulbs flashing everywhere - seems like K8 is becoming Intel's netburst.
"You moron hollow heads should realize that Intel cooked the books. 11000 laid off amounts to .6 billion operating cost that did not show."
I can assure you Intel does not 'cook the books'. The layoffs are phased over 4 quarters and charges are booked in the quarter in which they occur. This is required by GAAP. Even a "hollow head" knows this simple fact.
http://www.forbes.com/technology/2006/10/18/intel-earnings-amd-tech-cx_ck_1018intel.html?boxes=author
Santa Clara, Calif.-based Intel enacted its largest makeover in two decades this year, ultimately deciding to cut 10%, or 10,500, of its staff by the middle of next year in a bid to save $2 billion next year and $3 billion in 2008. Otellini said the plans are on track and that its staff will be down to 92,000 by the middle of next year. Severance costs ran to $98 million during the third quarter and are expected to hit $125 million in the fourth and $120 million in 2007.
What a BS, pointer. Where/when did I say $4B Netburst inventory? Do you understand that $3B~$4B means less than $4B?
yes, you said 3~4B netburst inventory. i just answer enumae question qualitatively, not quatitively. I refer sharikou and you as the one that think that almost all inventory are of netburst. read on.
Mr. Pointer, again, you are pure BS. I NEVER said work in progress are almost all Netburst. I've always said 75% of total worth.
refer to what you said about 3~4B netburst. There is about 5xxM raw material inventory, thus, finish good + WIP is ~3.9B. If you really meant 75%, you wouldn't have said 3~4B.
Furthermore, I believe you know that intel produce chipset, flash, PM, Yonah, etc, too. The claim i have on you is refering the CPU product, not inclusive of chipsets.
10 intel cpu sold, about 7-8 chipset sold along (guess). so, #4 cannot be zero or else intel will face chipset shortage again. lazy to deduce a number, but it is several hundred of millions (you can actually deduce the rough numbe r here, using chipset ASP of 20 and CPU of 1xx). assume 400M here, then left 3.5B.
I have not counted in flash and the rest of non IA product here yet.
Thus, your statement do give me the impression that you claim the WIP are almost all netburst.
I wonder how many of the $1.6B finished goods are Core 2, though? And how much those $1.6B would have worth before the 50+% Netburst price cut in late September?
I have no idea. But go read the excerpt. I beleive the CFO said the increase was due to ramp. What need ramp? of course new products.
Some short-sighted moron said...
"I can assure you Intel does not 'cook the books'. The layoffs are phased over 4 quarters and charges are booked in the quarter in which they occur. This is required by GAAP. Even a "hollow head" knows this simple fact."
Yea, sure. Believe everything the grand Oz tells you and oh yea, don't look behind the curtain. Please.
Intel is one of the most unscrupulous tech companies on the market, see AMD's long list of grievences against them in thier anti-trust suit case.(and they aren't the only company they are trying to screw over either, Via and Transmeta ring a bell) I wouldn't put anything past Intel if it means to their CEO's, CFO's and various other top staff that they will keep thier jobs and thier pay raises/bonuses and whatever else little perks they can earn. What's a little fudging of the numbers to them if they are willing to try everything they can to cut the throat of thier competition?
In short, go tell your fairytale to former Enron stockholders. I'm sure they'll give you a hand sign you're "number one" alright. Cooking the books ain't nothing new and you'll never hear about until a company gets busted, again see Enron for a shining example of this.
Alright- I'm trying to be as apples-to-apples as possible in the following analysis. Not guaranteed 100% accurate, but better than the original posting.
In comparing Q3 06 and 05, I have removed the memory group from 05 numbers, and Spansion minority interest from the 06 numbers. For 05, I have also added the impact of stock compensation (SFAS123) to the numbers (reduces them). For both, I have eliminated tax benefits, in particular because the 06 number was an extraordinarily large one time event, but wanted to be fair and did the same for 05.
The resultant earnings for AMD, memory business excluded:
Q3 2005 earnings: 116.5M
Q3 2006 earnings: 130.7M
% increase: 12, not bad...
Easier to adjust for, looking at operating income with the memory group pulled out, we find:
Q3 2005 OI: 129M
Q3 2006 OI: 119M
% decrease: 8, not what I would hope for. This is not a good sign with the revenue growth.
I would not paint as rosy a picture as our host. That said, I don't think the report was all bad.
As an investor, I would be most concerned about how management will keep focus on the core business (particularly during a major capacity upgrade and technology shift) while they are trying to digest an acquisition at a cost of just over HALF of their market cap. Technology, strategy, and synergy don't matter if management can't execute because they are distracted.
" This next quarter is going to be funny because is possible that if 65nm products prove to be more competitive, specialy Turions and 4X4, before K8L products arrive, Intel is going to have a rought 2007. Comments?"
Comments? 65nm will be only low-end part for the most of the H1 07, probably even longer if you don't count the K8L. 4x4 will use two 90nm 125W FX'es.
Some delusional conspiracy theorist said
"Yea, sure. Believe everything the grand Oz tells you and oh yea, don't look behind the curtain. Please."
Intel never announced they would cut 10.5k employees by end of Q306(http://money.cnn.com/2006/09/05/technology/intel/index.htm?postversion=2006090518). Besides, if they were truly unscrupulous, they would in fact take the full charge in Q3 when the numbers were going to be unimpressive anyway. This would artifically inflate the profit in Q406 and beyond when they expect things to further improve due to better product mix. The market is kind to those who exceed expectations.
The real short sighted Moron said
"Intel is one of the most unscrupulous tech companies on the market, see AMD's long list of grievences against them"
Please say why AMD is right in their claim vs INTEL's defence which say it was legal? Pick something more tangible. What specific grevience. ANd please provide the specific simple data.. Not some long drawn out accusation that AMD is throwing out.. please...
"What's a little fudging of the numbers to them if they are willing to try everything they can to cut the throat of thier competition?"
Have you heard of Sarbanes-Oxley? Have you noticed INTEL of all companies has had not one issue with backdating of options or any other funny stuff.. What data do you have that INTEL execs are crooked? Intel of companies has some of the most simple and trasparent business.
I guess I'm a little puzzled at how this could be good for Intel and bad for AMD.
As far as I can tell Intel's gross processor revenues for 2006 will be roughly equivalent to what they were in 2004 and well below what they were in 2005. In contrast, AMD's processor revenues for 2006 will be double what they were in 2004. This is the biggest drop in revenue that Intel has had since 2001.
The current margins are the worst that Intel has had since 2001 when they actually dropped below 48%. Intel's margins are down from a high of 61.8% in Q4 05 which is a drop of 20.6%. AMD's margins are down from a high of 58.5% in Q1 06 which is a drop of 12%. It is true that chipsets and flash memory affect margin but this didn't seem to be a problem in 2005 when Intel's average margin was 59.3%.
Secondly, non-processor related losses are not up that much from 2005. These can only be blamed for 25-30% of the drop in margin. It is very unlikely that chipsets are down since Intel has been short. Therefore we can assume that the primary cause is ASP on processors. This matches what AMD has said in the last two quarters about processor ASP being down. This also negates the argument since ASP for processors affects AMD the same as it does Intel.
However, as far as I can tell this is the largest overstock that Intel has ever had. AMD's inventories dropped from Q3 05 by 58% to Q4 05 while Intel's rose 11%. Intel's inventories have risen since Q3 05 which is not normal. Intel's inventories are 54% higher than they were a year ago while AMD's are less than half.
The outlook for Intel going into 2007 is that they have hit bottom and are now growing again. They grew at 8.8% from Q2 06 while AMD grew 10.1%. Even if Intel leaves the pricing where it is now AMD's margins should increase as it ramps capacity on FAB 36 since the costs are much lower on FAB 36. AMD should be back up into the 55% margin within 1 or 2 quarters. Beyond divesting which also costs money there is less that Intel can do during 2007 to reduce costs. FAB 36 will increase capacity until mid 2008 when FAB 38 will be ramping. In contrast, Intel won't have a lower cost ramp until the first 45nm FAB comes online in late 2007. Unfortunately, this would put Intel badly behind in terms of costs since it would only have 2 FABs out of 5 at 45nm while both of AMD's would be ramping 45nm. Intel may follow by converting the other 3 however this would be at a substantial capital cost.
Overall, I see Intel growing steadily from now and during 2007 and 2008. However, it seems most likely that AMD's growth will outpace Intel's. In fact, I haven't yet figured out any way that Intel can grow faster unless AMD stops being able to sell everything that it can produce since AMD's capacity will increase relatively faster than Intel's during 2007. I suppose it is possible that AMD's ASP could drop enough to offset increases in volume however even this would begin to affect Intel as it grows to 25% (most likely by end of 2007).
BTW, AMD's volume share wasn't 22%; it was actually 21.6%. While 25% is incorrect I believe AMD nevertheless had a small increase from 21.6% to 21.9% in volume share.
Not a bad quarter for AMD.
A lot better than Intel's, certainly compared to 2005.
http://www.fabtech.org/content/view/2039/
AMD has been understating its FAB36 ramp up, I wonder why,
Looks like they are positioning themselves to take more market share from Intel.
