Friday, June 30, 2006

Michael Dell: M & D

Read this on M & D. (Monopoly and duopoly)

Mr Dell smiles when asked about the decision, earlier this year, to finally start offering AMD chips and the effect that AMD's rise to prominence has had on chip prices. Intel has recently been stung into slashing prices. "When you have a duopoly rather than a monopoly, you have very different pricing. And that is good for consumers."

I now recall AMD said Dell was a victim of Intel's monopolistic behaviour.

I projected that DELL will go AMD across the board around Conroe launch time.

Thursday, June 29, 2006

Thinking about setting up a Woodcrest vs Opteron duel

What about an Opteron and Woodcrest duel?

There will be no SunGard ACR Intel Demo Version 0.1 guerilla benchmarketing stuff, no AMD crippling Intel compiler, just GCC, Linux (x86_64)+ PHP/Perl+Apache+MySQL+OpenSSL+Tomcat (add any open source stuff you want). I will set up the servers, connect them to the internet and allow people to login and check. People will sign up to get their time slots to run the benches, while others can observe the results via a web interfacce. I will invite the gurus visiting this blog to write the test scripts. Everything will be open, fair and reproducible. It will be an Open Source dual core duel.

The question is how do I pay for this. Maybe Intel and AMD can send their best, hyper tuned boxes. Or maybe we just buy an HP DL380 and an HP DL385. Or we setup a web page and put ads on it, and hope it will generate enough ad money to cover the cost...

The tests will be made as relevant as possible for real server usages: Web, FTP, MAIL, DNS, DB, Game Server, Video/Audio streaming...

This is serious business, folks. We can expand the idea from here to cover more products...

Wednesday, June 28, 2006

Woodcrest a paper launch

Originally, Paul Otellini promised a Woodcrest launch date of June 19, 2006. It was delayed to June 26, 2006. However, checking various online vendors via, all you get is a bunch of baggies. On, Woodcrest based servers have a ship date of July 18. IBM won't ship Woodcrest servers until August. On, the Woodcrest CPUs have a ETA date of July 31, 2006... If you see a Woody somewhere, let me know.

I remember Pat Gelsinger saying that he was going to ramp Woodcrest through the nose. So I expected at least 10 days of supply on launch. The world's total server shipment per quarter is only 1.5 million units, including non x86 ones. That's about 16,000 units per day. Assuming they are all 2P Intel, that's only 32,000 CPUs per day. Ten days supply is only 0.32 million chips. Judging from the availability, I would estimate that Intel accumulated less than 50,000 chips from June 19 to June 26. In other words, Intel is making less than 10,000 Woodcrest CPUs per day -- off its development FABs, I presume.

Even more surprisingly, with so much guerilla benchmarketing, two days after Woodcrest launch, not a single person made single purchase. As you can see there is zero back order for Woodcrest. Intel execs must be shouting at each other in panic now.

AMD sells 120,000 Opteron 8xx CPUs per month to Google alone.

Opteron 265 has a shortage, 917 back orders.

Plenty of Dempsey chips that no one wants.

Tuesday, June 27, 2006

Intel's Double Conroe will be literally hot

Clovertown will be 110 watts typical with a 1066MHZ FSB. Intel's 110 watts is like AMD's 150 watts.

You may wonder why the quad-core Clovertown has a slower FSB (Woodcrest's FSB is 1333MHZ). It is the same reason why Xeon MP has lower FSB than Xeon DP. On a shared bus, when you put more loads on it, the frequency has to be lowered due to interference from the loads. With Woodcrest, each core has 677MHZ bandwidth. With Clovertown, each core will have 1066/4 = 266MHZ (about 2GB/s).

Imagine you double the heatsinks (one for CPU, one for chipset) in the picture on the right, and you have Clovertown.

BTW: AMD's quadcore will be as cool as its dual core.

DELL Laptop Explosion Not Related to Battery

DELL claims that the Centrino explosion reported by INQ did not appear to have anything to do with battery issues.

I conjectured that a 53 watt Core Duo's thermal protection failed and caused two explosions.

AMD sells chips, Intel sells chips business

It's official, Intel sold its communications chip business for $0.6 billion. Since Intel paid a lot more for that business, it should record a large impairment on goodwill. Now, it seems GAAP loss for Intel's 2Q06 is quite certain. Going forward, I projected seven quarters of operating losses from 3Q06 onward.

Marvell was the buyer of Intel's money losing business. On the news, Marvell's stock fell $7.94 (15.30%).

Why did Intel bite the bullet? It has run out of cash.

I wonder who would buy Intel's Netburst and Itanim businesses.

While Intel generated cash by dumping businesses, AMD just sold 8064 Opteron 8xx CPUs to be used in 2016 4P Opteron servers. That's some nice profit there, but to AMD, it's a piece of cake. Opteron has popularized the 4P market with its low cost and high scalability. Google builds 30,000 4P Opteron servers per month, that's a nice 360,000 Opteron 8xx per quarter.

Intel's Woodcrest is for low end 2P market only. Did you notice that no tier 1 OEM cheered Intel's Woodcrest? They were all afraid of upsetting AMD, which commands the high end and where the money is.

Monday, June 26, 2006

Anand's pay day

Today must be Anand's pay day for this benchmark. I think he deserves this. DELL is buying AMD chips in the millions, and Intel is desperately in need of pumping. This is once in a life oppurtunity for review site owners to get rich. Intel now says on their web site that Woodcrest is up to 60% faster than Opteron, that seems to be Anand's MySQL negative scaling result.

Just looking at the above Intel web page, you will have to admit that Intel is lying in broad sun light. Intel is no server leader by any reasonable standard. First, even if its claims on Woodcrest were true, they were only for entry level 2P servers; AMD rules the high end with 4P and 8P Opteron. Second, we already busted AnandTech's bogus benchmarks, and and were simply repeating Intel's reasults. The independent GamePC benchmark on Apache web server performance show Woodcrest soundly beaten by Opteron.

The Opteron has changed the server landscape. Nowadays, a 4P Opteron server can be built for $5000. This has led to the rapid growth of 4P server market.

INQ reported that a Woodcrest may have suffered BSOD at its own launch demo.

Intel may bankrupt in seven quarters

DELL is going to market AMD like crazy. It's a business decision, business decision is based on money. Simple?

Intel execs are morons. They caused a price crash on Intel chips, instead of a price hike I proposed to them last year...What's the consequence of that? Dell's units and unit price will go down significantly. DELL's unit price will go down because Intel CPUs are dirt cheap now - a Pentium D 930 is only $100. DELL's units will go down because every small vendor can sell Intel PCes dirt cheap, DELL loses much of the pricing advantage...

Suppose DELL's PC unit price drop by 10% and units drop by 5%. The result will be 15% revenue drop. Can DELL survive that kind of spiral down? No. What can Dell do? Go AMD and expand the AMD market like there is no tommorrow. Since Dell's Intel business will suffer 15% drop, DELL must sell AMD like crazy to make up the difference. I have analysed this many many times.

Intel can be written off the book already. The prices of its chips have crashed by 60%, and units will be down significantly, as more vendors such as DELL are adopting AMD. The result will be continuing operating losses for Intel in the next 7 quarters. The end of that is an Intel at 1/10 of today's size, if it doesn't go straight to chapter 7.

Sunday, June 25, 2006

Henri Richard proposed a second life for Itanium

AMD's EVP said in this interview:

"Maybe the salvation for Itanium is to make it socket-compatible with Opteron. That would make it an excellent candidate for a Torrenza co-processing opportunity. That’s probably the only chance it’s got!"

