Monday, July 31, 2006

IBM to launch five Opteron servers tomorrow

It's Opteron time again. SUN introduced the 8P 16 core x4600, the 24TB x4500, the massive 4P Opteron blade 8400. The SUN x4600 smashes 16P HP Itanium Superdome and is very competive against 16P Power 5+, but at a fraction of the cost. Not to be left behind, IBM is going to unleash five new Opteron product lines, including two blade designs, which can be upgraded from 2P to 4P.

As I told you, 4P will be the predominant configuration in enterprise computing. AMD Direct Connect Architecture with glueless 8P ccNUMA made this possible. Just as Intel trying to catch up, the rule changes, and 4P is the name of the game. I expect IBM to EOL its x3 chipset -- too costly to follow Intel's changing FSBs. Woodcrest will be pushed to the ultra low end. The same is happening in performance desktop, moving from 1P to 2P.

In business desktop market, AMD64 is now inside ThinkCentre.

Digitimes reported that DELL's all out AMD64 flood will start in September 2006. That's only 30 some days away. Exciting times ahead, folks.

HP better hurry up.

Sunday, July 30, 2006

Opteron cleankills dual core Itanium 2 (Montecito)

4P 8 core Opteron, SpecInt_rate2000: 156; 4P 8 core Itanium 2 9050 (cost per chip: $3692) with total 96MB cache (24 MB per CPU) , SpecInt_rate2000: 134. Opteron leads by 16.4%.

Previously, we reported that the 8P Sun x4600 smashes the 16P HP Integrity Superdome with 16 Itanium 2 1.6GHZ.

In terms of floating point performance, Opteron ruins Itanium 2. the SUN x4600 got a SpecFP_2000 score of 3538. A Itanium 2 9050 got 3017. A 2.2GHZ Power 5+ got 3513.

Intel's Woodcrest Xeon 5160 got a SpecFP_2000 score of 3056 -- Opteron is 16% faster than Woodcrest and 17% faster than Montecito.

In SpecFP_rate2000, a 4P Opteron 885 (2.6GHZ) Blade got a score of 182. A 4P Montecito got 186, a statistical tie with Opteron 885. The best score for a Woodcrest server is 85.9. Of course, we know Woodcrest is only for low end 2P market, it's incapable of > 2 SMP. But you can see 2P Xeon 5160 is 47% of 4P Opteron.

AMD's denying move made 96% of Intel CPUs unwanted

As I analysed previously, AMD must deny Intel's oppurtunity to dump legacy CPUs (Netburst and Core Duo) . It's easy, just price the X2 3800+ at about $150 and sell a $50 Sempron. After that, 96% of Intel's CPUs will be unwanted.

Two days ago, I wrote that Intel top sellers on has been reduced from 3 to 1, with a crappy Pentium 4 remaining. As expected, as of today, all top five sellers at are AMD CPUs.

Top 5 Sellers

Intel has run out of ammo. Even if Intel sells Pentium XE 965 at $75, it won't help. X2 3800+ is better if you can consider total cost of ownership.

Saturday, July 29, 2006

65nm AMD parts in the wild?

Questions, questions.

AMD's 90nm X2 3800+ frags Intel's 65nm Presler Pentium XE965. I expect AMD's 65nm parts to see 40% performance improvement over 90nm.

Intel Core notebook in flames again

This page has the pcitues. Can someone identify the model number from this picture?
This picture shows the burnt hole.

We previously discussed this explosion and reached a preliminary finding that Core Duo exploded. FAA should ban all Intel Core notebooks on flights.

Friday, July 28, 2006

AMD winning the price war

A few days ago, at, Intel occupied three of the top selling CPU spots, they were Pentium D 930, Pentium D 805 and some Pentium 4 chip. After AMD's price cut, the only Intel CPU in the top five is a Pentium 4 524.

Top 5 Sellers
This is a no brainer decision: X2 3800+ frags Pentium XE 965.

Thursday, July 27, 2006

AMD market shares grows amid Intel pricing crash

We saw those Pentium D 805 at $110 which George Ou at ZDnet touted as cheap SUVs and the zillions of $30 Celerons flooding the 3rd world markets. Did they help Intel gain unit share? Joe Osha thought Yes. The reality is No.

These are the newest Mercury Research stats: AMD overall unit share 22%, Intel 73%; AMD revenue share 18%, Intel 81%; AMD server unit share 35%, DP server grew 45% in one quarter. Intel did gain 0.4% unit share in the mobile space. Turion X2 should push AMD's mobile share up in Q3.

Going forward, with AMD's massive price cut and DELL alliance, expect its unit share to rocket.

Conroe volume is negligible

Intel will ship 1 million Conroe CPUs in 7 weeks, or 0.14 million per week. Since the world's weekly PC consumption is 4 million, Conroe represents less than 4% of the market. It's non zero, but close. 550 PC makers will produce Conroe desktops, 1,000,000/550/7/7 = 37. On average, each PC maker will get 37 Conroe CPUs per day. Now you know another reason why Michael Dell has to go AMD. Only AMD can save Dell from going down like Enron.