They already have a higher margin than Intel and are making inroads to Intel stronghold over mobile market. As Scientia from AMDZone noted, 65nm should bring in more savings for AMD allowing them cut prices further if required
So AMD owns 20% of the overall market BUT earns only 10% of what Intel is earning? Those massive AMD price cuts, AMDtard!
Some koolaid drinking fool wrote...
"Intel never announced they would cut 10.5k employees by end of Q306(http://money.cnn.com/2006/09/05/technology/intel/index.htm?postversion=2006090518). Besides, if they were truly unscrupulous, they would in fact take the full charge in Q3 when the numbers were going to be unimpressive anyway. This would artifically inflate the profit in Q406 and beyond when they expect things to further improve due to better product mix. The market is kind to those who exceed expectations."
Blah, blah, blah. Whatever dewd, keep drinking the koolaid, Intel will say whatever it needs to save it's collective ass(i.e. the upper management) in front of it's shareholders. I wouldn't be surprised if it's part of thier upper management's creedo by now. As for it's true market value and how inflated Intel's stock price is, well I guess we won't know until the fit hits the sham will we as I already stated before. (perhaps as per Sharikou says when they hit chapter 11 BK? hah) I guarantee you'll see more layoffs though and that's only going to help Intel's stock so much before it starts damaging them in other ways.(like getting a rep where no one worth thier degree will want to work there, if it's not getting there already)
The market is VERY clear though for any bystander to notice in the long term trend, this is that Intel is shrinking, and the reasons aren't too hard to find.
"Please say why AMD is right in their claim vs INTEL's defence which say it was legal? Pick something more tangible. What specific grevience. ANd please provide the specific simple data.. Not some long drawn out accusation that AMD is throwing out.. please..."
Uhhhh lets see, THEY ARE BEING SUED FOR ANTI-TRUST VIOLATIONS OUTSIDE OF THE US AS WELL? Like in Germany and in Japan??? Transmeta is suing them for copyright infringment??? What the hell other evidence do you need??? It's all over the tech news! If you want the list of AMD's grievences, go read it on AMD's site your own damn self, I'm not going to hold your hand and waste my time explaining them in detail to you when AMD's already done it. Get off your ass and do it yourself, it's already out there and it's not too hard to find.
"Have you heard of Sarbanes-Oxley? Have you noticed INTEL of all companies has had not one issue with backdating of options or any other funny stuff.. What data do you have that INTEL execs are crooked? Intel of companies has some of the most simple and trasparent business."
Please. Again, keep drinking the koolaid and believe everything the Grand Oz tells you. Like I'm going to say again for the UMPTEENTH million time for you simpleton's to understand:
DO YOU HONESTLY THINK INTEL IS GOING TO ADVERTISE IT'S COMMITING VIOLATIONS IN REPORTING HOW IT FUDGES IT'S BOOKS OR CROOKED ACTS IN THE INDUSTRY TO THE SEC OR IT'S OWN SHAREHOLDERS??? If you do believe that I have some wonderful beachfront property in Arizona to sell you.
Here's the long and short of things so maybe you fools will grasp this concept. Intel is only concerned about ITSELF in the marketplace. They have shown time and time again they will use ANY MEANS necessary legally and illegially to further thier means to an end in thier continued dominance and monoply in the industry. For that matter most of the damn corporate world is and has been crooked.(I know because I work in it sadly) What the hell makes you people think that Intel is so innocent and can or have not done any wrong when there are GLARING, HARD EXAMPLES OF IT is my question!?!?!
I'm not worried about Intel going belly-up, I'm worried about *AMD* going belly-up.
The financial reports are gray and fuzzy for both companies, so we have to rely on current and upcoming tech more than anything, and AMD is kinda looking like it's going to be slapped silly this next half year or so.
This competition has been good for all of us. I can get a kickass CPU from either manufacturer for next to nothing. If AMD calls it quits (or Intel for that matter), we're all going to suffer.
You folks need to cheer on both companies, get it straight!
So AMD owns 20% of the overall market BUT earns only 10% of what Intel is earning? Those massive AMD price cuts, AMDtard!
Actually, AMD has close to 22% of the processor market and makes just over 18% of the money. I'm not sure where you got the 10% from.
How many of Intel's patents do you think Transmeta violated? Intel could probably sue a tiny company like Transmeta into bankruptcy. The outcome of the AMD vs Intel trial will take years to come out at a verdict. We all know that part of AMD's case has already been thrown out, that is not good news for AMD.
Intel fumbled around for two years with Prescott and it's awful derivatives. AMD had a commanding lead everywhere but in mobile processors. Yet AMD still didn't gain more than a few % points of marketshare from Intel. That's not Intel's fault, it's AMD's shoddy management.
Intel has the superior product line now (Core 2 Duo for mobile/desktop and Woodcrest for servers) with Quad Core processors due out in Novemeber - 45nm coming next year, with further increases in clockspeed, new SSE4 instructions etc in the 45nm shrinks of Merom and Conroe along with the Yorkfield native quad core design. In 2008 they have the brand new Nehalem architecture that we know very little about, but that will bring even greater performance boosts. Intel has the superior product now, and they're going to take back that market share from AMD, and a whole lot more market share.
Anonymous said...
Some koolaid drinking fool wrote...
blablabla ...
hey, cool down. We know u lose a lot of money in AMD stock .. but rumor and bad mouthing won't help though.
So AMD owns 20% of the overall market BUT earns only 10% of what Intel is earning? Those massive AMD price cuts, AMDtard!
Actually, AMD has close to 22% of the processor market and makes just over 18% of the money. I'm not sure where you got the 10% from.
Subtract the money Intel makes from Chipsets and Flash memory and Intel starts go
The financial reports are gray and fuzzy for both companies,
I'm not sure what part was gray and fuzzy to you. It seemed pretty clear to me. Intel's processor revenue's are flat from 2004 while AMD's have doubled. I can't think of any definition of performance no matter how twisted or convoluted that could say that Intel has performed better than AMD. The best you can say is that in spite of the worse shrinkage by Intel since 2001, Intel is still bigger than AMD.
so we have to rely on current and upcoming tech more than anything,
No. Intel's margins are not quite as bad as they were in 2001 when they dipped below 48%. AMD's margins will rise as their costs drop due to ramping on FAB 36. What (besides magic and arm waving) is going to make Intel's margins rise?
And, AMD has pretty solid future tech in K8L, HT 3.0, Torrenza, and DC 2.0. In contrast, Intel has just started a copycat initiative to Torrenza that won't even be approved for 12-18 months and then another 12 months to hardware. Torrenza could be a long way ahead by then.
Secondly, Intel has no substantial upgrade to its underlying architecture beyond moving to DDR3. Intel is going to be completely outclassed in the 4-way and up server market. Intel will also be outclassed in the quad core 2-way and up market. Intel should remain competitive for dual core 2-way and below. And, you are aware that even if CSI weren't delayed to 2009 that it is less than the current HT 3.0 standard? CSI may be facing HT 4.0 or even 5.0 by the time it is released on X86 processors.
and AMD is kinda looking like it's going to be slapped silly this next half year or so.
Okay. AMD is currently unable to meet demand. How is AMD going to get slapped silly if demand for its chips stays high while its costs drop? Most business analyst with common sense would see that as an ideal situation. The situation for Intel is far less ideal. They have no substantial cost drops until they reach 45nm. And, during 2007 they will have to face K8L plus increasing volume from AMD. Most analysts with common sense would not see that as ideal.
Sharikou:
Have you decided to re-cant your claim that AMD will have 40% market share leaving 2006?
http://sharikou180.blogspot.com
(A more balanced POV)
The outcome of the AMD vs Intel trial will take years to come out at a verdict. We all know that part of AMD's case has already been thrown out, that is not good news for AMD.
You obviously do not see the real strategy. AMD is not counting on a big settlement from Intel. If you read the article in my blog on why this isn't like 2002 I even dispute how much AMD dropped overall. I point out that AMD only dropped a couple of percentage points more than Intel. This might add up to a settlement of $100 Million but that would be trivial for both companies. The real point of the lawsuit is to prevent Intel from throwing its weight around while AMD grows. AMD's legal counsel is estimating that Intel's monopoly will break between 25% and 30%. The lawsuit only has to hold Intel in check until then.
Intel fumbled around for two years with Prescott and it's awful derivatives.
While it also developed Pentium M.
AMD had a commanding lead everywhere but in mobile processors. Yet AMD still didn't gain more than a few % points of marketshare from Intel. That's not Intel's fault, it's AMD's shoddy management.
And, you clearly have no grasp of manufacturing or markets. In 2003, the initial yields of 130nm SOI were not that good. This shows up as a revenue dip as the larger K8 die and lower yield displaced some K7 manufacturing. If you recall, AMD then began an expansion of the cleanroom space in FAB 30. The next year, AMD began work on FAB 36. AMD expanded capacity on FAB 30 by 50% which is extraordinary and even more extraordinary to do while under full production.
K8 processors didn't become the majority of AMD production until Q2 04. The push at this point was to deliver X2 which you may recall AMD did deliver ahead of Intel. AMD did this while also upgrading FAB 30 to 90nm, again while under production.