Very interesting idea indeed. But I wonder what is good about Itanium besides its 24MB cache.

Woodcrest officially fragged: performance 10-20% slower than Opteron

Performance, Price/Performance, Performance/watt

In March, I projected that Woodcrest should be 10-20% slower than Opteron in server performance. Now GamePC has done some Apache web server* benchmark for Woodcrest 5140, and Woodcrest is a dog. The Woodcrest 5140 is 14% slower than Opteron 280. In fact, Woodcrest 5140 is slightly slower than Opteron 270 (2GHZ).

Unlike those review sites which live on ad money, GamePC is a system vendor that sells servers, workstations, desktops and notebooks, both AMD and Intel. In fact. you can pre-order Woodcrest CPUs from GamePC's customers include big names such as Dolby, Hitachi, Lockheed Martin, FAA.

The Woodcrest 5140 is to be priced at $470. Opteron 270 is available for $441.

The key here is 64 bit server performance. Woodcrest is simply no match to Opteron. Now, you understand why Intel had to pre-arrange benchmarks for people to push buttons.

In 64 bit file compression/decompression tests under Linux, a 2.6GHZ Woodcrest is 40% slower than a 2.4GHZ Opteron. In 64 bit ScienceMark MolDyn test, the Opteron 275 (2.2GHZ) is 31% faster than Woodcrest Xeon 5140.

According to this report, Conroe/Woodcrest's enhancements such ops-fusion won't work under 64 bit.

As for power consumption, Woodcrest is getting close to socket 940 Opteron levels. The Woodcrest 5140 system consumes 23 watts more than the Opteron 270. I noticed that the test systems only had 2GB of RAM, so the extra power introduced by FB-DIMMs was small. In a fully loaded system, the Woodcrest system will definitely consume more power.

*Apache runs 61% of world's web sites (Microsoft IIS runs 29%). Apache is like the swiss-knife of web servers, it has a highly sophisticated design that allows almost infinite features and flexibility.

Friday, June 23, 2006

Supreme Court rules out IA32

Supreme Court of the so called world's largest democracy rules that Turion 64 is the only legal mobile computing solution.

My Guess on Direct Connect Architecture 2.0

It will be a Massive Multi-Scalar. Basically, you direct connect all execution pipelines in a multi-core multi-processors system into one massive pool of computation resource.

You should notice that on AMD's slides, there is no mentioning of multi-core beyond 2008. Why? The following is my educated guess.

A future AMD64 CPU will be single core, but with massive number of pipelines, say, 128. Each pipeline will appear to be a processor to the software. But these pipelines can group and ungroup dynamically within a time quantum. So you can have a single 128-issue processor, or 128 one issue processors, or one 127-issue and one 1-issue, or 32 four issue processors, or any combination, depending on the current computation pattern. Not only this, you can mix and match different pipelines, say, 9 FP and 1 integer, or 16 intger and 0 FP.

It doesn't stop here.

With DCA2.0, multiple CPUs can merge their massive number of pipelines to create one giant multi-scalar machine. Imagine eight such CPUs, total 1024 pipelines....

OK. The above might be my dream, but I bet it's the future direction of CPU architecture.

AMD builds up Task Force Wipe Out

It's official. After announcing astonishing plans to expand capacity by 400% through FAB36 and FAB38 expansion in Dresden Germany, AMD now signed a deal to build a new FAB in New York.

This would allow AMD to completely displace IA32 with AMD64. Pax AMD is nearer. Prepare to hail Hector Caeser.

After K8L upgrade next year, AMD64 will be soon upgraded to Direct Connect Architecture 2.0 in 2008. Details on DCA 2.0 are lacking, but Hector Ruiz, CEO of Advanced Micro Devices, reckoned that it will be a real killer.

In other news, Henri Richard, EVP of AMD, pointed out the genetic defects of Intel, which I analysed previously here, here and here. Intel execs are definitely not bright. Cutting price by 60% to gain 3% market share is not just irrational, it's suicidal. I projected Intel's demise in 7 quarters.

The INQ reported about inverse threading we have been discussing.

Thursday, June 22, 2006

AMD to frag Intel in price war

Digitimes reported that AMD has decided to eliminate multiple models in preparation for war against Intel. For socket 939, the Athlon 64 X2 4800+ and 3500+ will be the only ones remaining.

The fact that AMD will only produce X2 4800+ is very interesting. It means that AMD's APM process is so advanced, all its chips are virtually identical. There is no bin split. AMD simply mark some chips at lower speed. That was why people can overclock an X2 3800+ to X2 4800+ speeds with stock cooler.

Using APM, AMD can rapidly, accurately and automatically make adjustments as small as 1nm to the critical dimension ... In every transistor, In every chip,
On every wafer. That's roughly 21 billiontransistors per wafer.

Quiz: who said this?

I'm following the George Patton philosophy. We're not going to defend [trash], we're just going to attack, attack, attack. We're going to attack ...everywhere. We're not interested in defending. Somebody asked [Patton], "Hey, should we dig a trench or a hole?" He said, "You do that, you die." Same thing here.

UBS analyst humiliated

UBS analyst Thomas A. Thornhill recently upgraded Intel to buy. dug out his track record. It's hilarious.

Wednesday, June 21, 2006

AnandTech Busting Update

I just added the motherboard information for AnandTech's test of Opteron against Woodcrest. Anand used the MSI K8N Master2-FAR for the Opteron.

AMD has a validated server program. It's worth noting that in the Intel commissioned Woodcrest vs Opteron tests, Intel chose to build the Opteron server itself.

Dell Laptop explodes: Intel Inside

Looks like a Core Duo inside.

Remember this? You could literally get yourself killed by being an Intel fanboi.

Tuesday, June 20, 2006

Woodcrest semi-officially fragged before launch

After massive amount of benchmarketing fraud, Woodcrest is set to launch in a week. A French site got hold of two Woodcrest CPUs (2.6GHZ, 4MB L2) and did some compression/decompression testing against one Opteron 180 (2.4GHZ DC, Socket 939) on the linux-2.4.32.tar.bz2 file. Since the Opteron 180 is a one way dual core CPU, and a 2 way Woodcrest has four cores, we should only look at the numbers for 2 simultaneous processes.

For decompressing the bz2 file, the Opteron 180 dual core CPU finished two simultaneous tasks in 10 seconds. The two way 4 core Woodcrest finished in 14 seconds. Opteron 2.4GHZ is 40% faster.

For compressing the tar file into bz2, the Opteron 180 finished two tasks in 46 seconds, the Woodcrest finished in 58 seconds. Opteron 26% faster.

This should be a very reproducible and verifiable test. You should be able to get the linux-2.4.32.tar.bz2 file from

I know Intel fanbois are going to complain that "bzip" wasn't compiled with an Intel compiler, which would have crippled the AMD chip.

Monday, June 19, 2006

Inverse threading for AM2?

According to this source, all dual core AM2 CPUs have built-in inverse threading capability, ready to be turned on by software.

The idea is simple enough: superscalar across multiple cores, but at a much higher level.

Looking at typical piece of code:

v1 = f1 ();
v2 = f2 ();
// then do something with v1 and v2

Here, we have 2 calls. Assuming they are well written, re-entrant and thread-safe, there is a good chance that they can be run in parallel. So why don't we run f2() on the second core when we start f1()? It's just like a remote procedure call on a different core.

All you need is some logic to detect such code blocks. It should be quite easy to identify all those C runtime and Win32 APIs that meet the condition....

