Conroe-L suffers for lack of cache
Expecting 21 second SuperPi 1M for a 2.4GHZ Conroe? See this screen. A 2.4GHZ Conroe-L does SuperPi 1M in 32.6 second, 55% slower than the first result (After overclocked to 3.2GHZ, the result is 27 second). The difference? The former had 4MB L2 cache, the latter had smaller but seemingly enough amount of cache: 512KB.
As I analysed long ago, Conroe's so called performance is simply piled on cache. At the microarchitecture level, there is little advantage for the Core 2, a few percent at most. Review sites tout Core 2's 4 issue core and other architecture hacks, but the real truth has nothing to do with those. Core 2 is just an old dog trick, cache. I pointed out that SuperPi is not suitable for benchmark comparison between Intel and AMD for exactly this reason. It's obvious that SuperPi is very cache dependent.
That's why except those optimized for Intel's SSE, in real applications where working set is larger than 4MB, Conroe XE 6800 shows little performance advantage. AMD64 has 30% lead over Core 2 in SpecFP_rate, partly because the those complicated applications in SpecFP have large working set.
That's why I project 4x4 will frag Kentsfield. AMD64's scalable architecture will establish itself as the performance king. Anything less than that will be just second rate or mainstream.
After AMD 4x4 launch, the desktop computing landscape will be the following. At the high end, we have AMD 4x4 with performance supremacy, it will be the only high end machine, costing $2000 or more. Down below, we have mainstream CPUs, such as Kentsfield, Conroe and Athlon 64 X2. At the bargain basement, we have Pentium D and Semprons. Pentium 4 and Core Duo will be found in landfills.
An overclocked Kentsfield won't help. The limiting factor in Kentsfield is the choking FSB, with only 266MHZ allowance for each core.
posted by Sharikou, Ph. D. @ 11/04/2006 05:46:00 PM
74 Comments:
Well AMD had to pile on HT and IMC and is still behind on the desktop. It's a shame that AMD can't use cache as efficiently as Intel.
Thats about the same pi time as a similarly clocked A64 and it only costs around $60.
Also 1M version of Conroe-L are getting 20sec pi time and will be sub $100
http://www.xtremesystems.org/forums/showthread.php?t=121711
First things first, you had said before that a Super PI benchmark means absolutely nothing. But now that Intel has an engineering sample of a chip with 512KB cache and it is showing something in your favor, super PI is an acceptable benchmark?
WTF!
The sad thing is that you did not look up other Super PI scores for the FX62.
Here are some and it would seem the Conroe-L at 2.4GHz is doing very well against the FX62 at 2.8GHz...
-[link 1]-
-[link 2]-
So who should be worried AMD or Intel?
I understand that the performance is not equal to the E6600, but a single core (priced at around $150, assuming) in a single threaded test with a lower clock (2.4GHz) is challenging an FX62 ($700 2.8GHz) in single threaded performance.
If anything you are only showing that an AMD FX62 is overpriced and that the Core architecture is better than K8.
Way to make a case against Intel's Core architecture...lmao
PS: If this is the kind of scientific mind you took to court I will gladly refuse your services...Just kidding man :)
WTF??? You can't be serious dude. You are using a link to a picture of a benchmark for evidence. WOW! I vote that we replace this clown as the representative of AMD fanboys with Scientia. Scientia a real AMD fanboy who knows what the @$@#$ he is talking about.
Paid for by the: Let's get someone that actually knows what they are talking about commitee :)
This blog is so embarrassing for real
AMD fans. Sharikou is a joke and can't be taken seriously anymore, just look at his resent post. For a real discussion about pervasive 64 bit computing please go to....
http://scientiasblog.blogspot.com/
Don't let this clown represent the real fans.
Is this guy serious? There's hundreds of reviews out that has the Core 2 Duo easily beating anything Amd currently has. Can we just move on, I'm sure K8L will be very competitive against the Core 2 Duo
So intel will probably be about 40% Conroe, 70% Woodcrest, by the end of Q4. This has to be a problem for Amd since 40% Conroe is almost twice as much Conroes sold than Athlon X2s. I see Amd going BK by the end of 2007 if they can't get K8L to work. Bad times ahead for Amd. Hp will also take a huge amount of market share from Dell as well. Hp will now be intel's favorite OEM. Don't be surprised if Hp goes 100% intel by the end of 07.