The next step was 65nm and this is where AMD ran into problems. It was not cost effective to convert FAB 30 to 65nm because it was still used 200mm wafers. Since FAB 36 was 300mm anything new would be 300mm as well and any money spent upgrading FAB 30 with 200mm technology would be wasted. This delayed AMD until early 2006 when FAB 36 was able to start 65nm testing. What you are seeing now has been the result of this. Intel obviously was able to begin 65nm testing much sooner because of its dedicated D1D testing FAB. If you honestly feel that AMD had bad management then explain what you would have done differently. The only thing I've thought of is that if AMD had borrowed money to start building FAB 36 six months sooner that would have helped. However, I don't know if negotiations with Germany could have finished that soon.
Intel has the superior product line now (Core 2 Duo for mobile/desktop and Woodcrest for servers)
Not exactly. There seems to be no difference in the Celeron/Sempron range. And, once Turion is on 65nm there will essentially be no difference with Merom. Intel is ahead (by quite a bit) on the desktop. AMD is not currently competitive with anything above E6400. In servers, AMD is ahead on 4-way and up. On two way and below Intel would be ahead if they werent' being held back by FBDIMM. They need to replace the 5000 chipset as quickly as possible. Fortunately, Intel didn't make the same mistake as before and try to push FBIMM on desktop as it did with RDRAM. The only bad part is that by the time a replacement chipset is released, Woodcrest will be facing a K8L competitor and will still not be ahead. This was definitely a mistake by Intel.
with Quad Core processors due out in Novemeber
At too low of a volume to matter much. This will match up with 4X4 which will also be low volume. And, Clovertown will get trounced by Barcelona.
- 45nm coming next year, with further increases in clockspeed,
Strange because these supposed big jumps in speed don't seem to match up with the low clock speeds and lower FSB speeds for the E1000 series chips in Q3 07.
new SSE4 instructions etc
You overlooked AMD's new instructions.
in the 45nm shrinks of Merom and Conroe
A 45nm die shrink of Conroe doesn't match up with the Q3 E series. I'm wondering if this will only be for the server chips before 2008.
along with the Yorkfield native quad core design.
Which won't help as much with the architecture maxed out with 4 cores on two sockets. This design won't work well with 4-way.
In 2008 they have the brand new Nehalem architecture that we know very little about, but that will bring even greater performance boosts.
This is not likely at all. Intel wouldn't go to the trouble of changing Yorkfield to native quad core only to have it replaced six or nine months later. It seems more likely that the Yorkfield design will be the mainstay until Nehalem arrives with CSI in 2009.
Intel has the superior product now, and they're going to take back that market share from AMD, and a whole lot more market share.
This is not likely at all. AMD's demand exceeds its supply while Intel has still been unable to get rid of its overstock. Presumably you must be assuming that this situation will change either Q4 06 or Q1 07. BTW, if the Woodcrest design is that impressive where are the Woodcrest Supercomputing announcements? There have been three for Opteron. Also, are you expecting Dell to change its mind about AMD along with the two largest retailers in China? Are you also assuming that AMD's stable image image platform and bundled chipset/cpu solutions are not going to gain corporate contracts. Or are you just thinking that AMD will become unpopular because of magic? The current markets look pretty good for AMD. I should also say though that Intel looks good as well and should be able to increase revenue back up to the 2005 level's during 2007.
Two years ago, Intel's yearly processor revenues were 9.8X higher than AMD's. This year they should be about 4.8X higher. This means that Intel's advantage has been cut in half while facing a capacity constrained competitor. AMD's capacity improves all during 2007 and all during 2008. AMD even took 3% share away from Intel during Q4 05 which is the best quarter that Intel has ever had in its entire history. So, what is Intel going to do differently in 2007 to take back share from AMD?
BTW, if you think that 2008 is going to be so rosy then I think you should consider ramping. Both of AMD's FABs will be ramping to 45nm while only two of Intel's FABs will be. It is true that Intel could spend more money and ramp additional FABs. However, how does this square with AMD's recent increases in capital spending compared to Intel's cuts in capital spending?
"The financial reports are gray and fuzzy for both companies, so we have to rely on current and upcoming tech more than anything, and AMD is kinda looking like it's going to be slapped silly this next half year or so."
The problem with the current AMD is that, other than high-end Opterons, it has difficulty to sell chips at price $300 or higher. That is of course because at that price range AMD's K8 cannot compete with Core 2 (E6600 and above).
How much doomed is AMD due to that? Well, you just need to look at the volume percentage those sales represent. 5%, 10% maybe?
Going to 2007, Intel is pushing multi-core aggressively. But what is multi-core, after all? For Intel, it's just MCM-packaged SMP less memory controller. Seeing how strong Opteron is against Woodcrest at 4/8 sockets, I'd say at 8 cores or above Intel's Core 2 will have hard time to compete on power, performance, and price (incl. northbridge) against even K8 Rev.F, let alone Rev.H (K8L).
If you understand technically that Core 2 is FSB starved past quad cores you will see that the only advantage Intel has in 2007 is just its 45nm transition 6-12 months earlier than AMD.
The situation for Intel is far less ideal. They have no substantial cost drops until they reach 45nm.
Actually, 50% of production is at 65nm now. Assuming a nominal 30% cost reduction moving from 90nm to 65nm, if 100% production ison 65nm in 2 quarters, that would be a 15% cost reduction in overall wafer cost. Add in the increased dpw at higher ASP, and ...
Seeing how strong Opteron is against Woodcrest at 4/8 sockets
What a baseless statemet! Anyways, what can anyone expect from Edword...
Do you even realize that Woodcrest is 2P, and hence there is no data as to how it competes in 4P or 8P?
Additionally, Woodcrest is native dual-core, and hence those two cores look like single load to the FSB. Time and again Intel has shown that its single FSB is good enough for 2 loads. This implies, Clovertown won't be bandwidth starved. What's more, going forward, Intel can play the Dempsy trick--add an internal FSB to make the 4 cores look like a single load. At that point, you can put two of these 4 core processors onto a single FSB without any problem. That is 8 cores. Go, crwol back into the hole where you come from.
AMD retards keep on bringing up FSB as if that is the only architectural feature that matters. Go and educate those brilliant AMD engineers on prefetchers and memory reordering or how to make hyper-threading work. The bottom line is, AMD has a couple of macro-architectural wins, but overall, micro-architecturally, Intel is far superior. And Woodcrest has proven this hands down (and soon Clovertown will be proving it again).
AMD said net income was up 51 percent year over year, but that profit margins plummeted 5.4 percentage points to 51.3 percent, according to a Reuters report in the Los Angeles Times.
Despite the gains, Wall Street wasn’t happy at all, according to Forbes.com, which wrote “Pretty good just isn’t going to cut it any more for Advanced Micro Devices.”
The story went on to say AMD’s “easy gains are over” in the face of increased competition from Intel.
There’s been an increasing amount of discussions in tech sites about alleged problems AMD has had recently supplying its customers. The British tech site, Hexus.net has a smack in your face opinion entitled, “AMD helps Intel get back its market share.”
Author Steve Kerrison said, “AMD appears to be almost completely neglecting its consumer market in an effort to make inroads into the corporate and enterprise market. The cost to AMD could be catastrophic.” He wrote that one customer, MESH Computers, had to send out an apology to customers because it couldn’t deliver promised AMD systems.
I see good times for Intel ahead and BAD times for AMD. The AMD Dell deal and ATI merger will only continue to degrade AMD.
>> Actually, AMD has close to 22% of the processor market and makes just over 18% of the money. I'm not sure where you got the 10% from.
Who's talking about GROSS? Minus all those expenses and we arrive at NET INCOME.
Tell me, how is AMD gonna pay ATI shareholders? Also what about that 2.5 billion loan interest payments which amounts to roughly 200 million a year?
Not exactly. There seems to be no difference in the Celeron/Sempron range. And, once Turion is on 65nm there will essentially be no difference with Merom. Intel is ahead (by quite a bit) on the desktop. AMD is not currently competitive with anything above E6400. In servers, AMD is ahead on 4-way and up. On two way and below Intel would be ahead if they werent' being held back by FBDIMM. They need to replace the 5000 chipset as quickly as possible. Fortunately, Intel didn't make the same mistake as before and try to push FBIMM on desktop as it did with RDRAM. The only bad part is that by the time a replacement chipset is released, Woodcrest will be facing a K8L competitor and will still not be ahead. This was definitely a mistake by Intel.
At too low of a volume to matter much. This will match up with 4X4 which will also be low volume. And, Clovertown will get trounced by Barcelona.
A 65nm shrink of Turion 64x2 aka "Tyler", will not be enough to bring performance levels to Merom, as it is still hindered by the fact that it is a K8 Derivative (Rev.G), once we reach "Bulldozer" is when we will see the mobile market heat up in performance from the AMD camp.
Regarding desktop I agree, Intel will remain ahead for the time being, and is unlikely to be dislogdged from it's position until Rev.H parts "Antares" to go against the updated Conroe E6x50 Series.
Woodcrest will likely remain of niche of it's own, as there aren't any plans beyond Barcelona for Servers that I am aware of, AMD looks to be pushing Multi Core Rev.H only on the server front, which is fine as Rev.H is ging to be their top brand. What will be interesting is to see how Barcelona stacks up against Clovertown, with it's Native Quad Core and Rev.H tweaks I expect that it should be ahead, (and I would be disappointed if it weren't) but that is not guranteed until we see the final performance figures.