I was thinking about filing a patent application, but a reader pointed me towards this AMD patent(6574725).. it was filed in 1999 and granted in 2003.

Friday, June 16, 2006

Pentium 4 has zero market appeal

There is an interest poll at It asks what will your next CPU upgrade be. Not surprisingly, 24% of people want Conroe, which should make Intel marketing smile. Their guerilla benchmarketing paid off.

Equally expected is that less than 1% of people want Netburst (P4 and Pentium D). It's reported that Intel will however introduce a new Pentium D 9xx stepping (D0) by the end of 2006. It seems Intel fans will be stuck with Netbust for quite a while as Conroe's 4MB cache will ruin yields. Presler's double die approach does have a yield advantage.

AMD's socket AM2 enjoys a healthy 34% demand, and socket 939 has 16% popularity. This indicates that most people know that Intel is hiding the truth about Conroe, and they would rather choose a proven high performance future proof platform.

Thursday, June 15, 2006

Intel choked distributors to death

Intel stuffed the channel to make its 1Q06 revenue numbers. It turned out that in that stuffing-da-channel process, some resellers were choked to death and BKed. According to this interesting report:

Hamid Kutty Shajahan, owner of the debt-ridden Micron Computers, has come out strongly against the "Intel distributor cartel" for what he has termed as an attempt to monopolise the market. He alleged that Intel officials used to pressurise resellers to take stock of slow-moving products along with fast moving ones.

"If we refuse, the Intel official would warn us that it could affect our IPP (Intel premium provider) status and our pending rebate payments," he pointed out.

... "Intel team comes to our office and persuades us to take delivery of huge stocks beyond our capacity. If we tell them we don't have enough credit, they talk to distributors to come up with some extra credit lines and dump the stock on us. We have been trapped like this," Shajahan said.

Another leading local computer dealer said an Intel field officer had persuaded him to take the Intel Celeron CPU without informing him that a price drop was due after a week. "By force they dumped the stock on our head which we have to sell 30 per cent below the cost price."

Read more at the source.

Wednesday, June 14, 2006

DELL: AMD64 the fastest growing notebook market

Joseph Marengi, senior vice president of Dell Americas, said that "[t]he fastest-growing segment of the notebook market, which was the fastest growing segment of the PC market, was for machines with AMD processors."

Since its introduction last year, the Turion 64 controls 100% of the 64 bit notebook market. The recent introduction of Turion 64 X2 brings dual core 64 bit mobile computing to the masses. HP has a set of consumer and business mobile solutions based on Turion 64 X2.

I expect DELL to go AMD full force at the time of Conroe launch.

In other news, it was reported that AMD is moving away from dual core Rev F desktop chips with 2x1MB cache and will focus on the ones with 2x512KB. This is indeed a great move, as some readers have pointed out. AMD64 architecture is so efficient, it doesn't need big caches. In fact, even the next gen K8L has only 512KB L2 per core. With inefficient archirectures such as Intel's aging Pentium III based design, up to 4MB L2 is needed for achieving reasonable performance.

The 2x512KB Rev F still has a die size of 183mm^2, which is quite large. Using a geometry of 13.5 x 14, I found that a 300mm wafer can produce 336 dies, a 200mm wafer produces 144 dies. Each 1000wspm @ 300mm corresponds to 1 million dual core dies per quarter. FAB30 is 200mm at 30000wspm, it can produce 13 million dual core dies. FAB36 should be at 7000wspm by mid 3Q06. So AMD should be able to pump out 20 million dual cores per quarter by then. By 1Q07, FAB36 will be 50% converted to 65nm. At 65nm, the die size will be 95mm^2, which means a 300mm wafer can produce 690 dies. One 1000wspm at 300mm should produce 2 million dies per quarter. Thus, by 1Q07, AMD's capacity should more than double to produce close to 25 million dual cores per quarter. AMD will have capacity to supply both DELL and HP 100%. By July 2007, K8L time, FAB36 will be 100% 65nm, that FAB alone can crank out 30 million dual core dies per quarter. By 4Q08, AMD will be doing 45000wspm 300mm wafers.

BTW, I forgot Chartered FAB7 is delivering 1000wspm of 300mm wafers in July 06, 1 mil there. Based on demand, Chartered can crank a lot more chips for AMD.

Overall, AMD is poised to increase capacity to 100% of world's PC demand. I expect AMD to quickly cross 50% market share. Equally quick will be Intel's financial collapse, due to its higher cost and overhead.

Price war preview

At, a few days ago, Intel took #4 and #5 of the "Top 5 Sellers" in the CPU category. AMD made massive cuts on single core Athlon 64 Socket 939 and AM2. Now., AMD takes the #1, #2, #3 and #5 spot.

Top 5 Sellers

* AMD Athlon 64 3200+ 2000MHz HT Socket 939 Processor - Retail $99.99
* AMD Athlon 64 X2 3800+ 2000MHz HT Socket 939 Dual Core Processor - Retail $297.00
* AMD Athlon 64 3500+ 2000MHz HT Socket 939 Processor - Retail $114.99
* Intel Pentium D 930 800MHz FSB LGA 775 Dual Core Processor - Retail $190.99
* AMD Athlon 64 3000+ 2000MHz HT Socket 939 Processor - Retail $96.00

Expect AMD to take the #4 spot once it drops dual core prices.

Tuesday, June 13, 2006

The AMD64 Rev G Mystery

Looking at AMD's presentation in its June 1, 2006 analyst meeting, there is an interesting slide about Rev G production . Pay attention to the wafer picture on the right (the words say "High-performance, power efficient 65nm AMD64 processors on production 300mm wafer"). Those were definitely Rev G dies. One guy has written this analysis on the Rev G. Basically, the conjecture was that the extra circuits include one additional complex decoder, an OOO Load/Store buffer and OOO read/Write buffer. However, there is also this large block below the L2 cache, which is quite a mystery.

AMD spent a lot of time talking about Rev H (K8L) during the anaylst meeting, but it gave virtually no information on the capabilities of the coming 65nm Rev G. A side by side comparison of the original Opteron and Rev G is here.

I think AMD may have delayed the Rev F Opteron to pull in Rev G.

If AMD pulls out a Rev G on July 23, 2006 to frag Conroe, don't be surprised.

However, if AMD does it, you can write off Intel from history in four quarters instead of seven. Intel will continue to exist for a while, like DEC did. Then it will be gone with the wind.

Monday, June 12, 2006

On DELL and AMD again

INQ reported that DELL will start selling more AMD machines. I have analysed the situation previously in great detail at here, here, here and here. The logical conclusion now is that DELL will have to go AMD across the board (desktop, mobile and server) on about July 23, 2006.

Basically, Intel forced DELL to go AMD. DELL must go AMD or it will BK.

Intel's price slashing has two devestating effects on DELL

1) DELL's pricing advantage is reduced or eliminated as everyone now gets Intel CPUs dirt cheap. This vanishing pricing advantage means lower units and unit share for DELL.

2) Of course, PC unit price will also drop.

The above two factors coupled with AMD's increasing market share and slowing down of the PC market growth will lead to DELL's revenue decline. Continuing revenue decline will lead to the collapse of DELL, as its stock is valued at 17x price/book.

The only way for DELL to maintain revenue growth is thus by entering AMD market in full force. That's a 50 million unit/year market. DELL can expect to grab 15% of it to compensate the loss of revenue in the Intel space.