Looking at Intel's new price list starting from $30 for a Celeron, I seriously question the IQ of Intel management. There is only so much demand, once you sell a chip at $30 and at a loss, the market shrinks by 1. Intel's inventory has piled up to the neck - $8.66 billion worth of legacy chips and more are produced every day. At the end, a lot of these legacy chips will be sent to the landfill. A rational move should be selling the higher priced CPUs first, and give up the low end crap -- they will be written off later.

AMD's Sempron starts at about $50.

I reiterate my projection that Intel will post operating loss starting 3Q06, and BK in five to seven quarters.

Monday, July 24, 2006

AMD's denying move

As I proposed in this article titled "AMD: time for the kill", AMD must deny Intel the oppurtunity to unload its Netburst inventory at a good price. Specifically, I proposed that AMD should have low priced dual core to stop the Pentium D dumping, and a low priced Sempron to deny Celeron the market. Now, AMD has published its new pricing list. The AM2 X2 3800+ is priced at $152 and the AM2 Sempron 2800+ is priced at $47. With this pricing, one can safely say, Intel's $8.66 billion Netburst inventory is trashed overnite. And, Intel's FABs are producing more and more unwanted Pentium D chips.

AMD+ATI analysis: mobile is the key

My previous analysis on July 21, 2005 on AMD+ATI merger had a lot of unknown parameters, now the details of the merger are out, let's take another look. None of the analysis will change the projection that Intel will be slaughtered, but we will have more accurate insight on the near term impact.

I. Financial impact on AMD

AMD is going to pay $4.2 billion cash and issue 57 million shares of AMD stock. AMD will take a loan of $2.5 billion. For AMD shareholders, this is a great deal. With about 12% dilution, AMD now also becomes the world's largest GPU maker. This is a major boost to AMD's share value. Since ATI will add more than 12% in profit, the EPS impact on AMD will be positive.

AMD's cash position will be tough. However, Hector Ruiz must know something we don't, which makes him extremely confident. The DELL-AMD contract must be huge. I won't count AMD's monetary recovery from its anti-trust lawsuit at this point, but that's at least a few billion dollars in the bag.

The deal won't close until end of 2006, or even 1Q07. So there won' t be an impact for the current quarter. We should expect AMD's revenue and profit to increase for 3Q06.

II. Impact on AMD vs. Intel death struggle

The deal is definitely a major win for AMD. We should expect some major innovations come out from the AMD+ATI merger. This is not just about putting CPU and GPU on the same die, but fully integrating them in computing. Now AMD can provide full set of solutions in the PC: CPU+chipset+graphics at the highest end of the spectrum for commercial client and mobile market. Again, Intel is left to eat AMD's dust.

AMD(ATI)'s mobile chipsets will be a major boost to the Bulldozer platform due next year. The timing is perfect. Bulldozer will frag Merom like there is no tommorrow. INQ reported those Merom chipsets consume 13 watts. I expect AMD to FAB its mobile ATI chipsets on the same lower power 65nm SOI process in FAB36. Since mobile is the fastest growing segment of the PC market, Bulldozer+AMD Chipset+AMD graphics will be a major major frag.

One thing Intel will also hate is, AMD Crossfire Chipset+GPUs will milk the Conroe market for profit.

ATI had about $100 million revenue from Intel chipset business, that's a small piece to be thrown away. The revenue made from expanding the market (especially mobile) will be billions.

III. Impact on Nvidia

AMD will not ruin Nvidia's business. Nvdia owns 90% of the AMD64 chipset business with its proven Nforce technology. But this business is mainly in desktop and server, as AMD's mobile market share is very small, and ATI already has a big footprint in mobile AMD chipset. Therefore, expect Nvidia to continue to dominate the AMD desktop chipset business.

However, AMD+ATI merger does put Nvidia in a very tough position. Nvdia has nowhere to go, but to fully commit to AMD64.

(1) Nvidia can't get Intel's highend desktop/mobile chipset business. Intel has 100,000 mouths to feed and a large number of those depend on Intel's own chipset business.

(2) Intel won't and can't acquire Nvidia. Intel already owns 35% of the graphics market, adding Nvidia will cause a lot of overlap and will not pass HSR Act review.

(3) In the discrete graphics market, Nvidia's competitiveness will be dependent on performance, which is unrelated to this merger. On PCI-E, both Nvdia and ATI are on equal footing on both AMD and Intel platforms. On Torrenza, I expect AMD to license Torrenza to both Nvidia and ATI, even if there wasn't an AMD+ATI merger. If ATI produces some major innovations on Torrenza, Nvidia will be impacted--but that's not a result of the merger.

(4) Therefore, Nvidia's only oppurtunity lies with an expanding AMD64 market.

IV. Impact on the PC industry

Now everyone can depend on AMD to provide the full set of solutions to break the Intel monopoly. Expect AMD to ODM motherboards from Taiwanese and the circle is complete.
AMD will make CPU+hestsink/fan+chipset+graphics+motherboard. The only things you need to make a PC are memory, hard drive, case and power supply.

The HP+Compaq merger saved HP from certain death, HP and Compaq was a weak+weak merger.