Could it be that because it has 1/4 of the cache compared with a Core 2 Duo, a slower FSB... and only a single core it will be slower? Hahha. Nice try FUD king.
You didn't "analyze" anything. The fact is you "blathered" something and nobody cared.
Nobody cares. Most people are mature and intelligent enough to know that of course Conroe's cache helps it tremendously, it's part of the design decision made when the chip was made.
Your special brand of idiocy is akin to one 100 meter dash contestant complaining that the other 100 meter dash contestant won the race because "his legs are longer and more muscular".
silly kid this PhD..
Did he know the Core2 L still kicks AMDs ass and trounces anything near in price.
Keep it up.. and your interent school my revoke your degree Pretender
The K8 scores over 40 seconds with the same cache. So much for the K8 uarch.
SO, you admit that CONROE frags any AMD offerings..
The first step to healing is getting out of denial.. Good step Shar
Something is terribly wrong with that. Why are current 2M Conroes way faster than that?
Hey P.P.P.Pretender....
That's a super budget celeron replacement....under a 100bux! and still whoops on any single core that AMD has right now at only 2.4Ghz....
CHeck out the 2.8ghz version with only 1mb of cache
http://www.oc.com.tw/article/0611/readolarticle.asp?id=5557
That's right pretender, 22 second pi time....and suppose to be sub $100....
AMD's single core market is gonna get OWNED
Yeah, too bad this primitive cache intensive microarchitecture just demolished AMD's AthlonFX by the biggest margin and relegated it to compete with Netburts and Celerons.
You can point out whateve silly thing you can find against Intel and its processors, but the bottomline is your beloved AMD's products really doesn't have anything to show for.
Of course you would NEVER pratice guerilla marketing, of course not.
However, you still seem to forget that the performance offered is higher than anything any other AMD CPU can offer...
I really think most people posting here is way too shortsighted. I do believe Sharikou has a point. Being the large size of the cache the main reason for good performance is not a good sign for woodcrest. So, what is next step? 8M, 16M, and so on??? That doesn't sound like a scalable and future-proof architecture to me.
Calling all Best Buy and Circuit City Electronic sales boys..
The Pretender has made his prediction. You read it hear, use it on the sales floor
"That's why I project 4x4 will frag Kentsfield"
This is the same person that told you it was the INTEL CPU causing the laptops to catch fire. This is the same person that told you INTEL BK what was the prediction in 3 quarters.. but revised.
So when the buyers asks why you think AMD's 4x4 frags. YOu can its from the same person that said the INTEL CPUs causes laptop fires and that INTEL will be BK in 2008.
Yup nothing like references from past analysis to back up current and future prediction.
Thanks again to Sharikou PhD who is so stupid as he has NO credibility at all.
Pretender..
1000:1 I'll give you for betting on your BK prediction
100:1 on your fragging prediction.
You name the place I'll bring the dough. Man I'll even let you hold the money.. as I can't wait to expose your total cowarderly hiding behind a blog but won't put money where your prediction is.
Come on are you game for a bet.. Mr PhD Pretender!
Bottom line is profit. Big cache is expensive = less profit. Intel is forced into a situation of survival. 65nm & large cache is all Intel can do to compete with an old 90nm AMD design.
The clock is ticking and an Intel’s P3 bandage is running out of time. It’s evident by the every frustrated Intel fanboy comments of attacking the messenger that denial has become the flavor of the day.
Conroe’s 15 minutes of fame was fun while it lasted. I raise my beer mug to the Conroe and pose a toast.
Intel has some big issues to overcome like Vista, 64-bit, Graphics, FSB and scalability. The near future may find some temporary fixes for Intel, but realistically speaking, not today.
I wish they would run some superi 8M tests. With 4M vs 2M only 1M test has any kind of benefit. At 2/4/8M they are almost at the same speed and both vastly better than any K8.
"One more thing that is on my mind at the moment: if intel is soo ahead of AMD in their technology then how come they are soo firing people off? hehe."
To save money? After all, the less you spend the bigger the profits.
Posted by Enumae:
First things first, you had said before that a Super PI benchmark means absolutely nothing. But now that Intel has an engineering sample of a chip with 512KB cache and it is showing something in your favor, super PI is an acceptable benchmark?
WTF!