Regarding Clovertown, it will have around 2 Quarter advantage over AMD's Quad Core, so it will be interesting, and at reasonable pricing 455US - 1172US. Consider it is servers, I don't think this is cost prohibitive at all. Clovertown can't get trounced for the time Barcelona doesn't in retail.
For desktops Quad Core will remain a high end SKU, as prices start at 851US or above.
Strange because these supposed big jumps in speed don't seem to match up with the low clock speeds and lower FSB speeds for the E1000 series chips in Q3 07.
Your not utilizing valid information for comparison, the E1xxx Series is Single Core and is still based on Conroe technology so hence is on 65nm. The 45nm shrinks are going to be utilized for Core 2 Duo and Core 2 Quad first, just like how for the Single Core market we are still utilizing NetBurst technology on the desktop.
The big jumps in clockspeed are slated for the performance SKU's with Wolfdale/Ridgefield as a replacement for Allendale/Conroe as well as Yorkfield as a replacement for Kentsfield.
Which won't help as much with the architecture maxed out with 4 cores on two sockets. This design won't work well with 4-way.
The design currently doesn't exist in the MP sector, you have no idea how Core based technology scales in 4-way configurations.
This is not likely at all. Intel wouldn't go to the trouble of changing Yorkfield to native quad core only to have it replaced six or nine months later. It seems more likely that the Yorkfield design will be the mainstay until Nehalem arrives with CSI in 2009.
It depends on how you look at things, keeping to schedule and assuming Nehalem arrives in H2 2008, Yorkfield is arriving in Q3 2007, so that gives that product about a year's lifetime before Nehalem arrives. That is sufficient lifetime for an SKU and gives Intel practice for Native Quad Core, so the transistion to Nahalem won't be as complex. So it's more like 1 year later, and hell, it also serves to test out the 45nm process on existing technologies, which is very normal for Intel. As well it's not out of the ordinary for Intel to replace SKU's fairly quick, look at Yonah it has been replaced in 6-7 months after it's introduction. As well as Presler.
This is not likely at all. AMD's demand exceeds its supply while Intel has still been unable to get rid of its overstock. Presumably you must be assuming that this situation will change either Q4 06 or Q1 07. BTW, if the Woodcrest design is that impressive where are the Woodcrest Supercomputing announcements? There have been three for Opteron. Also, are you expecting Dell to change its mind about AMD along with the two largest retailers in China? Are you also assuming that AMD's stable image image platform and bundled chipset/cpu solutions are not going to gain corporate contracts. Or are you just thinking that AMD will become unpopular because of magic? The current markets look pretty good for AMD. I should also say though that Intel looks good as well and should be able to increase revenue back up to the 2005 level's during 2007.
I would say Intel has a good chance of regaining marketshare in the Q4 2006, and Q1 2007 and likely even Q2 2007, as Bareclona won't be a large percentage of the product mix at that point not enough to have much impact. Beyond that it is hard to say. We will have to see what AMD's schedule is for transitioning to Rev.H based SKU's as well as what Intel's product mix will be like at that point in time.
BTW, if you think that 2008 is going to be so rosy then I think you should consider ramping. Both of AMD's FABs will be ramping to 45nm while only two of Intel's FABs will be. It is true that Intel could spend more money and ramp additional FABs. However, how does this square with AMD's recent increases in capital spending compared to Intel's cuts in capital spending?
Sort of, there are more then just the 2 factories of Fab 28 and Fab 32 which will have access to production level 45nm technology. Fab D1C will also be changing over to 45nm in 2007, this is all in addition to the development Fab D1D for 45nm technology.
As well since 2 of these are new fabs alot of 65nm capacity is retained, as you still have Fab 12C, Fab 24-2 & Fab D2 on the 65nm node.
"What a baseless statemet! Anyways, what can anyone expect from Edword...
Do you even realize that Woodcrest is 2P, and hence there is no data as to how it competes in 4P or 8P?"
So Opteron is strong against Woodcrest on 4P and above, 100% to 0%, because Woodcrest isn't available. Why isn't, then? Because Woodcrest doesn't work well on 4P and above!
Your arguments on FSB "sufficiency" and dual-cores working as "single load" are pure laughable. Just because you read it somewhere on the Net doesn't mean you can use your imagination to extend/expand it. I'm too lazy to teach you, but you should really take more computer architecture course before you comment technically.
Scientia said: "The real point of the lawsuit is to prevent Intel from throwing its weight around while AMD grows. AMD's legal counsel is estimating that Intel's monopoly will break between 25% and 30%. The lawsuit only has to hold Intel in check until then."
Scientia, great comments.
But really, it's kinda useless to teach fanboys, who can know everything about Intel's roadmap and plans and fabs, yet unable/unwilling to grasp the points in your comments or to understand technology.
"If you understand technically that Core 2 is FSB starved past quad cores"
I thought it was going to be starved with quadcores - is this no longer the case? (or do you mean 4 socket?). AFAIK there is no 2007 roadmap for octo-core products (especially considering AMD will only be doing quad core ~Q2'07).
Perhaps we will see Intel do an MCM solution with 2 native quad cores but I don't see this until after they start conversion to CSI&IMC (although maybe they'll have an extreme edition desktop part to keep enthusiats happy)
"What (besides magic and arm waving) is going to make Intel's margins rise? "
To suggest a few:
- Completion of 65nm transition on CPU's (which is ~50%-60% right now)
- Reduction in expenses overhead (incl layoffs), this will be a one time hit short term (Q4'06/Q1'07) but will improve margin long term - Intel has said 2B in 2007 improvement but I would suspect this would be closer to $1-1.5Bil.
- Yield improvements. Sharikou, et.al have said yield is <50%; while this is a bunch of crap there will be some marginal yield improvements over the course of the technology life.
- Vista (if it actually does take hold in 2007); this will potentially drive desktop(and mobile?) to a higher end mix. Personally I think Vista adoption will be low and this impact will be minimal.
- Flash divestiture? If this happens this will have the biggest impact on margin. (while Intel's CPU:flash ratio is different, just look at AMD's gross margin before and after Spansion spinoff)
Did you want any other ideas?
Scentia, the answer to your biased statement is this... AMD is a corrupt and shitty company as history illustrates...
AMD sued Intel for "monopoly" on x86 products.. to get rights to reverse engineer Intel's 486 processor
AMD complained to FTC on Intel/DEC "monopooly" on technology. In exchange they get free license for EV7 and EV6. Guess what? EV7 became Hypertransport and EV6 becomes DDR IMC.
Innovation? Intel literally "owns" your Hypertransport and IMC.
And now they are suing Intel on "monopoly" again?
Scentia, get to know your AMD company better before you start praising.
"I thought it was going to be starved with quadcores - is this no longer the case? (or do you mean 4 socket?). AFAIK there is no 2007 roadmap for octo-core products (especially considering AMD will only be doing quad core ~Q2'07)."
It depends on the workload, whether quad core is going to be FSB starved. For servers, it's very likely the case; for desktops, it depends obviously on how much memory bandwidth the applications require. If you edit a 20MB picture, in very high resolution, for example, and the program utilizes multi-threads or processes, then FSB is very likely the bottleneck. If you transcode a 1MB MP3 to WMA, though, the FSB is rarely used once the file is read into the cache.
When I said "past quad cores" I really meant quad core per socket with 2 sockets or more. When CPU goes quad-core, Core 2 will start having difficulty to compete with K8 on 2P or more.
The best thing that can happen to Intel is that it ramps Core 2 fast and starts 45nm quickly enough in 2007 to lower cost and push AMD's profit down, and at the same time releases a good CSI-capable Core 2 in early 2008. It might still be able to sit on its quasi-monopoly if the execution above is well accomplished.
Personally I believe Intel has engineers working day and night right now to push CSI out, together with a revamped PCI-Express, as early as possible. Some said it's pushed back to 2009, but I think 2008 is more possible, or even late 2007. It will be yet another great product release.
I am sure if that happens, at that point all those Intel fanboys who's been crying "FSB is nothing wrong but great" will flip their positions accordingly. It is just the nature of fanboism - they choose side, not logic, in their arguments.
"AMD sued Intel for "monopoly" on x86 products.. to get rights to reverse engineer Intel's 486 processor"
Man... you got your facts totally wrong. AMD's last lawsuit against Intel was not what you described above at all. It has nothing to do with monopoly, but license. You should go back and read history better.
"AMD complained to FTC on Intel/DEC "monopooly" on technology. In exchange they get free license for EV7 and EV6."
Again you have the facts totally wrong. It was DEC who filed suit against Intel and the latter agreed to buy part of DEC, with FTC insisting DEC to license its technology to at least two other companies (including AMD).
"Guess what? EV7 became Hypertransport and EV6 becomes DDR IMC."
You got your facts wrong yet again. While Hypertransport is indeed based on EV7, the former adopts broadcast-based cache coherence while the latter is directory-based. There are other differences, too, but they are details.
EV6, OTOH, was not the source of IMC, which has been used since Alpha EV4 and IBM Power4. EV6 was the source of socket-A bus protocol. Socket-A is not an innovation, but it's a fine engineering; no other company other than AMD in the industry ever adopts and extends technology so well.