The AMD+ATI merger is a strong+strong merger. #1 CPU technology vendor merges with #1 GPU vendor, very little overlap. It's a killer enterprise.

Sunday, July 23, 2006

Expect AMD+ATI and AMD-DELL announcement at the same time

AMD must provide one stop solution under one vendor to support the commercial client market, CPU+Chipset+GPU+MB. That's why AMD needs ATI.

Once AMD has the whole package inhouse, DELL can eventually go 100% AMD: from notebook to 32 way server.

Right now, Intel owns the commercial space because it can provide all these things by itself. AMD is working with Nvidia on stable platforms, but that is still a multi-vendor solution.

AMD told us that 2006 is the year that it will break into the commercial client market in a big way.

Expect AMD+ATI and AMD-DELL announcement to happen at roughly the same time -- which also coincides with Conroe launch.

Friday, July 21, 2006

AMD and ATI merger may be a killer enterprise

Since Intel is out there trying to kill AMD, AMD's only way out is killing Intel. You are either quicker or dead, dead or alive.

Once AMD and ATI are merged, AMD can do a lot of interesting things in the GPU market

0) Use Z-RAM for graphics memory
1) Optimize ATI chips for HyperTransport (Torrenza)--You don't need GDDR3 anymore, just DDR2 via ccHT
2) Cripple ATI chips for FSB based technologies but keep them competieve against Nvidia ones
3) Charge lower price for Torrenza version which is higher performance and lower cost
4) Deny Intel licenses to ATI technology
5) Embed GPU functionality into the CPU, making GPU running at 2.8GHZ
6) Enable high end graphics on every AMD PC
7) Merge with Nvida too
8) Deny high end GPU to Intel market completely
9) Finish off Intel like the allies finished off the 3rd reich.

No one in the IT industry likes Intel. Nvidia certainly doesn't like Intel. Even the game developers hate Intel for pushing those crap Intel Extreme graphics to the market. With AMD+ATI tie up, I think it's just natural to make one step further and merge AMD+ATI+Nvidia. After that, all high end GPU chips will be either embedded or on Torrenza. Intel can enjoy 100% market share for GPU solutions for Intel platforms and become a monopoly of Intel graphics solutions. That will be happy ending for everyone.

There are some questions regarding this rumored merger.

1) Timing, why merge now? Possible reasons:

a) AMD needs a CPU+Chipset+Graphics solution to penetrate commercial market (DELL)
b) Intel was trying to acquire ATI, AMD had to prevent it
c) ATI has some great stuff for Conroe, so AMD decide to profit from the Conroe market

2) How can AMD afford the merger? Possible ways:
a) Borrow a lot of debt and pay cash
b) Pay the extra with cash, then do a stock swap
c) This deal will close in 4Q06 the earliest, HSR Act approval, shareholder vote, many other things, bet on huge 3Q06 earnings to prop the AMD stock through the roof.

Look at AMD's balance sheet, I found it very hard for AMD to take over ATI at this point.

So my guess is this: AMD will purchase a portion of ATI to do integrated graphics.

If AMD does take over the whole ATI, then AMD has shown its roadmap and plans to some rich dude, and the rich dude will put down $5 billion at 1% interest rate, convertible to AMD stock at $40 in 10 years.

Thursday, July 20, 2006

AMD ready to crush Intel in a price war

Chinese independently discovered that Con XE 6800 2.93GHZ leads AMD64 FX62 2.8GHZ by less than 9% with 4-4-4-12 RAM.

In other news, IBM is going Opteron in a bigger way.

Intel and AMD's pricing after July 24 will look like the following:

Since an Athlon 64 X2 3800+ frags a Pentium D XE 965, the pricing table above simply makes all Netburst CPUs 100% undesirable from all aspects. You should note that 90% of Intel's units in 3Q06 are Pentium D, Pentium 4 and Core32.

The fact that only AMD's FX CPU has 2x1MB cache indicates that AMD's yield is near 100%. Basically, AMD puts in a wafer and says let them be FX62s, and they all become FX62s. With APM3.0, AMD is super efficient. I expect the X2 4200 to become the top selling CPU.

AMD posts 53% year/year revenue growth

AMD posted its 2Q06 results, showing 53% revenue growth over 2Q05, compared to Intel's massive 13% year/year drop. Stockholder equity now stands at $4.85 billion, an over 40% increase from 4Q05. . Excluding inventory, current assets minus current liability is $1.8 billion (compared to Intel's $4 b). Current ratio stands at 2.3 (Intel: 1.98).

AMD expects a strong 2H06. Specifically, Hector Ruiz reckons that 3Q06 (the current quarter) will be above seasonality no matter what. Why? You ask. As I analysyed before, DELL will go AMD full force around the time of Conroe launch. Since Conroe launch is scheduled on July 27, 2006, AMD will see DELL selling AMD64 products for two full months of Q3.

On technology front, 65nm products and will start shipping and K8L quadcore will be demoed by the end of year.