In what part of Sharikou's post can you read that he is now arguing SuperPI is a valid benchmark? He is not using it to compare Conroe with other machines, nor to give it some kind of score. He just states that Woodcrest architecture performance is very cache-size dependant, and uses this application (the one that many people use to proclaim conroe's virtues) to prove it.
Please learn not to read what has not been written, and you will do a favor to this blog and the people who read it.
This is the same person that told you it was the INTEL CPU causing the laptops to catch fire. This is the same person that told you INTEL BK what was the prediction in 3 quarters.. but revised.
I have proved that Intel CPUs caused explosion by clear and convincing evidence. As for Intel's BK, the time is projected to be 2Q08.
With a bunch of apps running, my e6300 clocked at 2.1 Ghz does superPI 1M at 31 seconds. (2MB cache)
This is the result:
000h 00m 31s [ 1M]
Their test is seriously flawed for some reason.
Try running it on your 3800 Sharikou, post us the results. Would you be embarassed if it took more time?
The Pretender has made his prediction.
Dude, I think Intel folks have a lot of respect for me. I pinpointed the timing of DELL-AMD alliance. They did not believe it, now they do. People always respect the ones who have the wisdom and power to see the future. Intel's BK is set at 2Q08. That's the prophecy, if you like. But in fact, it's simple math..
Dude, I think Intel folks have a lot of respect for me. I pinpointed the timing of DELL-AMD alliance. They did not believe it, now they do. People always respect the ones who have the wisdom and power to see the future. Intel's BK is set at 2Q08. That's the prophecy, if you like. But in fact, it's simple math..
Yeah, you and a LEGION of other people that predicted the dell-amd deal. Everybody and their dog saw THAT one coming. Wow, you're amazing!
Actually, the battery issue was the stupidest conclusion I've ever heard coming from a half-intelligent person. You should concede defeat before you look even more like a fool.
Another person wagered a fair chunk of change and was willing to place it in escrow on the bankruptcy of Intel. If you're so smart and prophetic, why don't you take the bet? Chicken?
Why do you insist on spreading such blatent lies?
"AMD64 has 30% lead over Core 2 in SpecFP_rate, partly because the those complicated applications in SpecFP have large working set."
Really, thats funny cause AMD FX-62 SPECint_rate2000 score was 47.2, and an Intel X6800 is 63.6 which would equate to a 26% advantage for the X6800, now look at price FX-62 about $700 vs X6800 $1000, almost $100 for every 10% advantage, seems worth it.
All the while a lower clocked E6400 with 2MB cache got 44.1, so AMD only has about a 7% advantage, now look at the price FX-62 about $700 vs E6400 $220, almost $70 for each 1% advantage, the price worth is not worth it?
We all understand that Cache is there to help with the FSB bottleneck, but that is what they are using a primative and slow FSB, but at the same time it is beating AMD.
Now whether you want to admit it or not Intel has done a great job with the Core Architecture when comparing it to Netburst, and when you compare it to K8 thats alot of ground they made up, wouldn't you agree?
"That's why I project 4x4 will frag Kentsfield. AMD64's scalable architecture will establish itself as the performance king."
This is just retarded, frag?
The simple fact is K8 does not give 100% scaling (as I understand it no architecture does) so the most to look for would be about 80%, now factor in the 20% advantage core has already and they should be about even in FP, but Intel will still have an advantage in INT.
Now this is not all test but about 90% of desktop based applications should be equal or a little better for either camp, so your frag statement is just hot air.
"...At the high end, we have AMD 4x4 with performance supremacy, it will be the only high end machine, costing $2000 or more..."
To utilize 4x4 to the extent is was made for it will cost much more than $2000 even with lowend parts, so the statement of $2000 or more is rubbish, should be about $4000 or more.
For starters...
Motherboard $300 or more
Processors $1000 or more
4 GPU's (low end but SLI capable) $600
8GB Ram (low end DDR2-800MHz C5) $800
Power supply (low end quad SLI ready 850W) $281
Which totals about $3000 without a case, monitor/monitirs or keyboard, now all of a sudden your looking at about $4000, and still lowend.
"An overclocked Kentsfield won't help. The limiting factor in Kentsfield is the choking FSB, with only 266MHZ allowance for each core."
Good thing Asus and Nvidia are releasing the 680i chipset which has a 1333 FSB and should allow some good overclocking for Kentsfield.
Sharikou, I have a couple of questions for you.