"Innovation? Intel literally "owns" your Hypertransport and IMC."
Intel "owns" what? Socket-A bus protocol? ccHT? Or the plain HT which is already an open standard?
And you call Scientia biased when yourself had all these facts screwed? Way to go, man....
The reason that AMD is going to reach cross-over to 65nm faster while ramping Fab36 is because they announced in their Q306 analysts call that during in Q207 that they're are going to be reducing wafer starts in Fab30 to the tune of 10k wafers less per quarter until the conversion to 65nm is completed in Q208. At that time, Fab30 will be renamed Fab38. During this transition, Fab30 will dip to 40% capacity.
Now read this clarity and comprehension: A reduction of 10k-12k starts per quarter pans out to 40k-48k fewer wafers started per year. This clearly throws the proverbial wrench in Sharikou's prediction that AMD will exit this year with 30% and next year with 40% because AMD will not have the capacity to do so based on factory availability, wafer starts, larger die sizes for their quad-core, normalized yield losses seen in transitions from technology to technology (i.e.: 90nm - 65nm).
In fact, it is my belief that AMD's drop in stock price this past week isn't the result of bad news for the previous quarter. On the contrary, I firmly believe that the drop has to do with AMD's coming quarters and inability to meet demand due to capacity constraint. From the transcripts of the analyst call it clearly sounded that the reduced wafer starts of that scale caught the analysts by surprise.
Read the transcrips for yourself. It is hard to counter this point when it came from Dirk Meyer (AMD COO) at the analysts call with Hector Ruiz (AMD CEO) sitting beside him.
I'm glad to say that not everyone here is biased and blindly supports Intel. No one should just blindly support AMD either. Right now even the E6400 is superior to X2 if you really need SSE speed.
As far as Tigerton goes, I wrote a new article on this in my blog. Basically, with Tigerton you get the equivalent of 533Mhz FSB for Conroe or 266Mhz FSB for a single core. Considering that 266Mhz takes you all the way back to PIII and Athlon 2600+ that isn't saying much. Memory is going to be an issue for Intel.
At too low of a volume to matter much. This will match up with 4X4 which will also be low volume. And, Clovertown will get trounced by Barcelona.
Definitely, 4x4, EE, and FX chips don't drive the market. Also, not only will Clovertown get trounced but Tigerton will too on the Caneland platform.
A 65nm shrink of Turion 64x2 aka "Tyler", will not be enough to bring performance levels to Merom
It doesn't need to. There isn't much demand for SSE performance on laptops which is Merom's main strength. Besides, I'm sure you didn't overlook the 50% increase in demand for Turion in the third quarter. Demand should stay strong as AMD moves Turion to 65nm.
Regarding desktop I agree, Intel will remain ahead for the time being, and is unlikely to be dislogdged from it's position until Rev.H parts "Antares" to go against the updated Conroe E6x50 Series.
Agreed, the desktop should be at parity at all ranges in Q3 07.
Woodcrest will likely remain of niche of it's own, as there aren't any plans beyond Barcelona for Servers
No. K8L will also be released as a dual core part for servers. I don't know the name right off hand.
that I am aware of, AMD looks to be pushing Multi Core Rev.H only on the server front
No. Altair will be released in Q3 07 as quad core for desktop.
Barcelona stacks up against Clovertown, with it's Native Quad Core and Rev.H
This is a waste of time. Barcelona's competitor will be Tigerton.
The design currently doesn't exist in the MP sector, you have no idea how Core based technology scales in 4-way configurations.
Not well on Caneland.
assuming Nehalem arrives in H2 2008, Yorkfield is arriving in Q3 2007,
I don't think so. I think the native quad core Tigerton will be Intel's main product all during 2008. I think the CSI socket will show up first with Itanium in late 2008 and not for X86 until 2009. This would also be consistent with the Geneseo and FSB initiatives.
I would say Intel has a good chance of regaining marketshare in the Q4 2006, and Q1 2007 and likely even Q2 2007
Not Q4 06. It's still too early on the C2D ramp. However, I think Intel could regain some revenue share in Q1 07 and Q2 07 but I doubt the volume share will change much because AMD will be seriously ramping in 2007.
Fab D1C will also be changing over to 45nm in 2007, this is all in addition to the development Fab D1D for 45nm technology.
D1D doesn't matter; it isn't a production FAB. I guess D1C could help but that still puts Intel behind. Intel can't compete with three FABs versus two unless Intel is planning to let AMD have 40% of the volume share.
As well since 2 of these are new fabs alot of 65nm capacity is retained, as you still have Fab 12C, Fab 24-2 & Fab D2 on the 65nm node.
It wouldn't surprise me if AMD announced construction for a NY FAB in 2007 which would be ready about 2009. I think it is highly likely that not all of the current 300mm FABs will be upgraded to 45nm.
I would agree that Intel's 65nm ramp with C2D will help some but only about 1/3 - 1/2 of the cost reduction that AMD will get. Remember AMD gets cost reduction both from 65nm and from 300mm whereas Intel is already at 300mm. The divestments are good but their effect won't show up for several quarters. This could make Intel's late 2007 margins look much better. On the other hand, the cost of maintaining a dedicated R&D FAB is probably going to become a detriment during 2008. Using APM, AMD will be able to run tests on any of its FABs and immediately move production to the others. Intel will not be able to do this.
Now read this clarity and comprehension: A reduction of 10k-12k starts per quarter pans out to 40k-48k fewer wafers started per year.
I'm sorry Sheep but you misread the transcript. FAB 30 does 30K wspm. That is 90K per quarter or 360K per year. FAB 30 will have a total reduction of 18K wspm.
This reduction won't matter. By mid 2007 FAB 36 will be at 15K wspm. Since it is 300mm versus FAB 30's 200mm this would be equivalent to 30K wspm or equal to FAB 30's total output. However, FAB 36 will also be at 65nm by then so this doubles again to twice FAB 30's total output. Since FAB 36 would be 67% of AMD's total capacity the reductions for FAB 30 won't matter. Also, Chartered will be producing 65nm by then.
One final thought however. Intel just cut back on capital expenditures while AMD increased its. Less cap ex now means less production capacity in the future; more cap ex now more means production capacity in the future. I would say that Intel is planning to lose marketshare and I would say that AMD is planning to gain it.
Scientia from AMDZone said...
"Besides, I'm sure you didn't overlook the 50% increase in demand for Turion in the third quarter."
I think your making the numbers look better than they are.
50% increase of a segment that grew about 25-30%, in which they (AMD) hold only 13%.
I looked but it was not clear, is Turion AMD's only mobile processor (example... PentiumM, Core Solo, Core Duo and Core 2 Duo)?
It doesn't need to. There isn't much demand for SSE performance on laptops which is Merom's main strength. Besides, I'm sure you didn't overlook the 50% increase in demand for Turion in the third quarter. Demand should stay strong as AMD moves Turion to 65nm.
Merom strength is that it outperforms Turion 64x2 overall, not just in SSE performance, AMD will need more then a 65nm shrink of Turion 64x2 to become competitive in performance. Anyway in Q3 2006 Merom wouldn't have had that much of a chance for an impact anyway, we will need to see Q4 2006 to see how the mobile sector does with Merom out in full force.
By the time Tyler comes out on 65nm Intel will have the Santa Rosa Centrino platform ready to give the mobile sector a nice boost, since Tyler is schedule for Q2 2007.
Definitely, 4x4, EE, and FX chips don't drive the market. Also, not only will Clovertown get trounced but Tigerton will too on the Caneland platform.
You quoting yourself here so I don't know what to say to that. Not to mention your making alot of assumptions about Bareclona's performance despite the fact that all we have now are tech sheets on these processors.
No. K8L will also be released as a dual core part for servers. I don't know the name right off hand.
There is currently no indication that Rev.H will have a Dual Core part for Servers when Bareclona launches. Sometime after that perhaps but not at Barcelona's launch.
No. Altair will be released in Q3 07 as quad core for desktop.
You misinterprted my statement, I am saying that for the moment AMD looks to be pushing Quad Core Rev.H only for the Server front due to the distinct absense of any Dual Core Rev.H Cores for Servers. I don't see any information otherwise that conflicts with this. I am aware that there will be Rev.H parts slated for Desktop in both Dual Core and Quad Core form in Q3 2007.
http://www.amd.com/us-en/assets/content_type/DownloadableAssets/PhilHesterAMDAnalystDayV2.pdf
Take a look at that, slide 12 of 46 states that there will be Quad Core for Desktop and Servers while only desktop will recieve the Dual Core variant. So if you have any new roadmap information that there will be a Dual Core Rev.H part in Mid 2007 please show it to me, I will not simply take your word as sufficient.
This is a waste of time. Barcelona's competitor will be Tigerton.
No I disagree it isn't, Barcelona since it is a Socket F SKU will be slated for both the Opterton 2xxx and 8xxx Series, so both Clovertown and Tigerton will be viable competitors. Nevetheless I would like to see how Barcelona handles both of Intel's Core based products in both of these segments
From the looks of things the Clarksboro Chipset of the Caneland platform allows a dedicated link between each Quad Core processor, and the chipset quite interesting.
http://www.vr-zone.com/index.php?i=4186
Intel demos 4P Tigerton Server
http://news.zdnet.com/2100-9584_22-6128125.html
Not well on Caneland.