Wednesday, July 19, 2006

Some entertainment

Someone asked me to do another phone call transcript, but I don't think I can outdo this one here -- Intel's 2Q06 CC transcript. If you have the time, you should listen to it, much more fun than Jay Leno. One thing I noticed was that Joe Osha was allowed in after he called for Paul O's head lat Q and was banned. Joe asked a philosophical question which seem have irritated Paul.

John Lau asked a DELL-AMD question. Adam Parker got quite emtional, I wonder why. Any way, Paul O and Andy B both think $4.33 billion inventory is very healthy.

Intel 2Q06 revenue dropped 13%, profit dropped 60%

Intel's 2Q06 result is out, far below analyst expectations. Worse of all, its 3Q06 outlook is only $8.3 to $8.9 billion revenue. Mid point $8.6b, 4Q06 follows seasonal pattern, 8% increase over Q3, so it will be $9.3b.

One thing I was wondering is Intel's sale of the coms division to Marvell. I hope Andy Bryant didn't book a profit from that. It should be at least $1.6 billion impairment on goodwill. Any way, I couldn't find it from the report. There is no loss of goodwill shown in the balance sheet.

Looking at the balance sheet, Intel is fragged for good. Inventory is now at $4.332 billion at cost, increasing 21.4% from 1Q06. With a margin of 49%, that's $8.66 billion of chips. I laughed out loud when Andy Bryant said that was only a whole quarter of sales. Only $0.2 billion inventory is Conroe. Its designers are working in bomb shelters. 3Q06 inventory level will go up further. 4Q06 gross margin may be even lower. Those Pentium D chips are piling up, folks. Cash is reduced by $1.2 billion, long term investment reduced by $0.4 billion.....Excluding inventory, current assets minus current liability is $4b. Total long term liability $2.8b, long term investment $3.5b. One step closer to cash burn out. Keep in mind that Uncle Sam says Intel owes $2b taxes.

Man, this company is going BK faster than I projected.

From this graph, you can see after AMD filed its lawsuit in 2Q05, INTEL started going straight downhill.

AMD is definitely gaining market share. Now, AMD's revene share of the CPU market is 18.6%.

I admire Hector Ruiz

The guy is a great military strategist. The massive attack with $149 X2 3800+ and even lower X2 3600+ coupled with DELL's massive volume capability will truly deny Intel the oppurtunity to clear its legacy inventory and result in massive operating losses for Intel. AMD is gonna ramp FAB36 and Chartered FAB7 like crazy. The goal is clear, deny Intel from selling Pentium Ds at a price above $90 at all costs. Since an X2 3800+ frags Pentium D XE 965 at much lower power consumption, AMD's move will make the Pentium Ds (which will remain 75% of Intel's volume) 100% unwanted.

I declare Intel clinically dead in 5 qaurters instead of the previous 7 I projected.

There are some hand waving arguments by some people on AMD's capacity, which are clueless. AMD's chip die sizes are always similar to Intel's. The Pentium D and Conroe's die sizes are about 160mm^2 at 65nm due to their 2x2MB cache. AMD's X2s have die size of about 180mm^3. Assume AMD is now at 25% X2 and 75% single core, the average die size is 117mm^2. Suppose AMD starts to produce 65% X2s and 35% single cores, the average die size is also 158mm^2. The increase in average die area is 35%. However, look at page 5 of this slide, you see AMD's wafer output increases by 50% in mid 3Q06, add another 10% increase from Chartered, AMD's chip unit output will increase by a factor of 1.6/1.35= 118.5% by mid 3Q06. By 2Q07, AMD's total wafer output doubles, however, because FAB36 is 50% converted to 65nm, the total equivalent output will increase to 2.5x of current level, leading to 2.5/1.35 = 1.85x current unit volume, throwing in some more from Chartered, 2x output is an easy goal. That's not all, AMD's capacity will go straight up from 2Q07 till 2009.

It's impossible for Intel to grab more market share, since an X2 3800+ frags a Pentium XE 965, so Intel will continue to lose market share quickly. Intel's ASP was $150, and AMD seems to have deliberately priced X2 3800+ at $149, below Intel's ASP. For Intel to maintain its ASP, for every chip it sells at $100, it has to sell another at $200, which is impossible, because Conroe's volume is limited. Lower units coupled with lower ASP, will lead to massive operating losses for Intel.

For AMD, its ASP now is below $100, so every chip it sells at above $100 is a winner. Actually, if AMD can simply sell one X2 3800+ at $149 plus one Sempron at $51, its ASP will hit $100 and an increase from current levels.

Woodcrest out in the woods

Woody, woody, where are you?

You were born 25 days ago,
but no one heard you wail,
except the Sam the Uncle,
who sent you back to Intel.

Tuesday, July 18, 2006

INQ and I are in agreement on Itanium

I agree with their insight quite often, Charlie at INQ wrote his analysis on Itanium, which I agreed in March 2006 more eloquently in a post titled "Intel execs are brain dead". But unlike Charlie, I gave Intel the benefit of doubt and assumed that they sell Itanium at $5000 with a cost of $4000. Let me repost my old article below:

Intel execs are brain dead
SUN and IBM make all the money on Sparc and Power, they are system solutions providers, they do the whole hardware and software stack, CPU, server, networking, storage..., operating systems, enterprise software stack and services, the gross margins are very high.