AMD to chainsaw Intel

This is huge. Athlon 64 X2 3800+ to be sold at $169. That chip can beat Pentium 965 XE. Such cuts will definitely deny Intel the oppurtunity to unload its legacy inventory. I bet Hector is calling Chartered to load more wafers.

One may wonder what will happen to AMD's revenue and profits. You have to keep in mind that AMD's ASP is about $100, which means most of AMD's units are Semprons. Now with the price cut, it may increase the units on Athlon 64 and X2. In other words, the mix changes. So the net effect on ASP is less. Coupled with fast expanding capacity, I expect AMD to achieve sustained revenue growth and unit share gain.

By 4Q06, AMD's capacity will double from 1Q06 level. I expect AMD to exit 2006 with 40% market share.

The first victim of this AMD-INTEL death struggle will be Intel. The second will be DELL. With Intel's price crash, Dell will have close to zero advantage over competitors. Dell's revenue will fall because of lower unit price. To survive, DELL will have to go AMD across the board to maintain revenue growth by increasing its market share.

Intel will simply walk out 2006 dead.

Conroe getting really close to be busted

One month from Conroe launch, we finally have one independent 32 bit benchmark comparison between Conroe XE 6800 2.93GHZ 4MB and Athlon FX62 2.8GHZ 2x1MB. Remember I told you that Conroe will only do better in single threaded loads where the working set is comparable to 4MB cache?

First look at ScienceMark results,the Con E6800 is 9.2% faster than FX62 in Primordia, but they got almost identical scores in Molecular and Cypher (less than 1.5% difference).

Next, in PiFast and SuperPi, E6800 is 36.6% faster in 10M PiFast, and 88% faster in 1M SuperPI. Both PI codes have very little branching and memory usage has high locality.

In Cenema 4D, FX62 and Con E6800 scores were close, 11 and 10 seconds, E6800 is about 10% faster, but the time measurement is not precise enough to determine exact numbers. This is clearly a multithreaded test, as the single core CPUs have about half the performance.

In Kribibench, Con E6800 is over 100% faster than FX62. Seems to be another Super SuperPI case to me.

In Cinebenh 2003 CPU Render, Con E6800 is 47.9% faster than FX62. However, we have seen other 32 bit Cinebench tests, a 3GHZ Woodcrest was 22% faster than a 2.6GHZ Opteron. The version of Cinebench differs, but the discrepency between these two results is too big. A question mark here.

Doom3, Con E6800 39% faster. Doom is single threaded.

Quake 4, Con E6800 9% faster.

FEAR, Con E6800 5.8% faster.

One thing you should notice is that the above gaming tests were all done at low resolution of 640x480. At higher resolutions and high quality, the FX62 beats Con E6700 by 7% in this Intel arranged Farcry test. Higher resolution requires more memory, thus reduces the effectiveness of cache.

The Con E6800 is 2.93GHZ. The E6800 has 4.6% clockspeed advantage over FX62.

Judging from the results, the Con E6800 offers minor speedups in modern games. Except a few simplistic tests, the E6800's lead over FX62 is around 10% in majority of the tests.

What's missing from the picture is those SysMark tests and multi-tasking tests Intel is hiding from us. I expect FX62 to bust Con E6800 in such tests. I like to see some tests with quicker memory (lower latency).

Another missing item is 64 bit performance. Windows Vista is near and we expect AMD64 to achieve 10-40% performance gains running in 64 bit. Mooly Eden's claim of a major performance gap is closer to be busted. Those who bought Intel's hype will be disillusioned.

The FX62 is on sale now at $1390. Intel must be mad seeing that price tag,as its Pentium D chips are selling at 10% of that price. Intel has to wait 40 days to sell Con E6800 at $999. I think by then, AMD will have a few 65nm parts off its test runs just to push Con E6800 a bit.

4x4 will definitely set Intel back by 50%. The 4x4 tag will add a 20% premium to AMD chips, just like SLI cards sell at higher price.

Then comes AMD's next core, with 50% integer performance increase and 300% FP performance increase per core.

AMD has slashed prices on its single core CPUs. An Athlon 64 3500+ AM2 is now only $120 at I am glad that AMD is executing Operation Kill. AMD is going to deny Intel the oppurtunity to unload its legacy inventory. I expect AMD's ASP to fall by 15% in 3Q06 to $84 due to massive price cuts on desktop chips. However, both FA36 and Chartered FAB7 are cranking, AMD's capacity should increase by at least 40% from the 1Q06 level. The Turion 64 X2 and Opterons should continue to give AMD some nice profits.

Sunday, June 11, 2006

Joe Osha paints grim picture for Intel's 2Q06

Joe Osha wrote in his analysis of Intel :

We are lowering our Q2 revenue estimate to $7.8 billion, below the company’s forecast range of $8.0-8.6 billion, and our earnings estimate goes from $0.14 to $0.10.

In 4Q05, Intel had revenue of $10.2 billion; in 1Q06, $8.9 billion. Now we are looking at $7.8 billion for 2Q06. But, Joe Osha missed something important here. As I pointed out, Intel may see GAAP loss in 2Q06 due to impairment of goodwill and other charges. From 3Q06 onward, Intel should see operating losses.

Back in Feburary 2006, I projected 15-20% fall of Intel's revenue in 1Q06. This has come true. At that time, I also projected even larger revenue drop in 2Q06. In March, I projected operating losses for Intel from 3Q06 onward.

Looking at Free Cash Flow (FCF), Intel and AMD were almost at the same level as of 1Q06.

I reaffirm my projection that Intel may suffer losses in the next seven quarters. If AMD executes the kill strategy I proposed, the total collapse of Intel can be expected by mid 2008.

Joe Osha re-iterated his "buy" rating on Intel and "sell" rating on AMD, while his firm is busy dumping and buying. Joe's bonuses may get reset to much higher this year, I guess.

Saturday, June 10, 2006

Conroe is a stopgap solution, AMD 4x4 is a permanent solution

A "4x4" sticker is just like a SLI sticker: it's a differentiation that devalues the competition. -- Sharikou

Some people say that AMD 4x4 is a stopgap solution. Such a view is naively wrong.

From 2002 to 2005, PC performance gained very little. The Pentium 4 reached 3GHZ in 2002, at the end of 2004, it was about 3.6GHZ. There was only 20% improvement in two years. AMD's Athlon 64 did outperform the P4, but the margin was not huge, around 10-30%. Then in 2005, AMD introduced dual core technology, system performance almost doubled overnight.

Today, personal computing is in a new era. CPU core improvement is still important for incremental gains, but from now on, performance jumps will come primarily from increasing the number of computing engines. Multiple cores is the answer for the future.

You never get enough performance. On the graphics front, we have moved from dual graphics cards to quad graphics cards. On storage, we have multiple hard drives forming RAID to increase speed and reliability. On memory, we have dual channels to increase bandwidth. The theme is clear, more is faster. But strangely, so far, we have only one CPU inside a typical PC.

To fully utilize the computing power of multiple cores across multiple CPUs, you must have an inter-core, inter-processor, core-memory and core-I/O communication platform to deliver the compute cycles to the outside world. This is distributed computing. AMD's Direct Connect Architecture, with its crossbars, cache coherent hyper transport, integrated memory controller and hypertransport, is the most elegant communications platform exists today.