Suppose you undertook the task of managing Intel. Then...
1) What actions would you take?
2) Do you think these would avoid your predicted BK?
Thanks in advance.
Anonymous said...
"In what part of Sharikou's post can you read that he is now arguing SuperPI is a valid benchmark? He is not using it to compare Conroe with other machines, nor to give it some kind of score. He just states that Woodcrest architecture performance is very cache-size dependant, and uses this application (the one that many people use to proclaim conroe's virtues) to prove it."
He has stated in the past that Super PI is not a valid benchmark, look through his old posts and you will find it.
Please learn not to read what has not been written, and you will do a favor to this blog and the people who read it."
Now to your rude comment...
Is this your first time here?
Though its nice to see you approach a rude statement so politely, you fail to realize that Sharikou likes to try and spin information/comments/arguments to favor his view.
Had you any problems with my comments you would have said something, correct?
So I will assume that my points made sense to you and you just have not been here long.
Hang around a little while and you will see this, Sharikou the spin doctor.
And look where Dell has taken them! Rock bottom margins and hostility with the channel.
1) What actions would you take?
2) Do you think these would avoid your predicted BK?
I answered these long time ago already.
1) layoff 40K workers
2) Since AMD can supply 30% of the market, I will cut Intel's production to 65%, creating a 5% shortage. Basically, I will stop all 90nm production and start convcerting them to 65nm.
3) make a ton of profits due to the price hike
4) convert all FABs into 45nm using the money made
5) make quadcore the standard...
" And look where Dell has taken them! Rock bottom margins and hostility with the channel."
"the channel" has been always the second class of citizens, well known.
The same second class that kept their margins up and took them when no one else did.
Anonymous said...
" And look where Dell has taken them! Rock bottom margins and hostility with the channel."
"the channel" has been always the second class of citizens, well known.
1:03 PM, November 05, 2006
Sad, but I suspect that you, Sharikou, make up anonymous aliases to support your conspiracies.
Ok, finally I decided to get an account to stop posting anonymous (I was lazy because a barely posted, but this way it's easier for others to reply). Ok back on topic:
He has stated in the past that Super PI is not a valid benchmark, look through his old posts and you will find it.
Did I deny that fact? I actually agree with his opinion. This makes me think that you didn't read my post properly. Summarizing, what I say is that Sharikou's post does not endorse SuperPI and a proper benchmarking tool, therefore your criticism doesn't make sense.
Is this your first time here?
Though its nice to see you approach a rude statement so politely, you fail to realize that Sharikou likes to try and spin information/comments/arguments to favor his view.
No, it is not my first time here. I read this blog very often, but I barely make posts. And I didn't say I agree with each and all of his statements. Therefore please don't act so condescendingly.
Had you any problems with my comments you would have said something, correct?
So I will assume that my points made sense to you and you just have not been here long.
Hmmm, I don't understand what you mean. I don't have any problems with your comments, I just disagree with you (at least the part of your post I referenced), and that is what my post expressed, showing you why.
By the way, one last question: do you consider SuperPI as a valid benchmarking tool? (do not look for any irony in it, just want to know your opinion).
Thanks.
Summarizing, what I say is that Sharikou's post does not endorse SuperPI and a proper benchmarking tool, therefore your criticism doesn't make sense.
Yes. It is indeed sad that most of the Intelers are so low IQ, they can read, but they can't comprehend.
1) I actually never said SuperPi is an invalid benchmark. Any program, including BogoMIPS, can be used as benchmark. The only question is relevance to real use scenario. I repeatedly stated that Conroe is the SuperPi chip. if Intelers run SuperPi all day long, they can celebrate 365 days/year. But for sane users, SuperPi is irrelevant.
2) From this message I posted, the observation I made is that the CORE2 microarchitecture is not good at even SuperPi after all, Conroe's superpi score originated from the cache size, as I analysed long time ago.
2) From this message I posted, the observation I made is that the CORE2 microarchitecture is not good at even SuperPi after all, Conroe's superpi score originated from the cache size, as I analysed long time ago.
What's wrong with that, if it performs like a bat out of hell?
alb said...
"Ok, finally I decided to get an account to stop posting anonymous (I was lazy because a barely posted, but this way it's easier for others to reply)."
Sorry to come across rudely, it is just that anonymous posters really bother me.
"Ok back on topic:In what part of Sharikou's post can you read that he is now arguing SuperPI is a valid benchmark?"