And you know this how? Core architecture shows no indication of being bandwidth bottlenecked at this time.
I don't think so. I think the native quad core Tigerton will be Intel's main product all during 2008. I think the CSI socket will show up first with Itanium in late 2008 and not for X86 until 2009. This would also be consistent with the Geneseo and FSB initiatives.
If your talking about MP servers 4P+, I would agree that it is likely that Tigerton will remain the product for MP servers, however for Desktops where the FSB isn't much of a limiting factor if at all, Nehalem is likely to arrive in H2 2008. Yorkfield is replacing Kentsfield which is a Desktop product in Q3 2007 and Harpertown will be replacing Clovertown as the 2P segement product.
Not Q4 06. It's still too early on the C2D ramp. However, I think Intel could regain some revenue share in Q1 07 and Q2 07 but I doubt the volume share will change much because AMD will be seriously ramping in 2007.
I think Q4 2006 has a good chance of getting some marketshare back. I expect continue improvement throughout the holiday season. This is the 1st Quarter Core 2 Duo will be avaiable for the entirety of the Quarter vs only 2/3 or 1/3. As Woodcrest, Merom and Conroe continue to saturate their intended markets, I expect the situation to improve for Intel.
D1D doesn't matter; it isn't a production FAB. I guess D1C could help but that still puts Intel behind. Intel can't compete with three FABs versus two unless Intel is planning to let AMD have 40% of the volume share.
While D1C is about the same size as AMD's Fab 36 and future Fab 38, Fab 28 and Fab 32 are both larger then AMD fabs so that should help offset some of volume share. You also have to consider Intel would have much more 45nm capacity by the time AMD ramps up.
It wouldn't surprise me if AMD announced construction for a NY FAB in 2007 which would be ready about 2009. I think it is highly likely that not all of the current 300mm FABs will be upgraded to 45nm.
We will have to see on this, I wonder where AMD is going to be getting the money from for such a project after buying ATI.
I would agree that Intel's 65nm ramp with C2D will help some but only about 1/3 - 1/2 of the cost reduction that AMD will get. Remember AMD gets cost reduction both from 65nm and from 300mm whereas Intel is already at 300mm. The divestments are good but their effect won't show up for several quarters. This could make Intel's late 2007 margins look much better. On the other hand, the cost of maintaining a dedicated R&D FAB is probably going to become a detriment during 2008. Using APM, AMD will be able to run tests on any of its FABs and immediately move production to the others. Intel will not be able to do this.
That's only if your talking about Fab 38, when it comes online with both 65nm production and 300mm wafer processing, and that is if it can process the amount of wafers necessary to surpass old 200mm wafer production right away. The effect won't be immediate. Fab 36 is already on 300mm wafers so gets the normal savings for going to the 65nm node. Having a dedicated fab allows you more dedicated leeway to make changes and tests without affecting production.
scientia said ... As far as Tigerton goes, I wrote a new article on this in my blog. Basically, with Tigerton you get the equivalent of 533Mhz FSB for Conroe or 266Mhz FSB for a single core. Considering that 266Mhz takes you all the way back to PIII and Athlon 2600+ that isn't saying much. Memory is going to be an issue for Intel.
i'm quite surprised by you all that like to use the direct divide method to get the FSB freq availabilty per core. Have you all heard of trunking or the Erlang unit?
To elaborate this, at any given time, we have to consider how many core needs to access the FSB and what is the probability. The larger and better cache surely reduce the FSB accesses need. In simple and straight forwards word, most of the time, the availability of the FSB freq to a given core is surely much more higher than the simple divide by number of core calculation. For the 4 cores instance, a divide by 2 is more realistic.
Edward said ... CSI ... I am sure if that happens, at that point all those Intel fanboys who's been crying "FSB is nothing wrong but great" will flip their positions accordingly. It is just the nature of fanboism - they choose side, not logic, in their arguments.
BS. Most people has never defied the need of CSI at later stage and the real stetement is that for most current desktop or laptop or some server usage, the FSB is still sufficient. It is and will not be sufficient for high MP system. It is not sufficient for whatever desktop/laptop application that need VERY HIGH FSB access which i do not know of any example or the number or even the how frequent user would use such apps.
When you argue with other poster on the fuxed cost stuff, one statement stand out: the dominant part stand out and you claimed fixed cost is much dominent thatn the varaible one (not necessary i agree with your whatever other statements, but i have no comment there because i dunno). Same case here for the deskptop/laptop usage here. It is the same too for some server apps.
AMD fanbois likes to distort the statement and exaggerate the possitive side of AMD, and espeically the case for the current situation when Intel has the overall lead.
scientia said ... One final thought however. Intel just cut back on capital expenditures while AMD increased its. Less cap ex now means less production capacity in the future; more cap ex now more means production capacity in the future. I would say that Intel is planning to lose marketshare and I would say that AMD is planning to gain it.
What about looking at it differently? Intel current capacity can provide about 90% (guess) of the 2007 IA market and AMD target 40% (guess) capacity. Why should Intel spend extra money to make its capacity towards or even ever the 100%? If intel plan it that way, Intel will surely has excess capacity and just lose money out of it.
The 2007 will be an interesting year where supply will surely exceed demands and either one or even both will be surely hit by the excess capacity. Anyway, another great year for cheap CPUs! :)
"However, FAB 36 will also be at 65nm by then so this doubles again to twice FAB 30's total output."
Scientia - your analysis overall is pretty good but please don't fall directly into the Sharikou 2X 90nm-65nm coversion hole. In thoery you have 70% linear scaling and 49% area scaling but in practice AMD won't see this entire benefit. If you look at IBM's reported SRAM cell size scaling I think 65nm is on the order of 35-40% smaller than 90nm (logic should behave similarly to SRAM scaling). So while it will still give a significant improvement, it won't quite be 2X (maybe ~1.6X)
"D1D doesn't matter; it isn't a production FAB."
Actually it does both prodcution and development (production on 65nm, development on 45nm). It is not on the order of Intel's other's fabs in terms of output, but it will be substantial (~10KWSPM); note this is 1/2 of F36 fab capacity.
"It wouldn't surprise me if AMD announced construction for a NY FAB in 2007 which would be ready about 2009"
AMD themselves has said that production will ramp over 2010-2012. (which likely means ground breaking in 2008). Also keep in mind I believe AMD said they expected to start making either chipset or GPU's in house (I think they said starting on 65nm in 2007 - can someone confirm this). If so not all of AMD's fab capacity will be CPU.
"One final thought however. Intel just cut back on capital expenditures while AMD increased its. Less cap ex now means less production capacity in the future; more cap ex now more means production capacity in the future."
This could mean a myriad of things - cheaper tooling, output improvements leading to less tools, push of some ramp wafer starts from Q4'07 to Q1'08, better than expected reuse betweeen 65nm and 45nm equipment, lower expenditures on research tooling.... You also might want to look at capex ratio between the 2 companies. All this means is spending in 2007 is down.
"This reduction won't matter. By mid 2007 FAB 36 will be at 15K wspm. Since it is 300mm versus FAB 30's 200mm this would be equivalent to 30K wspm or equal to FAB 30's total output. However, FAB 36 will also be at 65nm by then so this doubles again to twice FAB 30's total output. Since FAB 36 would be 67% of AMD's total capacity the reductions for FAB 30 won't matter. Also, Chartered will be producing 65nm by then."
ust keep in mind you are completely ignoring impact of greater mix of products moving to dual core (and small amount moving to quad core). K8L is also bigger die size than K8 for a given tech node too...
pointer said: "BS. Most people has never defied the need of CSI at later stage and the real stetement is that for most current desktop or laptop or some server usage, the FSB is still sufficient."
FSB has been the bottleneck since the end of P-III. This bottleneck is the sole reason that Intel had to use RDRAM and DDR2 3 years earlier than AMD.
Highly clocked FSB with large cache and memory prefetch is suitable for the kind of benchmarks where you get some data (possibly high-rate), process them heavily but sequentially, and then be done with them. It suffice to say that such a scenario is not common at all in realistic computing usage.
One fact remains that Intel's Core 2 will be far better off if CSI was available today. AMD engineers (and fans) are not exagerating when they think HT is a key to processor performance under heavy (data access) load; OTOH, Intel fanboys like you downplay the importance of such technology only because your beloved Intel couldn't come up with it in good time.
Intel's promise of that 10GHz wonder isn't 2 years old yet. Maybe you ought to say that at the time of 2 years ago 10GHz was still sentible, and P4's fast-to-clock but slow-to-run design was the way to go? I bet you had said that before, hadn't you?
Fanboys choose side, period. Because of that, they (you) are blind to truth and reality.
pointer said: "To elaborate this, at any given time, we have to consider how many core needs to access the FSB and what is the probability. The larger and better cache surely reduce the FSB accesses need."
Wrong. Cache reduces effective memory access latency, but it does not help memory bandwidth. You would've known this if you have ever read a standard graduate-leve computer architecture textbook.
The only case where a larger cache would help reducing main memory access is when a smaller cache would experience some thrashing (capacity miss), which is a very application-specific phenomenon.