What does INTEL get for Itanium?

Even at $5000 per CPU, Intel is only getting $1 million bucks for the 1000 CPUs it is selling. Right now, HP is making all the money on Itanium.

With its primitive copy-exact methodoly, Intel is tying a FAB line for making a few hundred chips a quarter. This is bad business.

However, without Itanium, Intel becomes a pure AMD64 clone maker. Intel has to continue Itanium for the last bit of pride it has left.
Posted by: sharikou Posted on: 03/23/06

SGI is a reflection of the Itanium

SGI became a pure Itanium play, and it went bankrupt because of it. Intel is proudly presenting SGI as the proof that Itanium rules.

I wonder why Intel didn't invest a couple million bucks in SGI, so the world's only pure Itanium backer can at least feed its workers without incurring debt.

Intel in turmoil and facing possible financial disaster

Tommorrow, Intel will post its 2Q06 financial results. As I analysed previously, GAAP loss is a substantial possibility. I also expect substantial decline in net cash and stockholder equity. The expected revenue was $8.3 billion, down from $8.9 billion in 1Q06. Some analysts projected a revenue of $7.8 billion.

INQ reported some stories from inside Intel. Some very competent but outspoken folks lost their jobs at Intel. I can imagine that place is full of fear and uncertainty, everyone is an enemy. If your co-worker loses job, your chance of staying increases. Especially for those folks who spent their youthful years there, their chances are slim. Craig Barret once said that "[t]he half-life of an engineer is only a few years." Those who lose their jobs will find it hard to find another one -- AMD is not hiring that many.

AMD already forecasted that its 2Q06 revenue will be $1.21 billion, down 9% from 1Q06. Looking forward a few months, 4x4, 65nm, K*L, Bulldozer. Intel won't be able to walk out 2007 in one piece.

Monday, July 17, 2006

AMD64 4x4 ready to frag Conroe XE by 60%

INQ reports that AMD 4x4 will boost performance by 80%. Since the 2.93 GHZ Conroe XE 6800 is about 10% faster than the 2.8GHZ Athlon 64 FX62, the 4x4 will outperform Con XE 6800+ by about 64%. In other words, Intel needs at least a Conroe at 2.93*1.6 = 4.8GHZ to compete against 4x4 at 2.8GHZ. As far I know, 4.8GHZ is unattainable with Con XE even if liquid helium is used for cooling.

As I have pointed out, 4x4 is a permanent solution to pin down Intel at 50% of AMD's performance or less by taking advantage of the Direct Connect Architecture.

Saturday, July 15, 2006

Another missing Conroe benchmark: stability

We heard stories about Woodcrest servers crapping out all over the place and got rejected by US government. What about Conroe, which is essentially the same chip?

INQ reported its first hand experience with the new Intel chip.

Compared to Athlon 64 FX62, the 2.93GHZ Con XE 6800 shows 25-31.8% lead in FarCry, 0% lead in FEAR, 6.7% lead in Quake 4, 10.7% lead in UT 2004 16-bot match. With the exception of FarCry, the CON XE 6800 and Athlon 64 FX62 are within 10% of difference. With K8L promising massive performance and scalability increases over K8, we will see AMD commanding a 40% lead over Conroe in just a few months time, even before Intel is able to ramp Conroe production to 25% of its volume.

Compared to Pentium XE 955, Athlon 64 FX62 leads 47% in FarCry, leads 30% in FEAR, leads 65% in Quake 4, leads 60.9% in UT2004. No wonder an Athlon 64 X2 3800+ can frag a Pentium Exteme Edition 965. However, I don't remember ever hearing the kind of excitement before from Anand, Tom or anyone else about AMD's massive lead over Pentium XE.

Athlon 64 FX62 defeats Con XE 6800 in ScienceMark 2.0, demonstrating superior floating point performance. This result is consistent with the SpecFP_rate_2000 results, where an 2P Opteron 2.6GHZ outperforms a 2P Woodcrest 3GHZ. Woodcrest is for entry level 2P market only. x4600 with 8P Opteron outperforms 16P HP Superdome.

As I pointed out earlier, the Conroe benchmarks prove only one thing: the $7 billion Pentium 4/XE/Celeron inventory Intel has at hand and was touting is a pile of crap and now Intel admits it.

But that is not all, read what INQ had to say on stability of the Conroe:

"However, there is still one thing that disturbs me about the Intel Conroes. The load times on our Far Cry tests simply took ages, regardless of whether we used a NetWurst craptecture or Core marchitecture. AMD loads FarCry in a matter of seconds, and in this particular game, we feel like watching Athlon 64 to complete SuperPI after running Conroe for three years. B-o-o-o-ring. A big surprise for me was the fact that Quake 4 gameplay also experienced more hiccups on Conroe platform than it did on a competing AMD platform."

I suggest SuperPi runners always check the digits output from Conroe and make sure the numbers are right.