In his previous post (around the time Conroe first came out) and now look at his post below your's you'll see him say that...
"The only question is relevance to real use scenario."
Yet he uses it as the basis for an argument against Conroe-L, like I said, now he can use it because it supports his opinion that Intels core architecture is "simply piled on cache", but my point is that a lower clocked 512KB cache chip is competing against an AMD FX-62, so the performance might be worse than with the "piled on cache", but the core architecture is much more competitive, and actually better than K8.
"Summarizing, what I say is that Sharikou's post does not endorse SuperPI and a proper benchmarking tool, therefore your criticism doesn't make sense."
If he is using it to support his claim or to base a debate then yes he is supporting the benchmark.
"Therefore please don't act so condescendingly."
I am trying to be very polite considering your initial statement...
Please learn not to read what has not been written, and you will do a favor to this blog and the people who read it.
So who is really being condescending?
I have no hard feelings, but it works both ways :)
"Hmmm, I don't understand what you mean..."
That was mainly out of frustration, sorry about that.
"...do you consider SuperPI as a valid benchmarking tool?..."
No I do not, I just have a hard time seeing his faulty claims and constant spinning.
Still you fanboys deny the benchmark results all the time. AMD beats Conroe in science benchmarks and AMD beats Conore on Memory bandwidth.
And in the end, I dont run benchmarks and I dont know many end users that run benchmarks.
I run complex multitasking software and thats where Conore lags behind AMD.
In fact, a conroe with a 945 intel chipset and onboard video cant even run google earth.
A AMD with onboard video, ATI can run google earth just fine.
I only point this out since it is another Intel conroe failure.
I guess its just reworked and overworked pentium 3s, conroes just cant handle modern software.
Intel drowning in a 5 billion dollar conroe inventory and they cant give them away cause nobody wants them.
Sharikou said...
"Yes. It is indeed sad that most of the Intelers are so low IQ, they can read, but they can't comprehend."
What is really sad is that you don't see your own spinning, I used to think you were intelligent and had an understanding of what you did.
Now I understand that you are really an obsessed AMD fan who can not accept defeat no matter how short lived.
----------------------------------
"... But for sane users, SuperPi is irrelevant."
Then why post something showing Super PI scores and have that as a base for an argument?
"From this message I posted, the observation I made is that the CORE2 microarchitecture is not good at even SuperPi after all, Conroe's superpi score originated from the cache size, as I analysed long time ago."
And like I had pointed out a 2.4GHz Core architecture processor is competing with an AMD FX-62 running at 2.8GHz.
So how can you say that Core architecture is not good?
Or are you saying that K8 is worse?
I try and be polite with you, but it would seem to be pointless.
lol, and I remember all intel fanboys running around screaming angrily claiming superpi was a correct benchmark, now that Mr.Sharikou finds something that sucks at superpi, now they now whine back claiming superpi is not a good benchmark tool?
ROFL!
PATHETIC!!!
So if it's not invalid, but has "relevance to real use scenario"..What exactly is an invalid benchmark?
What exactly is an invalid benchmark?
A benchmark is invalid when its correctness can't be independently verified. Guerrilla benchmarketing done by Intel at IDF is an example. We busted that long ago.
"Guerrilla benchmarketing done by Intel at IDF is an example. We busted that long ago."
Is Core 2 not 40% better than previous Intel technology (P4)? It was the reviewers, not Intel, who twisted this to say 40% better than AMD.
Hw exactly did you bust the IDF benchmarks again?
"That's why I project 4x4 will frag Kentsfield. AMD64's scalable architecture will establish itself as the performance king."
Well this is obvious as demo's show when I want to play 2 simultaneous games AND watch a high def video at the same time this clearly is the system to get.
Right now I'm only watching TV, posting on this blog, running a FPS game, watching a high def video and ripping a DVD to my hard drive; this is simply not enough productivity (I could easily be playing a second game no?)
Anonymous said...
"...Mr.Sharikou finds something that sucks at superpi,...
ROFL!"
Keep laughing because the fact of the mater is a 2.4GHz Conroe-L is on par with an FX-62 at 2.8GHz.
The only thing this test shows is how much better the Core architecture is than K8 without the extra cache, the extra cache is just icing on the cake.