AMD 4x4 utilizes the glueless ccNUMA capability of the AMD64 on the desktop. AMD 4x4 doubles the compute engines on desktop. With two dual channel memory controllers and double HyperTransport links, 4x4 doubles memroy bandwidth to 25.6GB/s and I/O bandwidth to 16GB/s. 4x4 also allows one to have 8 slots for non-registered DDR2. 4x4 almost doubles system performance with the efficiency of the Direct Connect Architecture. As both AMD and Intel are expected to be closely matched in multi-core CPU development, 4x4 is a long term solution to pin Intel at 50% of AMD's performance on the desktop. As we move to quadcore or octal core, the same 2x multiplier on the 4x4 platform will set Intel back 50% yet again and again.

The concept of 4x4 isn't limited to two CPU sockets. One can have 4 graphics cards, why not four CPU sockets if needed? Also, one of the socket may be used for Torrenza cards.

Now, what about Conroe? Conroe is merely an improvement on the execution engine. It sits on the same communication platform: the shared FSB. Back in 2003, Intel was talking about 10.2GHZ Pentium 4 by 2005. According to Intel, Conroe will be a 40% improvement over Pentium XE. Intel is clearly behind schedule on their performance goal. Intel's FSB based approach is not scalable and doesn't represent a challenge to AMD64. Eventually, Intel will have to follow AMD and create a communications infrastructure to deliver the compute cycles. Just like what Nvidia's SLI did to ATI, AMD 4x4 will force Intel to react. A 4x4 sticker is just like a SLI sticker: it devalues the competition.

AMD 4x4 will deliver an equivalent of 18 GHZ Pentium 4 cycles with an efficiency factor of 90%. It will more than double that with the K8L quadcore. Now, that's compelling.

In 2Q07, the K8L will be out. AMD is expected to widen its lead at single core level. Coupled with 4x4, that will be 3x peformance lead over Intel. Conroe will have to be scrapped and redone. Its shelf life is no longer than that of a banana. AMD 4x4 will continue for the forseeable future, at least two to three years.

Now tell me, which is a stopgap measure? 4x4 or Conroe?

Intel's 60% price slashing reflects true value

I have analysed the consequence of Intel's pricing collapse. It is unfortunate that Intel will see red. But Intel's chips indeed only worth about 40% of AMD's chips. Here is why.

1) Total cost of ownership: Considering the power consumption difference and the electricity costs, running an AMD CPU saves at least $150 over three years, assuming energy prices remain stable.

2) Performance and longevity: Athlon 64 is faster than Pentium under 32 bit Windows, that's the situation now. Going forward, with Windows Vista approaching quickly, Intel's P4 with EM64T will see 10-20% performance drop under 64 bit, and AMD64 will see 10-40% performance increase under 64 bit. As a result, current EM64T owners will get a big downgrade and current AMD64 users will get a free upgrade. Under 64 bit, an Athlon X2 3800+ will run like a X2 4400+, and a Pentium 965 XE will run like a Pentium D 805. A Pentium D 805 will run like, well, a Celeron D.

If you look at AMD's AM2 chips, they have even higher potential value

1) They support VT. You can run Windows XP and Windows Vista the same time.
2) They support 4x4. You can't get that from anywhere else.

Friday, June 09, 2006

Must watch video: Dual core stress test

Intel vs AMD dual core stress test video by

Thoughts, Intel overclockers?

You may also want to read this on heat and motherboard life.

PS: This video is super amazing.

Intel Tulsa/Truland Xeon MP profile

Intel announced today that it will pull Tulsa's 4Q06 launch into 3Q06. Tulsa, Intel's codename for:

  • 435mm2 die size
  • 1.328B transistors
  • Dual Core, 4 threads
  • 2 x 1MB L2, 16MB shared L3 cache
  • 1.25V / 150W TDP
  • 667/800MHz 3-load FSB

  • Imagine a DELL PowerEdge 6850 with 4 such suckers. 600 watt typical heat from the processors alone. Total 8 cores, 16 threads, sharing 667/800 MHZ bus. Each core gets an average of less than 100MHZ FSB bandwidth, each thread gets less than 50MHZ bus bandwidth, or 0.4GB/s, less than that of a 80486.

    You interested in this big piece of silicon? It's dirt cheap.

    But be prepared to pay 600 * 24 /1000 *365 * 0.18 = $946 a year for electricity. No, double that for cooling cost. If Google used Tulsa, it would need 30 mega watt of power.

    Thursday, June 08, 2006

    Intel may suffer losses in the next seven quarters

    Intel is cutting its own throat. It's confirmed that Intel will slash P4 and Core Duo prices by up to 61% on July 23, 2006. The stated aim is to gain back 3% market share. However, according to analyst report, Intel had already cut prices by up to 40% since April, however, that price cut failed to stimulate demand. On the contrary, Intel's volume in April dropped 21%.

    In my opinion, INTEL made the drastic cuts because it's deperately short of cash, and it needs to convert its legacy inventory into cash real soon.

    So far, AMD hasn't officially responded to Intel's cuts.

    As I analysed here, Intel will see operating losses because of a price war. 3Q06 and 4Q06 will be very tough for Intel. After that, AMD will unleash its K8L super frag.

    AMD is an all new beast. In the x86 space, who commands the high ground has the say on the market. AMD rules the high end server market and has 100% of the 64 dual core mobile market. It can easily deflate Intel's attack at the low end desktop. AMD 4x4 will command the desktop high end also. You don't have a full line of AMD products? Sorry, our Opterons and 4x4s are in short supply. Combined with the momentum of Torrenza, AMD is poised to unify the whole computing industry under one architecture - AMD64.

    Intel's recent behaviour is a puzzling. It made everyone know AMD64 is better. Now it announces deep price cuts 45 days ahead, which poses a dilemma even for die hard Intel fanboys who wish to buy a Netbust today. Such early announcement might be an attempt to halt AMD. But, the world still needs PCes, and most people want to buy the best.

    Honestly, I think Intel management has gone nuts. They were never bright, but what they are doing now is suicide. AMD has lower cost structure and near 60% gross margin, Intel's grosss margin is projected to be 49% for 2Q06. With fragile financials, Intel's long term viability is in serious question. I will not be surprised if Intel files Chapter 11 in two years.

    It's time for AMD to make some aggressive moves.

    Pentium D 830 consumes 166 watts

    A Chinese site measured power consumption of 14 CPUs to determine which one is most suitable for frying eggs for breakfast. The results showed Pentium D 830 as the champion spewing 166.14 watts of thermal energy with both cores fully loaded. Note the measurement was for CPU power only, it didn't include power consumption of other parts, such as chipsets.

    AMD quadcore seems to be coming real soon

    AMD's Henri Richard stated at Computex that 4x4 will soon get an upgrade.

    "This platform, which is second to none, will soon be (upgraded to) '8x4,'" Richard said. "It will have eight CPUs very, very soon."

    I don't think this will be 4 sockets. It must be quadcore. During AMD's analyst meeting, AMD's FAB Guy,
    Daryl Ostrander said "Rev G was production ready". Also the recent INQ report that AMD is delaying Socket F opteron makes me wonder if Rev G is pulled ahead instead. It makes sense. Even if yields are low at 65nm, AMD should have no problem producing enough Opterons.

    Intel confirmed its massive 60% price cut on July 23, 2006. Previously, I projected Intel to suffer losses in the next seven quarters.

    Wednesday, June 07, 2006

    A true retard at ZDNet

    I run into this blog entry at ZDNet: 4GHz Dual Core Game PC at $825! It is the Pentium D 805 overclocked to 4.1GHZ, which has performance comparable to Athlon 64 X2 4600+ in a lot of tests.

    But the author fails to consider the electricity bill when performance/watt is the buzzword today. The sucker consumes 474 watts and needs $300 of electricty per year.