Friday, July 14, 2006

The missing Conroe benchmark

Right after FACE Intel's field day, Intel fanbois had their field day as 32 bit Conroe benchmarks popping up yet again for the 100th time. Nor surprisingly, Conroe's lead over Athlon 64 diminished a notch again. A previous 20% lead is now reduced to 5-15% lead. But the noise this time gets louder.

Where are the 64 bit benchmarks? Vista is coming in just a few months, and there is no 64 bit benchmarks for Conroe? Or, only Japanese know how to do 64 Conroe benchmarks? (shows Conroe 40% slower running 64 bit code, EM64T really stands for emulated 64 bit?).

The Intel fanbois will tell you: "I am happy with 32 bit, I will never need 64 bit, I will stick to IA32 for the rest of my life".

The new Conroe benchmarks only proved one thing: the Netburst P4s Intel was pumping are total crap. But Intel is stuck with $7 billion of that crap and is still producing majority of its chips with that architecture. By the time Conroe reaches 40% of Intel's volume in five months, AMD's 65nm products will be out. Before that, AMD will sell two Athlon X2 AM2s frag the Conroe by 70% with the 4x4 platform.

According to this, AMD's 65nm parts will out for testing in August.

Thursday, July 13, 2006

FACEIntel got a boost today

FACE Intel had a field day today, adding 1000 experienced managers. Intel was also a winner, saving 1000x license fees on Powerpoint and MS Project. Intel needs to sell 10 top end Woodcrest CPUs to feed one manager for one month, there aren't enough chips or demand out there.

AMD and Intel are in a life and death struggle. Intel is slashing its own throat to hurt AMD. AMD is rapidly expanding capacity. If AMD manages to sell its capacity at any price, Intel will suffer massive losses and BK in a few quarters. Assuming both AMD and Intel sell CPUs at $1 each (giving them away), AMD can last one year and Intel can last one quarter. The average cost of an AMD CPU is $40, the average cost of an Intel one is about $70. If I were Hector Ruiz, the decision would be simple: cranking up capacity as fast as possible and sell at any price: deny Intel the oppurtunity to unload Netburst inventory at any cost, also do the decisive battle before Intel successfully laying off 50% of its workforce.

Suppose AMD successfully grabs 35% of the PC market, leaving Intel at 65%. Suppose both AMD and Intel sell at $50 ASP for CPUs. World's quarterly CPU consumption is 50 million. This will lead to AMD having $0.9 billion quarterly revenue. Let's assume Intel's ASP for chipsets is $25 and Intel has 100% of the Intel chipset market. Intel's quarterly revenue will be ($50+$25) * 50mil * 0.65 = $2.4 billion. At these revenue levels, AMD is close to break even and Intel will suffer $4 billion loss per quarter. As a result, Intel will BK in less than two quarters. Intel has run out of cash.

The choice for Hector is clear. It's to AMD's ultimate advantage to fight the price war at all costs. Remember it was Intel who started this war, but now it's AMD's job to end it.

I saw that Athlon 64 3000+ was being sold at $65. I suggest AMD sell Athlon 64 x2 3800+ at $135. AMD should also sell Turion X2s at around $100 to disrupt Intel's mobile profit engine. If Intel reacts by cutting price further, I suggest AMD sell X2 3800+ at $75. What da heck. It's war, you are either quicker or dead.

Wednesday, July 12, 2006

U.S. House passed CPU electricity bill

The very first article I wrote for this blog was titled "Why isn't DOE imposing power consumption limits on CPUs?" I emailed that article to SUN, AMD and Intel's execs. I got feedback from some of the folks.

I am glad to know that the lawmakers passed a related bill today. I expect the Senate and the President to rubber stamp this -- unless Intel spends millions lobbying against it.

Once the bill becomes the law, DOE can do something about it. Those who use CPUs rated above 125 watts should be required to pay energy taxes. Also, any company who attempts to cheat US government using unproven chips that crap out all over the place should be heavily fined.

Tuesday, July 11, 2006

SUN Galaxy x4600 smashes Superdome

Some say SUN's servers are too expensive. Let's check SUN's pricing on the 8P 16 way 4U rackmount x4600 server, which outperforms HP's 16P Superdome with Itanium 2 1.5GHZ by up to 20%.

For $67,495.00, you get 8 Opteron 885 2.6GHZ dual core CPUs, 32 GB memory, 2x HDD with 4GbE ports. For $35K, you get 4 processors and 16GB ram. In case you don't know, the 16P Superdome costs over $1 million.

Go to and try to configure a 2P Woodcrest server with the same amount of memory, your total runs up to $40,000.

Why are people paying $1 million for one 16P Superdome, an amount of money that can buy 25 fully loaded 2P Woodcrest machines? Are they stupid? No, they pay 25x because the 16P Superdome is 25 times more productive than a 2P Woodcrest. One wolf is more powerful than 25 sheeps combined -- because sheeps don't have technology to integrate the power of 25 sheeps into one wolf.

But, the x4600 is 20% faster than the 16P Superdome, and only cost slightly higher than a 2P Woodcrest. Folks, this is something fundamental.