I am not justifying Super PI, just showing your misunderstanding of the post.
for (int i = 0; i < 1000000; i++) {
do nothing;
}
There you got a valid benchmark. Is it of any use? Not at all. You cannot compare two systems by saying system A runs that piece of code faster than system B.
I like to compare it to cars and tracks. You may build a track consisting on a straight road 1 km long, and use it to time cars running it. A car may do it real fast, but it may suck when running on "real" tracks.
Sharikou provides good entertainment, but Scientia's blog is where you will get real information without ridiculous FUD. It bothers me when you have parrots actually repeating the same garbage that you spew. Like this guy.
" Anonymous said...
Still you fanboys deny the benchmark results all the time. AMD beats Conroe in science benchmarks and AMD beats Conore on Memory bandwidth.
And in the end, I dont run benchmarks and I dont know many end users that run benchmarks.
I run complex multitasking software and thats where Conore lags behind AMD.
In fact, a conroe with a 945 intel chipset and onboard video cant even run google earth.
A AMD with onboard video, ATI can run google earth just fine.
I only point this out since it is another Intel conroe failure.
I guess its just reworked and overworked pentium 3s, conroes just cant handle modern software.
Intel drowning in a 5 billion dollar conroe inventory and they cant give them away cause nobody wants them.
3:24 PM, November 05, 2006 "
He forgot the exploding chip theory of yours, not to mention the BK. What a joke.
come on to rubyworks @ www.rubyworks.com/forumz. Join and help this community grow!
I guess it would be useful to see how Conroe-L behaves with other apps, not just SuperPI.
"There you got a valid benchmark. Is it of any use?"
Actually you could use it to see how good is a CPU in branch prediction (if loop optimizers work) and in cjmp. Depending on how is it compiled you could even see if macro-op fusion works.
Though I agree that it doesn't show much else. Perhaps in the future with reverse HT and speculative threading this might really become a nice benchmark.
A bit better benchmark would be several nested loops so branching units would be stressed a bit more.
I think people are missing the point. Conroe-L has actually LESS cache than the FX-62 (512KB vs 1024KB per core), 400 MHz lower clockspeed, low FSB and no IMC. Still, it is on par with the FX-62.
Draw your own conclusions.
for (int i = 0; i < 1000000; i++) {
do nothing;
}
i++: Milion of integer incrementations?
Off topic but interesting
AMD Ships Graphics Card For Older Macs
http://www.crn.com/sections/breakingnews/dailyarchives.jhtml?articleId=193501916
We all know that SuperPI is useless as a benchmark. But as we are talking about it. My X2 3800+ oc'd to 2.6Ghz does 1M in 33 seconds.
I would really like for someone that owns a C2D to run the following test though!
Run 3 instances of SuperPI 32M. This would actually stress the CPU to the max. I'm really interested to know what the results will be.
I'll do the same, let's compare.....
Whilst this blog does have it's shall we say interesting days at least it is updated very regularly so there's usually something interesting to read is somewhat skewed:)
"Guerrilla benchmarketing done by Intel at IDF is an example. We busted that long ago."
Intel never said it's processors are 40% faster than AMDs CPUs. They said that the Core 2 Extreme X6800 would be 1) 40% faster than a 3.6Ghz Pentium D and 2) Use 40% less power. It is much faster and uses much less power than that Pentium D. They were not lying at all.
That said the Core 2 Extreme X6800 still commands a lead of some 20 -> 25% over the Athlon 64 FX-62.
Anónimo dijo...
Hey P.P.P.Pretender....
That's a super budget celeron replacement....under a 100bux! and still whoops on any single core that AMD has right now at only 2.4Ghz....
CHeck out the 2.8ghz version with only 1mb of cache
http://www.oc.com.tw/article/0611/readolarticle.asp?id=5557
That's right pretender, 22 second pi time....and suppose to be sub $100....
AMD's single core market is gonna get OWNED
3:14 AM, November 05, 2006
yeah, everyone wants to run superpi every day, all the time just to see how big their e-penis gets!!!
you're ridicoulously dumb!! :/
INTEL's predicament
AMD's FAB36 has been in production since Oct 14, 2005, finished chips should hit the market soon. No doubt, there will be an oversupply of CPUs in the market, even Joe Osha is worried about the excess capacity. What will customers buy? I think price/performance will be a deciding factor. AMD is used to selling Sempron 64s at $50 or less. No matter how you look at it, INTEL will lose market share and units.