    Busting AnandTech's Woodcrest vs Opteron benchmark

    Anand reproduced MySQL bug #15815, and quickly termed it a feature called "negative scaling"

    We have proven that Anand is a paid Intel pumper here. Now, AnandTech has done a series of Woodcrest vs Opteron benchmarks. Again, AnandTech showed Intel having substantial lead.

    Really? Can a paid pumper be trusted ever again? Let's forget about the past, and look at AnandTech's newest presentations without prejudice.

    One year ago, in June 2005, AandTech's Johan De Gelas did similar benchmarks on dual core Opteron. You can see from that page that Opteron MySQL performance was 17% lower under Gentoo Linux than under SuSe Linux. On the same page, AnandTech compared the MySQL (InnoDB) performance of two (2) dual core Opteron 875 processors against one (1) dual core Opteron processor under Gentoo Linux. The conclusion was that adding a processor led to about 10% total performance increase. This was extremely poor scaling, but there was at least some increase.

    AnandTech's system configuration on Woodcrest and Opteron is here. Pay attention to the motherboard AnandTech chose for the Opteron. It was the K8N Master2-FAR. What's special about this board? It is probably the only 2P Opteron board with only one bank of memory. With this board, the second Opteron always has to get its memory from the first CPU, incurring high latency. This MSI board (pictured below) is definitely not designed for server use, it doesn't even have onboard video.

    A typical Opteron 2P server board looks likt this Tyan S2892, with two banks of memory, one for each CPU. The only AMD recommened MSI Opteron board is the K8D Master3-FA4R.

    For both Woodcrest and Opteron, Gentoo Linux 2.6.15-gentoo-r7 and Mysql5.0.21 were used. Gentoo has about 1% of the Linux market and is getting more popular. Now, look at the MySQL performance numbers AnandTech got here.

    First, we look at the comparison between one(1) 3GHZ Woodcrest Xeon 5160 and one (1) 2.4GHZ Opteron 280. The Woodcrest did 996 queries/second, while the single Opteron did 805 queries/second. The Woodcrest had 25% clockspeed advantage and 23% lead over the Opteron 280 on MySQL query speed.

    Now, look at the next row, which was TWO Woodcrest CPUs against TWO Opteron 280 CPUs. The result? The 2P Opteron 280 system showed a 23% performance DECREASE from 1P Opteron 280. It did 622 queries/second. Adding a CPU in a 2P system leads to 23% drop in total MySQL performance? Two processors slower than one?

    Anyone with half a brain would immediately conclude that something was seriously wrong with this benchmark. The result was simply stupid. If some moron told his boss that adding a CPU actually causes performance drop, he should be fired immediately. Because he should have used his brain to fixed the sucker instead of making such stupid conclusions. You need to dig deeper, you need to debug the software and see where it's hanging, and then solve it. MySQL has a lot of parameters to tune, so does the Linux SMP kernel. You either try out all available combinations to find a rational outcome or you use your brain to identify and fix the problem. Of course, doing this requires deeper understanding of the operating system and MySQL. We don't expect AnandTech's people possess this level of knowledge, otherwise they'd be working as software or system engineers who make a lot more money and accumulate a lot of stock options. But, they should at least contact the OS and DB vendor and Google the WEB* to find a solution. If the problem was identified as a major bug in the OS or DB or was otherwise persistent, they should try some other OS such as RedHat Enterprise, SuSe Enterprise, or different versions of OS and DB. A nonsense software result simply can't be used to represent CPU performance. A nonsense result only makes the benchmarker look stupid.

    Yet, AnandTech concluded that "[q]uad-core and Dual-core x86, you'll notice that the scaling is negative... It seems like an anomaly, but this is not the case. These benchmarks have been checked, verified and checked again. "

    So, not only AnandTech is a paid pumper, they are also surprisingly incompetent*.

    Stupidity lends no confidence.

    Actually, if you look at the first graph on this page, you can see that the 2P Opteron 280 quickly reached a saturation point at concurrency level of 2. But the 1P Opteron 280 scaled up quite nicely up to concurrency level 8. Apparently, there was a resoucre contention problem in the 2P setting. You just need to dig deeper...

    Performing a Google search with the words "Linux opteron dual core MySQL bug", I quickly found this's page for bug #15815. The #15815 bug is highly reproducible, it was a bug due to mutex locks and cache line "ping-pong". There was a fix for it. In the discussion of the "negative scaling" on AnandTech, "mutex ping-pong" was mentioned. This is indication that someone at AnandTech actually knew about this. At the bottom of the bug #15815 page, a user reported:

    "I'm using MySQL 5.0.21 on a 2.6.16 Gentoo system, I've tried both Gentoo compiled and precompiled Mysql distribution, and the Xeon machine (dual xeon 3.8Ghz with 8GB RAM) acts too much better then the dual Opteron (Opteron 275 with 16GB RAM)...better as 640 tps (Xeon) versus 110 (!!!) tps for the Opteron."

    Note this guy was using the Gentoo OS and the same MySQL 5.0.21* Anand was using. He was also running dual Opteron 275.

    Maybe someone noticed this bug report on and instructed Anand to use the same setup to reproduce it but present it as an Opteron benchmark?

    The most current version of MySQL is 5.0.22.

    Tuesday, June 06, 2006

    SUN seems to be planning a Blitzkrieg

    According to this Register report, SUN is planning some Opteron blades with "vastly higher I/O bandwidth, thermal capability and RAS features than anything else on the market today."

    A SUN insider claimed that "[t]his will let us go through multiple processor generations from quad to octal core*, and multiple memory and I/O generations without requiring changes in chassis, power supplies and other components." Hmm, John Fowler said these things during the AMD analyst meeting. SUN is going to bring system longevity to x86.

    Exciting times ahead.

    *Indicates socket F (1207)

    Our Coverage of Anand

    On June 5, I published the SysMark 2004 comparison between Con E6300 and Athlon 64 X2 3800+. Many visitors from read that page. On June 6, Anand showed SysMark 2004 comparison between Conroe XE and FX62.

    Some people keep quoting Anand's stuff as gospel truth. I have proven that Anand is a paid Intel pumper. So please stop quoting Anand as reliable source here, except for critical examination.
    • Dempsey vs Opteron benchmark: Why did AnandTech handicap the Opteron? The responses I quoted in the comments were from Anand himself. We know now, Dempsey is no match to Opteron in 64 bit performance.
    • Anand's benchmark on Yonah: X2 3800+ won by 16:6, yet Anand initially gave X2 a negative conclusion. After seeing those big Centrino ads on the same review page, it became all too clear. Anand's primary source of income is ad money.
    • IDF: Was Anand duped by INTEL? Anand pushed the Intel arranged buttons, instantly wrote that "Intel Regains the Performance Crown". After the BIOS issue, Intel called Anand back and had the machine BIOS flashed and tests redone. No one else got such VIP treatment from Intel. I suggested to AMD that it should subpoena Anand for information regarding the IDF test as part of the discovery process in the anti-trust lawsuit.
    • One month later, Intel quoted Anand's words and showed them to Wall Street analysts and investors. Intel market cap increased about $100 million that day. You notice that was also quoted by Intel (see page 51 of this Intel presentation to Wall Street).
    • Today, Anand published what he claimed to be an independent Conroe benchmark, while others could only push Intel arranged buttons. How could Intel trust Anand so much more than others? Anand claimed that he gathered the Intel parts. How? The Taiwanese were suddenly not afraid of Intel any more and just gave Anand the Conroe CPU?