The SUN x4500 is also reasonably affordable for most people. You get 2x Opteron 285, 16GB ram, 24TB storage for less than $70K. If you buy 10 of these, the price drops to $47K per box. The x4500 with Solaris 10 ZFS can sustain 2GB/s read data rate, which means the server can examine every byte of the 24TB in just three hours. Suppose your average file length is 1GB (such as those huge Yahoo mail mbox files), and your application needs to search within the file, the x4600 can do it in 0.25 second on average.

I think that the x4600 is perfectly suited for the online video industry, one such box can store 20,000 DVD quality movies (compressed), or 40,000 VHS quality movies. There are zillions of web sites that are streaming video, each of them needs at least one such box. Maybe SUN can bundle some video streaming software with the box and make it a turnkey solution.

SUN X4600, X4500 servers and 8000 Blades frag everyone else's x86 system

Folks, checkout the new SUN x64 products, the X4600 8P 16 way server, the X4500 4U server with 48 SATA disks, the Sun Blade system 8000 with 4P 8 way blades, take the virtual tour and examine the insides. The SUN folks are so clever. If I wasn't so impressed by x4100, I am definitely awed by the creativity Andy Bechtolsheim showed in these new products.

I was wondering how one can fit 48 hotswappable disks into 4U and I guessed right (well, that's the only way possible). 48 SATA disks requires a lot of bandwidth, over 20GB/s, for system-to-disk I/O alone. No Intel server with obsolete FSB technology can handle a fraction of that. Only AMD64 technology can enable such massive amount of I/O and still able to pump the data over the network. Each Opteron 2xx CPUs has two 8GB/s non-coherent HyperTransport links, a 2P X4500 thus have a total of 32GB/s I/O bandwidth. My understanding is SUN is taking full advantage of the HT links found in Opteron 2xx CPUs. One application for x4500 immediately come to mind is mail and online forums. Yahoo can serve 100K users with one of these boxes.

The X4600 8P 16 way server and 8000 4P 8 way blades are equally impressive. Take your time to examine them. The x4600 supports 8 socket 940 Opterons, but it can be upgraded to quadcore Socket F Opterons by swapping the CPU modules. No HP machine can touch these. According to SUN's published benchmarks, the 8P Sun x4600 beats the best 16P HP Integrity Superdome Itanium 2 server in both integer and floating point performance by up to 20%*. DELL? Not even worth mentioning, Intel doesn't even have scalable 4P technology to start with.

These SUN products kick ass!

Monday, July 10, 2006

SUN and HP make me laugh

Read the exchange here. It's amusing.

Any way, SUN's announcement will be huge.

AMD64 everywhere is the vision endorsed by SUN. Sparc will co-exist with AMD64 via Torrenza. HP better give up its Itanium soon and brace the future with 100% AMD64.

AMD and SUN have a lot in common -- great technologies, very smart and creative people, unquestionable ethics.

DELL laptop explosion was a chip problem

INQ reported that DELL's investigation showed that the Intel laptop explosion was caused by a chip problem, not battery. As I previously analysed, the incident was probably caused by an exploding Intel Core Duo processor.

The buzzword is customer-centric innovation

Even Michael Dell is following Hector Ruiz's "customer-centric innovation" doctrine. This must be a big change for Dell.

There is no doubt that going forward, AMD will dominate the computing landscape and AMD64* will become the industry standard, not only because AMD64 is vastly superior than IA32, but because AMD will grow the ecosystem and allow others to prosper within the AMD64 food chain. The news that SUN Sparc and AMD Opteron will share a common socket is huge and has profound implications to the enterprise. Imagine that you can buy a single SUN server that runs SPARC and AMD64 at native speed the same time. AMD's ccHT will be the fabric that interconnects different types of compute cycles, like CISCO routers that connect computers over the internet. AMD's Torrenza will combine and multiply the forces of its partners.

* AMD64 is not just about the AMD64 instruction set, which Intel has partially copied, it is more about Direct Connect Architecture.

Friday, July 07, 2006

Intel Core 2 self-fragged: stability and reliability problems with Woodcrest and Conroe

It's been 11 days since Intel launched Woodcrest, and the chip is nowhere to be found. However, before we could possibly get a Woodcrest to do some real drill, the Intel Core2 marchitecture reportedly fragged itself. It seems that in a rush to catch up with AMD, Intel has cut some corners with CORE2, and its next gen Woodcrest chips would come up with unexpected errors and run into undefined behaviours during normal operation*. We have seen reports that Woodcrest servers suffer from BSODs before. But we are talking about production Xeon 5100 systems delivered to US government for evaluation here. Woodcrest is definitely not your chip for running a nuclear reactor or a heart surgery.

You say, what about playing games? Well, Conroe reportedly posted some good frame rates, but no one has verified that Conroe was playing the games correctly. I am not even sure people verified that their Conroe got the right SuperPi digits. Intel quality is very much in question here. If Intel sent government of USA a bunch of bug laiden server chips and hoped to get millions worth of contracts, what kind of confidence can you have for Intel's consumer chips?