However, what really matters to a company is not units or market share, it's revenue and profits. If INTEL can sell 100 chips for 10 billion, it will be super rich.
So, facing the inevitable decline of units and market share, the only way INTEL can maintain revenue and profit growth is to hike CPU prices.
AMD's capacity is limited and can only supply 25% of the market next year, INTEL has the say on the rest 75%. Furthermore, corporations tend to prefer INTEL brand. INTEL shouldn't have much problem increase the ASPs by 20%, to negate the effect of unit decrease (ASP increase can be done by simply lowering production of low end models).
But there is a problem: INTEL has made it a top priority to keep DELL happy. When the DELL dude makes noise about AMD, INTEL folks cave in and give more discounts, or in other words, selling at lower ASPs. And to prevent others such as HP and Lenovo from being too disgruntled, INTEL must give them some albeit less discounts also.
So, INTEL is facing a dilemma here. On one hand, market share drop is an ineveitability and it needs to hike ASP to maintain revenue growth. On the other hand, feeding DELL is a futile effort to retain market share and will result in lower ASP. (With DELL being the favourite darling, others get pushed over to AMD).
The two are in direct conflict.
Clearly, keep feeding DELL with lower priced products coupled with losing overall market share will lead to revenue and profit short fall.
It seems that the only way out is for INTEL to drop the most favoured treatment for DELL and increase prices cross the board, and make prices the same for everyone, making it possible for every one (instead of just DELL) to profit in the INTEL space and create a more balanced INTEL market.
posted by Sharikou, Ph. D @ 12/09/2005 10:35:00 AM
"In what part of Sharikou's post can you read that he is now arguing SuperPI is a valid benchmark? He is not using it to compare Conroe with other machines, nor to give it some kind of score."
alb... give it up. some people not just not capable of logical thinking. It's useless to explain to such people.
We all know that SuperPI is useless as a benchmark. But as we are talking about it. My X2 3800+ oc'd to 2.6Ghz does 1M in 33 seconds.
I would really like for someone that owns a C2D to run the following test though!
Run 3 instances of SuperPI 32M. This would actually stress the CPU to the max. I'm really interested to know what the results will be.
I'll do the same, let's compare.....
I did Super pi 1M with my E6300 C2D overclocked to 2.04Ghz in 29 seconds. I did it with a bunch of apps running yesterday and was at 31s.
I will do the 32M tonight or when I have a slow period today and post the results tomorrow (if I can remember).
Give me some proof on that Kentsfield's 266MHz FSB is choking performance, cause I haven't seen any.
Actually the four cores scale very well.
I did Super pi 1M with my E6300 C2D overclocked to 2.04Ghz in 29 seconds. I did it with a bunch of apps running yesterday and was at 31s.
SWEET!
dont forget to run 3 instances, you'll need to create 3 seperate folders, one for each instance of superpi.
I'll post mine shortly!
Edward said...
"alb... give it up. some people not just not capable of logical thinking. It's useless to explain to such people."
Edward... Why do you do this?
Do you have anything to contribute or just the ability to bad mouth someone for not seeing things as you and Sharikou?
"If someone showed an qx6700 system running something like what AMD showed its system doing (2 instances of city of heroes (a mostly cpu bound game), 2 high definition videos, 1 multi-threaded hd video encode, and then other miscellaneous apps added on top of that"
You just described usecases that get scale almost perfectly with 4x4 NUMA architecture. NUMA works well when it has two distinct data sets it can distribute between its two memory pools.
I'd like to see how well 4x4 works when encoding a HD video with all four cores working with the same data all at once.
OK,
Here's my set up:
Athlon X2 3800+ @ 2.5Ghz
TWINX2048MB @ 208Mhz
Memory timings Timings 1T & 3-8-8 ish (not trying too hard)
3 simultaneous instances of SuperPI and here are the times for each:
1: 40m 48s
2: 47m 56s
3: 48m 22s
I find it interesting that one of the threads seems to take seniority and get extra processor time.
As I said, I'd like to see what results a C2D gets.
"Edward... Why do you do this?"
Because I was pointing out that your previous responses to alb was totally without logic. Anyone with proper logic training could see that; but then, maybe you can't. Or can you?
BTW, some people prefer good grammer than good logic. They should really go grading papers for English teachers than talking about logic or intelligence.
Edward said...