"most of the time, the availability of the FSB freq to a given core is surely much more higher than the simple divide by number of core calculation."
What matters is not the availability of FSB freq to a given core, which seems to me a phrase that you invented yourself out of ignorance, but the effective delay it takes to fulfill a memory request.
The FSB can be seen as a big fast queue for memory accesses from the increasing number of cores. If we use M/D/1 queue to approximate FSB memory access, then the queuing delay is proportional to a/(d-a), where a is the arrival rate (i.e., memory request rate, proportional to number of cores) and d is the departure rate (i.e., access complete rate, proportional to FSB bandwidth). Thus the queuing delay is actually super-linear with respect to the number of cores.
Thus it is reasonable to assume that, to maintain the same memory access delay, the FSB speed should increase at least twice for double number of cores, unless some cores are idle or memory starved.
As a side note, the Erlang unit measures the capability of a system (above which the queue length grows to infinity), but it doesn't measure delay. The memory system almost never have the enough Erlang units to fully serve the core (the Core 2 core can theoretically process four 64-bit instructions per cycle, thus demanding up to 64GB/s at a mere 2GHz clockrate). Instead, the core self-adjust its speed when waiting for a much slower/delayed memory response. Suppose you already had all the memory access data ready and start up a simulated run on both the core and the memory system, the core will definitely finish up much, *much* sooner than *any* memory system (since it wouldn't need to wait for memory anymore).
You will learn that sometimes a simple straight-division is more accurate than your fancy Erlang trunk theory.
"50% increase of a segment that grew about 25-30%, in which they (AMD) hold only 13%."
enumae, stop this kind of high-school maths, will you? I mean I did much better than that when I was in high school...
Industry-wide mobile shipment grew 25%, but AMD's mobile shipment grew more than 50%. That means, in 1st order estimation, 50% growth for AMD, and a bit more than 20% growth for Intel. How did you get the 13% again?
AMD have both mobile Sempron and Turion, but the former has lower unit ratio, and much lower revenue ratio in the market.
"Core architecture shows no indication of being bandwidth bottlenecked at this time."
What are you talking about? Core architecture, as a matter of fact, has little improved memory system, asdie from its super large cache. How is it then to avoid memory bandwidth/delay problem? In fact, the more powerful the core is, the more severe the problem will be. Core 2 has a good core, but not a good architecture that works around and with the core. As such it is not a good server chip.
Servers today have gigabit links; a simple calculation would show that it can take less than an average packet RTT to have a 16MB cache refreshed. So the next time I remotely access the server, data from my previous access is likely totally gone from its super big 16MB cache. And I thought we had expected quad-core 16MB L2 to support virtualization, which will only intensify the problem. This is just one example of the case where fast core plus big cache could rock on simple benchmarks but flat in real-world scenarios.
Edward said...
"enumae, stop this kind of high-school maths, will you? I mean I did much better than that when I was in high school...
Industry-wide mobile shipment grew 25%, but AMD's mobile shipment grew more than 50%. That means, in 1st order estimation, 50% growth for AMD, and a bit more than 20% growth for Intel. How did you get the 13% again?"
Edward... Ok, here we go.
For starters I didn't run the numbers, but I will now.
AMD held 13% market share of the mobile segment at the end of quarter 2.
Market share growth
50% growth of that 13% would then be 6.5% growth, which would equal 13 + 6.5 = 19.5% market share.
Intel at 87% of the market grows the remaining 18.5% growth, which would equal 87 + 18.5 = 105.5% market share.
105.5% + 19.5% = 125%
100 * 100 / 125 = 80%
Adjustment for new market share
AMD adjusted market share 19.5% * 0.80 = 15.6%
Intel adjusted market share 105.5 * 0.80 = 84.4%
I hope this makes sense now, the 50% equates to about 2% market share gain, and why I said it was blown out of proportion.
"AMD have both mobile Sempron and Turion, but the former has lower unit ratio, and much lower revenue ratio in the market."
Thanks for answering my question.
One fact remains that Intel's Core 2 will be far better off if CSI was available today. AMD engineers (and fans) are not exagerating when they think HT is a key to processor performance under heavy (data access) load; OTOH, Intel fanboys like you downplay the importance of such technology only because your beloved Intel couldn't come up with it in good time.
it takes a AMD fanboi like you to blind enough, even to your own statement. Yes, it is better to have CSI now. But, what is the impact for the current usage on the desktop/laptop or some server apps? It is neglegible enough, say 1%. In case you argue some more, i'll increase the numner to 10%, but it is still neglegible ... errr sound familiar? :)
Edward said...
Wrong. Cache reduces effective memory access latency, but it does not help memory bandwidth. You would've known this if you have ever read a standard graduate-leve computer architecture textbook
Wrong ... may be you read a predtaed, or wrong textbook, or some textbok written by some AMD VP, expecially the marketing one :)
When you have cache hit, in Intel C2D case, either the core itself or other cores read that in, you would not need to access memory through the memory controller. In such, the NEED to using FSB is reduced, as compared to AMD's CPU (IMC) under exactly the same scenario. By reducing the need, the FSB availability to the cores are higher.
What matters is not the availability of FSB freq to a given core, which seems to me a phrase that you invented yourself out of ignorance, but the effective delay it takes to fulfill a memory request.
Nope, what really matter in term of delay should be view from CPU end, not the bus end. Intel has been able to balance this with the cache as i said. One has to look at it statiscally, for a given usage scenario, what is the delay of a CPU do a load from memory/cache. You are so ignorant, out of the AMD fanboism, ignore the fact that the availability of the BUS when you need it has effect on the effective latency.
pointer: "When you have cache hit, in Intel C2D case, either the core itself or other cores read that in, you would not need to access memory through the memory controller. In such, the NEED to using FSB is reduced, as compared to AMD's CPU (IMC) under exactly the same scenario."
I don't care how wrong you are, but I could not let you spreading wrong information. Well, not really your fault per se, since 90% of enthusiasts on the Net are doing similar grasp-and-amplify things on computer architecture concepts. Sometimes those rough ideas are about inline with true analysis, but many times they are not.
Lets look at the cache size and memory bandwidth problem from more scientific point of views. If you look at the average global capacity miss rate in benchmarks (not real-world tests), the difference between 512KB and 4MB is something like 0.5%, with the former having an absolute values around 1%. (Thus a ~50% reduction in FSB access).
However, for the same cache size, among different benchmarks, miss rates could differ as much as 8% (i.e., 16x of the average value) from one benchmark to another. In those high miss rate cases, larger cache size, i.e., 4MB vs. 512KB, generally don't help (with gcc be a notable exception).
What that means is that there are essentially two types of programs: one that can be helped with larger cache, and one that cannot. For benchmarks, the former clearly dominates; for real-world applications, however, it depends on how you use the computer, but these two cases are about even.
Thus my previous comment said it clearly: it is highly workload specific. For office applications with a document being just a few hundred KBs, you need little memory access out of the L2 cache. For streaming, media editing, and server applications, cache mostly cannot help bandwidth.
"Nope, what really matter in term of delay should be view from CPU end, not the bus end. Intel has been able to balance this with the cache as i said."
Do you even know what you are saying? I was looking at the memory subsystem as a queuing system in whole. There is no "bus end" at all, because the bus is the system. Everything was analysized from the "CPU end".
Intel was able to "balance" delay for single-process, benchmark-oriented workloads with super large cache size, which helps in those cases. It doesn't work well for servers or applications aforementioned that requires high bandwidth.
Everyone speculating about AMD calcs vs. Intel?
Current Assets- Current Liabilities- Inventory:
Intel: 4.07B
AMD: 1.55B
Not bad for AMD, not great. Still- not an accurate reflection of liquidation value of inventory, but lets work with it.
Let's look at some common metrics used by those losers on Wall Street who don't understand liquidity:
1. Current Ratio, the ratio of CurrAssets/CurrentLiab. <1 is considered to be an indicator of liquidity issues and an inability to pay bills:
Intel:1.94
AMD: 2.07
Both do reasonably well here, and are fairly close. Neither is showing signs of liquidity issues.
2. Quick ratio- this is what Sharikou is getting at, without even recognizing it. Also known as the acid test, a more conservative look at liquidity. (Current Assets - Inventory)/Current Liabilities
AMD: 1.82
Intel: 1.45
Here, AMD is doing better, BUT- Intel is still far > 1.0. No danger at this point.
3. The most conservative liquidity ratio is the Cash Ratio: Cash+Marketable Securities/Curr Liabilities. This is only important in case of a demand payment-unlikely in either case.
AMD: 1.25
Intel: 0.91
AMD is doing fine here. Intel is a bit sub-1, but again, no risk here- suppliers would be likely to extend terms before they let Intel go under- too big a customer to lose.
Now to the scarier part for AMD: getting beyond the short term. Recognize that these numbers do not include taking on ~2.5B in debt +consuming .5B in cash and diluting equity for the remaining 2.4B to acquire ATI...