Since there wasn't any report of similar problems with Dempsey/Bensley, the Woodcrest stability and reliability issues appear to be intrinsic and specific to the Woodcrest/Bensley platform. Clearly, Intel America has a lot more server chip design experience than the Israeli folks, who spent all their careers patching up Pentium III.

So we have a bunch of amateurs who have been working on 32 bit laptop chips all their lives to come up with a 64 bit server CPU, we know it is not going to be easy. I personally expected Intel to announce delays for Conroe/Woodcrest. Since Intel's server chips are based on desktop designs, Intel normally launches corresponding server chips many months after the desktop part is out. For instance, the Xeon Paxville was launched eight months after Pentium D. Yet, the marketeers at Intel decided to push out a Conroe dressed up as Woodcrest Xeon 5100 in June, even ahead of Conroe launch. Such a reckless move is deemed to fail -- it's like politicians directing hopeless military efforts. Intel has lost the server market period, to challenge the advancing AMD forces with unproven troops will only invite greater defeat.

I won't be surprised if the Woodcrest has messed up some cache coherence logic in association with its complex memory preloading mechanism. Intel's competition, AMD, is a server CPU company from ground up.

I smell a Woodcrest recall soon.

I also expect Intel to send a pile of manufactured buggy Woodcrest chips and Woodcrest wafers into the dumpster - or maybe they get remarked as Conroe and sold in China?

Intel is so predictable.

* Please note that reported problem was not RAID alone. RAID was causing performance issues, but Woodcrest was crapping out all over the place. As INQ wrote: "Worse of all, several problematic situations occurred during the trial period and Intel was heavily criticised in internal memos, all ending up in a really unexpected manner."

Monday, July 03, 2006

Intel brags about Woodcrest's 83MB/s memory system

According to Intel's featured quote, Woodcrest has an amazing 667 megabit per second (or 80MB/s) memory system which once again raises the bar. Seven days after the Woodcrest launch, finally, someone in NJ ordered two chips - ETA August 2006.

In other news, Woodcrest's unborn brother, the Conroe, was touted as the SuperPi computer in a PC by some Intel investors.

Sunday, July 02, 2006

Monopolistic Exclusion and Nazism

Another report alleges that Intel excluded AMD from the PC market in Germany. German retailer Metro tells its suppliers that it has an agreement with Intel which bars them from purchasing products containing AMD CPUs.

I think this kind of exclusionary acts are worse than price fixing. Exclusion completely denies competors' access to market. In the movie "The Pianist", the Nazis denied some people's access to food regardless of what they were willing to pay. It's a very similar situation here.

There are three main features of Nazism: (1) A sense of absolute superiority and strong hatred and contempt towards competition (2) A urge to exterminate the competition regardless of the cost (3) A methodical and even scientific plan to carry out the extermination.

One can compare monopolistic exclusion to Nazism. Hitler had a philosophy of racial exclusion and Aryan monopoly of racial existence. Monopolists have similar doctrine in the market. Hitler wanted to wipe out competing races by violence. Monopolists don't want fair competition, they only want monopolisitc control. Intel execs' pathetic behaviour* and their fixation on excluding AMD may be traced back to their deep rooted hatred towards competition.

Reading the book "Inside Intel" by Tim Jackson, you know that Intel has a ruthless and often unethical way of doing things, internally and externally.
Intel doesn't look like an American company in many ways. Intel's corporate culture was cultivated by Andy Grove, a survivor of Nazi and Soviet oppression. During his times, even high level Intel execs are routingly shouted at and chided in meetings and punished by being put into CAP ("corrective action program"). Intel had a system to constantly weed out the least productive 10% of its work force, which led to the creation of the FACE Intel organization.

Today's Intel is the same old Intel. Its execs are still following the old way of doing business.
Intel is not expanding its horizons into other areas, it is not focusing on growing revenue and profit. Instead, Intel is solely working on gaining CPU market share, or reducing competitor's market share, even though Intel already has 75% of the market.
They are again trying to restrict free trade and commerce and limit people's freedom to choose by exercising their monopolistic powers.

There is a fundamental difference between expansion and exclusion. AMD expands the computing ecosystem through efforts such as pervasive 64 bit computing, x86 everywhere, 50x15, multicore scalable systems, HyperTransport Consortium, Torrenza, GreenGrid, AMD Live!... Intel always tries to exclude competition through programs such as Intel Inside(stop buying AMD and get $$$), Centrino (Intel CPU+Intel chipset+Intel Wi-Fi), VIIV(Intel CPU+Intel chipset+Intel LAN card)...It's no surprise that AMD is growing and Intel is shrinking.

Intel execs openly said they are angry. I think they are not just angry, they are hateful. Such attitude has even poisoned the Intel user community, where fanboism often turns into fundamentalism.

The Third Reich's collapse was swift.

* Henri Richard:
"I think that Intel executives refusing to attend a meeting, for example, if AMD is on the menu, or on stage, is just simply pathetic, but it happens time and time again. I don’t think that a company that’s worth US$120 billion or more, that’s one of the top Fortune 500 companies, should ever condone that kind of behavior, and I don’t understand why Intel is not welcoming free and open competition, especially from a much smaller competitor."