"Because I was pointing out that your previous responses to alb was totally without logic. Anyone with proper logic training could see that; but then, maybe you can't. Or can you?"
Could you specify what was illogical about what I had said, I understand that I am not perfect, nor are you, so please explain.
"LOL, that's funny edward, it took training to give you your fast logical abilities?"
Actually, the better term is "education." Since I know that not everyone commenting here is capable of high degree of education, I used the word "training" instead.
Lack of higher education should not be an excuse of ignorance, however. If you don't know what logic training is about, it is your problem (which could've been easily solved by 5-min web searching).
"I actually did a personal study on Gordon H Clark's, Logic. It's an excellent book, and I think you'd like it."
I see. My apology if I misjudged you. I was too tired of those illogical arguments (whose posters often pull in all sorts of "evidence" to "support" them endlessly), and acted on your comment a bit bluntly.
Let me use this chance to do some basic logic practice on enumae's comments above.
First, Sharikou said that Conroe-L, with a much smaller cache size, runs 55% slower than Conroe on SuperPi. Then, enumae questioned "super PI is an acceptable benchmark?"
To understand this logically, there are 3 statements directly involved:
A. Conroe-L has smaller cache (than Conroe)
B. Conroe-L runs SuperPi slower (than Conroe)
C. SuperPi is a good benchmark
Sharikou's claim is "A->B", where enumae's question is "but C?" It's not only illogical, but irrelevant.
Then someone (alb) responded to enumae at 10:07 AM, November 05, 2006, pointing out precisely that the A->B has nothing to do with C. Then enumae, not giving up, followed up with a statement saying "but Sharikou previously claimed 'NOT C'..."
Well, maybe you were right, that it takes no vast logic at all to find out the problems of enumae's comments here. Just "fast" logic will do. :)
Eward which do you believe...
1. Sharikou cares about Conroe-L's performance in Super PI with only 512KB cache?
Or...
2. Sharikou has chosen this Conroe-L Super PI benchmark as an opportunity to criticize the Core architecture?
If you choose 1, then I can not help you.
Edward and Ashenman...
Edward we have been down this road before and I would like it to stop, I do not know how but there has to be some way to get along here.
Ashenman, I really do believe Sharikou to be intelligent, but it is when he starts to spin the information to suit his needs that it bothers me.
I understand what it is both of you are saying.
If I had misunderstood his post, and continued to defend my interpretation it is not out of ignorance or to be illogical.
So I am sorry if that is how you had interpreted my comments.
I may have assumed his post was more about the lack of Core architectures performance than the scalability of the caches and there effect on performance.
I still believe Sharikou to be very biased, and that is what made me draw the conclusion originally.
I would like you both to know I am not a fan boy, I just get really frustrated seeing someone post false statements, and since this site is pretty much an AMD site it may come across that way.
Again I do apologize for any misunderstandings.
I've tested SuperPi on a 2.0Ghz K8. It showed no sign at all of cache dependency. Apparently, at 2.0Ghz a K8 can access memory quickly enough to not need cache. So, cache size wasn't an issue. However, I registered a logarithmic scaling factor which renders SuperPi useless for comparison. SuperPi would need to be true scalar to allow comparisons; logarithmic is useless. Sorry to all of the OC'ers who love to post SuperPi scores but those scores may only be valid in comparision to the same architecture.
Real testing of SuperPi to see if there would be some way to fix it would be nice but I doubt this will ever be done just as it is doubtful that anyone will do true load testing on Kentsfield. At least, not any earlier than Q3 07.
At any rate, the fact that a 2.0Ghz K8 is not cache sensitive with SuperPi is not proof that Conroe L isn't. And, I'm sorry but I can't be certain on the basis of two scores. To actually prove cache sensitivity you would need to test a range and show a step in the performance.
I've tested SuperPi on a 2.0Ghz K8. It showed no sign at all of cache dependency.
Scientia, I’m but what that means?
So K8 with 128kb of L2 or 1MB not important? But for Core 2 Duo is very important?
That means better cache efficiency (or use) on AMD (or Intel)?
Limited L2 bandwidth on Intel Core 2 with less L2 cache?
Because you have to admit that Intel with Core 2 Duo we are talking of 1/4 to 1/8 (and more) difference of L2 cache sizes. 1MB maximum for K8 and 4MB to Core 2.
Post a Comment
<< Home