    Let's make some quick comparisons. In this test by PCStats, an Athlon FX62 got a SysMark 2004 Office Productivity Overall score of 261 . Another independent result for a slower Athlon 64 X2 5000+ (2.6GHZ,2x512KB) agreed with PCStats' results quite well. There, the X2 5000+ with a 7200RPM HD got 230 points in SysMark 2004 Office Productivity Overall. In Anand Lal Shimpi's test, FX62 (2.8GHZ, 2x1MB) only got 210, Conroe XE (2.93GHZ) got 266. Look at the FX62 sub scores from PCStats, they were 263 for communications, 297 for document creation, and 214 for data analysis. However, in AnandTech's results for FX62, the sub scores were respectively 178, 280, 185.

    Let's look at result on Business Winstone 2004, the FX62 got a score of 36.4 there. However, at AnandTech, FX62 only got 27.9, while the Conroe XE got 32.8.

    Why were AnandTech's scores on FX62 substantially lower than the scores obtained by others?

    Anand is not dumb. He knew that AMD64's main advantage is low memory latency due to the integrated memory controller (IMC). AMD64 doesn't need huge cache in general because it can access memory quickly. AMD estimated that IMC's low latency gave its CPUs 20% performance edge. Intel Conroe's solution is to use large cache to compensate the lack of IMC. With this knowledge, Anand decided to use high latency 5-5-5-12 DDR2 memory for his test. As a result, FX62's low latency IMC advantage was almost eliminated.

    As you can see from this memory shopping page, most DDR2-800 memory in the market today has 4-4-4-12 or lower latency. In fact, on, out of 59 DDR2-800 memory products, only 15 models have CAS latency of 5, the other 44 products have CAS latency of 4 or lower. 4-4-4-12 memory is 25% quicker than the 5-5-5-12 used by Anand.

    AnandTech's results on FX62 should therefore be considered invalid if not fraudulent. Based on AnandTech's SysMark 2004 results on Conroe XE (2.93 GHZ) and PCStats' results on Athlon 64 FX 62:

    SysMark 2004 Office Overall: Conroe XE scored 266, Athlon 64 FX 62 scored 261
    Business Winstone 2004: Conroe XE scored 32.8, Athlon 64 FX62 scored 36.4

    This agrees with our previous findings.

    Charlie at INQ commented that "[i]t would most likely be cheaper to buy all the hardware sites out there off". I guess some company has already done that. What did Charlie know? A hardware site, a script kiddie finished reading "How to Upgrade Your PC" pushing benchmark buttons. I doubt his annual income is big. He should be cheap, as our Charlie observed.

    Intel desktop chipset consumes 23 watts typical

    INQ reported that the Conroe chipset, the P965 consumes 23.1 watt typical. The other day, Intel reported that they have an ultra low power CPU, the immediate question came to mind was "how much the chipset consumes?".

    I bet that the Blackford chipset for the Bensley platform will consume close to 40 watts.

    It's tough business for Intel for run these chipsets at 1GHZ on junk yard 130nm process.

    Monday, June 05, 2006

    Conroe close to be really busted

    Remember I told you that Conroe is only good for those single threaded applications with working sets comparable to 4MB, so they can run from cache? Now, there is more proof. A Chinese site got hold of one of this Con E6300 chip (1.86 GHZ) and benchmarked it with BAPCo SysMark 2004.

    BAPCo SysMark was created by Intel. In July 2002, AMD was allowed to join in its development. Today, BAPCo's members include INTEL, AMD, Microsoft, HP, DELL, CNet, ZDNet, Nvidia, ATI and others. SysMark is about doing a series of real activities, mimicking real computer usage. These tests have large memory requirement, as they chew on large documents, etc. Also, multiple applications may be running at the same time during a SysMark test. Intel had avoided using this benchmark in all the Intel controlled benchmark runs.

    In SysMark 2004 test, the Con E6300 got 210 points. In comparison , an Athlon 64 X2 3800+ got 232 points (2/1.86 = 1.07, 232/210=1.1). In Business Winstone 2004, the Con E6300 got a score of 25.2, an Athlon 64 X2 3800+ got 26.2. In Content Creation Winstone 2004, the E6300 got 32.2, the Athlon X2 3800+ got 33.2 (note that WinStone is not a multi-tasking load, it just runs through more memory). Recall that the X2 3800+ has only 512KB L2 cache per core. Suddenly, the Conroe doesn't look so rosy at all. Intel priced this chip lower than X2 3800+ for a good reason.

    No wonder Intel so far doesn't allow reviewers do their own benchmarks. The most recent comparison of Con E6700 and FX62 was again done with Intel configured hardware and benchmark suites. When a lot of amateurs and script kiddies cheered Intel's IDF guerilla benchmarketing, I smelled something fishy from the very beginning. Later I pin pointed the origin of Conroe's scores.

    Why is this so important? Intel's Israeli team has bragged about the 4 issue core and various minor hacks. The amateurs cheered about a 33% IPC improvement. They say 4/3 = 133%. But anyone with a good understanding of CPUs know this is impossible. The Conroe has 14 stages, 4 issue, this means 56 instructions in flight at any moment, the amount of dependency is inherent in normal code. AMD has long concluded that 3 issue is efficient and cost effective.

    So, again, just as we have seen from Woodcrest benchmarks, from SysMark results, we see CORE2's IPC is not any higher than K8. CORE2 marchitecture is no miracle. Clockspeed and bandwith will determine Conroe and K8's relative performance.

    I expect that once Conroe is released in the wild, it will be soundly fragged by AMD's offerings in most multi-tasking loads. Remember, the whole purpose of going multi-core is for multitasking/multithreading.

    Of course, the AMD 4x4 technology will permanently pin Intel at half of AMD's performance.

    PS: I am glad that the Chinese have started running benchmarks. What da heck. It's just installing and running programs. Any kid can do it.

    PPS: Let's enjoy some music.

    PPS: FX62 benchmark results can be found here.

    DELL customer Rackspace went HP for AMD64

    Intel is set to launch Woodcrest on June 19th, 2006. I saw some vendors have Woodcrest servers ready for June 23, 2006.

    But, Rackspace, a long term DELL customer, who has 20,000 servers and is adding 800 servers a month, had to go HP for Opteron today. On May 18, DELL announced plans for MP Opteron (4 processors and more) by the end of 2006, but DELL did not announce plans for 2P Opteron servers. Rackspace couldn't wait any longer, it went HP for the 2P Opteron HP DL385.

    Didn't Rackspace read the wonderful Woodcrest stories? Didn't DELL show Rackspace those wonderful Woodcrest stories? Rackspace can't wait just a couple of weeks?

    Or, maybe Rackspace actually got a Woodcrest in their labs?

    "We've really been a sole-source buy [with Dell] for years, and going to another vendor is a big change from a technology standpoint," Lew Moorman, SVP of Rackspace said. "It was something we needed to do."

    Sunday, June 04, 2006

    AMD Torrenza a huge wave

    I discussed the implication of openning ccHT before at here and here. Now, read this news about Tarai's ASIC content processors. This is huge, folks. ASICs have much better performance, and should cost a lot lower than FPGA.

    Tarai's content processors are much more "general purpose": XML, Perl-Compatible RegEx, ZIP/GZIP, UTF8/UTF16, RSA, DES/3DES, Media encoding/decoding...

    Imagine PERL boosted by this chip...This is a revolution!