4. Debt/Assets Ratio:
AMD: 40.2% (42.6%, post-ATI, with substantial goodwill on the books)
Intel: 25.3%
5. Debt/Equity Ratio:
AMD: 67.2% (74.1%, post-ATI, with substantial goodwill on the books)
Intel: 33.8%
These numbers are far less promising for AMD if they can't begin generating a substantial cash flow to address the debt payments that will come due in addition to that needed to expand capacity. Current liabilities includes the current portion of long-term debt. Intel needs to get out of it's current inventory problem, but long term appears to be on stable ground. Given that the CC indicated a lot of the inventory is flash-related, there is an easy solution to improving the numbers- dump flash. AMD already took that route and are the better for it- don't be surprised if Intel does the same and ends up with some spectacular results due to this...
These numbers are far less promising for AMD if they can't begin generating a substantial cash flow to address the debt payments that will come due in addition to that needed to expand capacity.
Many view AMD's ATI acquisition as a risky move, however, as cautious as Hector Ruiz, I know he has done the math. AMD FABs will crank like hell and billions of profits will flow in... As I said before, Intel's fate solely depends on AMD's capacity, which is sky rocketing in the next two quarters. Just wait and see.
Many view AMD's ATI acquisition as a risky move, however, as cautious as Hector Ruiz, I know he has done the math.
Article of faith. As a shareholder, I'd be pissed as hell- this acquisition is a HUGE distraction for management, dilutes my equity, and puts a huge amount of goodwill on the books, some of which I likely will have to write down very soon as Intel Intel no longer wants to (nor is obligated to) support my offerings for Intel CPUs. Financially, Hector may have bet the company on this one- the equivalent of pushing all in with pocket 9s. It might pay off, but it might just bust you out.
AMD FABs will crank like hell and billions of profits will flow in...
Undoubtedly the fabs will crank like hell. Whether or not billions of profits will flow in is again an article of faith.
And with AMD now leveraged to the hilt, they won't be able to borrow their way out if they get into trouble this time. Intel is lightly leveraged, so even if they get into short term trouble, they still have the option of floating debt and using the funds to wait while AMD's debt load crushes the company.
In a perfect world, both companies will make enough money to be able to continue to invest in the forward march of CPU technology. But AMD's latest moves may make that more difficult. Then again, they may not...
Article of faith.
Dell is not faith. DELL deal is money. Also, ASP is not a concern, because AMD cost is lower and Intel will have to raise prices to delay BK as it quickly loses mkt share.
Edwards said Intel was able to "balance" delay for single-process, benchmark-oriented workloads with super large cache size, which helps in those cases. It doesn't work well for servers or applications aforementioned that requires high bandwidth.
a typical AMD fanboi, trying to spread false statement, especially now. Are you really read-challanged? First, those benmarks out there include both single threaded and multi-thread results, which intel beat AMD by mile. Second, i have said Intel current implementation is good enuf for desktop/laptop and some server apps and you just state It doesn't work well for servers or applications aforementioned that requires high bandwidth as if i never said that. Good try for an AMD fanboi that run out of facts.
"Good try for an AMD fanboi that run out of facts."
You should spend less time calling pointless names and spend more to read my comments where you'll at least learn the way of scientific analysis.
The thing you said closest to "fact" or "analysis" other than folklore imagination was probably the term "Erlang," which unfortunately you had obviously no idea of.
Just you know I run both an X2 4200 and E6600; I use them for many workstation tasks, including graph computation, source/channel coding, simulations, and some cryptography, and my analyses are all in line with facts. The real false statement here is your "AMD fanboi" one.
To my observation, Core 2's true superiority is in media transcoding, or when SSE2 is supported. For all other workloads, Core 2 is less than 15% faster than X2 at the same clockrate - if multiple processes arerunning, or if the apps are bandwidth demanding, Core 2 is actually losing. No, I don't do SuperPi, nor do I bench games in low resolution, and I really don't care their toy results.
My suggestion to you is, first, get some good education in computer architecture, then, get both an X2 and a Core 2 and do something meaningful works. Other than that, you are just miles away from any "fact" you talk/imagine about.
Edward said ... You should spend less time calling pointless names and spend more to read my comments where you'll at least learn the way of scientific analysis.
It's so hilarious. The one that started all out calling ppl fanboi, making false and distorted statement, now begging ppl not to call him a fanboi.
Just you know I run both an X2 4200 and E6600;
good for you that you have the great C2D system. I have exactly the same one too, but not the 4200. I;m not rich enough to buy few system just for testing anyway.
if multiple processes arerunning, or if the apps are bandwidth demanding, Core 2 is actually losing.
Cool. please publish the test sets and i believe AMD would love you very much. People can also analyse your test sets and may be they will be convince to buy an AMD one if they found that's exactly what they do in REAL life?
No, I don't do SuperPi, nor do I bench games in low resolution, and I really don't care their toy results
Well, i now convine you have no computing background because you have no idea on the CPU bounded or GPU bounded, etc. In case of the GPU bounded case, an upgrade of the GPU will see the performance gain right away. Expect to see updated benchmarks into the C2D advantage when the better GPU is out later. My advice for you is to take some computer related courses (if you are qualified). I'm not really into teaching but engineering thus i can't help you much.
"In case of the GPU bounded case, an upgrade of the GPU will see the performance gain right away."
So you do know games are GPU bound, don't you? Thus using a (pair of) better graphics cards will improve gaming performance for BOTH X2 and Core 2 (precisely because the reason above). What's your argument favoring Core 2 for gaming in normal resolution then?
"Cool. please publish the test sets and i believe AMD would love you very much."
How can you be so rock-head? There are abundant test results that show, for server workloads, even S939 Opteron is more efficient than Woodcrest at 2P. Here is just an example of the apache benchmark, Woodcrest vs Opteron.
So in your attempt to bad-mouth me, you actually proved yourself having bad logic. Way to go.
So you do know games are GPU bound, don't you? Thus using a (pair of) better graphics cards will improve gaming performance for BOTH X2 and Core 2 (precisely because the reason above). What's your argument favoring Core 2 for gaming in normal resolution then?
Exactly. By having a better GPU, in a point when it is no longer GPU bounded, you surely see big difference in the game experience. If you insist on this doesn't matter because of the current GPU bounded stuff, then i guess the gamer can go buy a much cheaper PD and good GPU then ... but i guess you won;t say that and you never think of that when there are no C2D few months ago.
How can you be so rock-head? There are abundant test results that show, for server workloads, even S939 Opteron is more efficient than Woodcrest at 2P. Here is just an example of the apache benchmark, Woodcrest vs Opteron.
Abundant? where? btw, I don't run apache in my desktop/laptop. My argument was always base on the desktop/laptop and some server apps. why not just publish what you did here instead of go 'taichi'. I really not sure who is the rock haed here. If you are really read-challenged, please ask your frens to inteprete for you on what is said :for most current desktop or laptop or some server usage, the FSB is still sufficient. It is and will not be sufficient for high MP system. It is not sufficient for whatever desktop/laptop application that need VERY HIGH FSB access which i do not know of any example or the number or even the how frequent user would use such apps.
effient? what kinda vague comparison statement. What is the effcient that you are talking about here? better scaling in 2P and above? even so what is the end result? by pitying you, lets say AMD has 15% (i'm being very generous here, i think ~10% is more reasonable) better scaling in 2P as compared to Intel's. But the end result will still be the Intel CPU win because it's much better performance which is greater than 15%. You can say all you want that you own those system and yet not able to give concrete statements but vague ones.
"Exactly. By having a better GPU, in a point when it is no longer GPU bounded, you surely see big difference in the game experience."
As long as you get 60fps at low resolution from a game, it will not be CPU bound.
"btw, I don't run apache in my desktop/laptop. My argument was always base on the desktop/laptop and some server apps."
Based on those benchmark tests run by those websites (not even yourself), your arguments were only on the desktop. OTOH, my analysis of Core 2 being FSB limited has always been referring to server and (some) workstation workload. You couldn't understand, and you wouldn't read properly. I really don't know what's your problem. Maybe you should spend more time with a shrink instead of here?
"why not just publish what you did here instead of go 'taichi'."
1. You are worthless to let me compose and publish my results.
2. Go test it yourself, that is if you can. I am telling you facts, whether you approve it or not.
3. You wouldn't be capable to understand the results anyway.
"What is the effcient that you are talking about here? better scaling in 2P and above? even so what is the end result?"
If you don't understand, maybe you should stop talking.
"by pitying you, lets say AMD has 15% (i'm being very generous here, i think ~10% is more reasonable) better scaling in 2P as compared to Intel's. But the end result will still be the Intel CPU win because it's much better performance which is greater than 15%."
What do you think scaling is? Rebate price comparison that you can add/substract percentages at will? Only clueless amateurs will calculate like this.
I really doubt how much formal computer engineering education you've been exposed to. I guess it's very little, but well it's none of my business. I'm tired of your pointless & worthless arguments. Consider this be the last comment I will write to you.
I really doubt how much formal computer engineering education you've been exposed to. I guess it's very little, but well it's none of my business. I'm tired of your pointless & worthless arguments. Consider this be the last comment I will write to you.
It's really a waste of my time. When i posted my comment on something else, then you start firing back, name labelling ... and then now concede to my analysis of Core 2 being FSB limited has always been referring to server and (some) workstation workload.
it's really worthless to talk to someone that will admit that he/she is wrong when it is. This is my last comment to you here too. Last advice to you - Go take some computer course or any other smaller course in order to qualify you for such course.
Post a Comment
<< Home