    Intel runs into thermal wall again with Conroe XE aiming at 130 watts

    Intel demoed overclocked Conroe XE at 3.5GHZ with liquid cooling to journalists. David Tuhy at Intel said that the Extreme Edition would be able to run at up to 130 watts.

    No matter how much Intel can overclock Conroe, it can't match the performance of 4x4. You see, overclocking Conroe from 2.9GHZ to 3.2GHZ is only a 10% improvement, but power consumption increases by 30%. Power consumption rises non-linearly as one increases clockspeed. From 3.2GHZ to 3.5GHZ is just another 9% increase, but liquid cooling is required. Intel is stuck with the old school of thought. The way they try to increase the number of compute cycles is again through jacking up frequency. Soon Intel runs into yet another thermal wall, deja vu all over again.

    Intel never learns.

    AMD's thinking has been different since 1999 when they designed the Opteron. AMD uses more compute engines to increase performance . With AMD64, you double the performance by doubling compute engines, thus multi-core and multi-CPU designs. The key to the success of this is to make sure when you double the engines, you have a delivery system that allows the compute cycles to be fully utilized, instead of being wasted in FSB collisions. That's the Direct Connect Architecture. IMC, ccHT, XBAR and HT technologies ensures near linear increase of computing power as one increases the number of compute engines.

    AMD's 4x4 technology gives you near 100% improvement at the blink of an eye. 4x4 permanently pins Intel at half the speed of AMD. At 130 watt limit, AMD can put together two 65 watt AM2 X2 4800+ processors in a 4x4 PC, which almost double the performance of a X2 4800+. The beauty of Direct Connect Architecture allows AMD to build such a system at very low cost. No additional chipset is needed for 4x4, you just connect the ccHT wires. I estimate a 4x4 motherboard will cost $150-$200. 4x4 will be faster than a 2P 4 core Opteron, because it uses unbuffered RAM.

    If Intel competes against 4x4 with a cheaper two socket shared 800MHZ FSB design, performance will be very poor, as each core will only get 200MHZ bandwidth. If Intel pulls Bensley/Blackford into desktop, the cost will be very high.

    3.73 GHZ Dempsey Xeon 5080 fragged by 1.8GHZ Opteron 265

    The much hyped Dempsey (Xeon 5080) with the Blackford chipset (the so called Intel Bensley platform) is finally available for independent tests. The result is embarassing for Intel. The 3.73GHZ Dual core Dempsey server got squarely beaten by 1.8GHZ Opteron 265 in Apache web server benchmarks. Opteron 265 is AMD's lowest dual core server chip. In less bandwidth demanding workstation application tests, the 3.73GHZ Dempsey did better.

    Clearly, Intel's imitation of Athlon MP 's dual bus is not going to solve Intel's FSB bottleneck problem. Please note that Woodcrest (Xeon 5100) share the same Bensley platform. No wonder Intel only allows controlled benchmarks of Woodcrest in its own labs.

    Saturday, June 03, 2006

    Intel may suffer losses in 2Q06, 3Q06, 4Q06, 1Q07, 2Q07, 3Q07

    For Intel, the common scene for the next 12 months will be impariment of goodwill (abandonment of acquired technologies and businesses), writing off inventory, writing off assets (old FABs, etc), one time charge for layoffs, price crash, unit share loss, etc, etc. Intel's stock holder equity will vaporize faster than that of Enron. GAAP loss is a certainty, but I also see operating loss from 3Q06 onward. I previously projected mass layoffs in 2007, but since Intel failed to execute a survival strategy, I projected a faster collapse in Feb 2006.

    2Q06: 40% price drop, 20% volume drop, impairment on goodwill, one time charges for massive 16,000 layoff. GAAP loss in the bag. Expect Intel to warn soon. Currently, Wall Street expects Intel to earn $0.15 per share. Expect ($0.05) loss at minimum.

    3Q06: Conroe ramps to 10%. 20% additional price drop of P4, mobile CPU price crash, 10% additional volume drop. Operating loss expected. Plus more impairment on goodwill, plus one time charge on inventory write off.

    4Q06: Conroe ramps to 20%. Merom ramps to 10%. Further reduction of P4 and Core Duo price. Revenue flat from 3Q06 in a seasonally up quarter. Operating loss. AMD exit 2006 with 40% market share (run rate).

    1Q07: Conroe ramps to 35% of desktop units. Revenue down more than seasonal as AMD's 65nm parts flood the market.

    2Q07: Intel server market share drops to 40% as AMD ramps Rev H quadcore. Bulldozer hits hard.

    3Q07: AMD FAB36 ramps to 20,000wspm at 65nm. FAB38 ramps. Chartered FAB7 ramps. AMD grabs >50% of market share.

    4Q07: Go figure.

    Friday, June 02, 2006

    Global Crown Capital Equity Research's David Wu needs to have some IQ

    According to INQ, David Wu at Global Crown Capital Equity Research (GCRE) wrote that

    We wereperplexed by AMD's performance-per-watt leadership claim as 1) AMD deliberately compared its Opteron to Paxville (soon-to-be 2 generations old) and avoided Intel's extremely competitive Woodcrest processor that will be launched in a few weeks; 2) AMD again decreased Opteron's total system power consumption by ignoring North Bridge components and increased Woodcrest's by doubling the number of memory slots (16 FB-DIMMs instead of a typical 8 FB-DIMMs for DP). According to trusted 3rd party reviews, Woodcrest DP system consumes roughly 20% less power and outperforms at least 20% at the same time.

    Such low IQ analyst is extremely annoying. First, David Wu said AMD avoided Woodcrest, then David Wu said AMD doubled Woodcrest memory slots. David Wu's brain coherence can't be sustained within one sentence. A stupid analyst is harmful even to the con-analyst profession.

    Marty Seyer's presentation slides are here. The webcast video is here. On slide 7, he compared the currently shipping processors, the Xeon Paxville and Opteron. Then, he said since Intel is hyping Woodcrest, he would take the oppurtunity to set the record straight. On slide 8, Seyer compared the upcoming chips, the Intel Woodcrest and AMD Santa Rosa Opteron (Socket 1207).

    Seyer's estimate of Woodcrest is based on the following

    1) The Blackford north bridge chipset in Woodcrest consumes a minium of 22 watts. In AMD system, the north bridge is inside the CPU. Both AMD and Intel have similar south bridge chips for I/O.

    2) Each 1GB FB-DIMM (for Woodcrest) consumes 11 watts. Each 1GB DDR2 DIMM (for Opteron) consumes 4.4 watts. Seyer used 16 GB memory for the 2P server for the estimate. If you look at typical Bensley systems, most of them have 16 FB-DIMM slots, they can hold up to 64GB. During the Q&A, Seyer was asked about the average memory size. Seyer said if you only fill half the slots (8), the Opteron would still consume 15% less power (Woodcrest: 144+22+ 8*11 = 254, Opteron: 190+ 8* 4.4 = 225, then add 28% for conversion loss). It obvious that for Woodcrest to consume less power, the system must have less than 4GB of memory.

    3) Seyer further pointed out AMD's TDP spec is for max power. Intel's spec is for typical power.

    4) David Wu talked about independent 3rd party tests on Woodcrest. So far, all Woodcrest tests are done in Intel's lab with Intel configured system under Intel's watch. We exposed Intel's fraud previously.

    I hope this set the record straight.

    Some analyst actions:

    Sumit Dhanda of B of A, rates INTC at $25, dump below $20, buying AMD like crazy.

    Joe Osha of Merrill Lynch, rates INTC at "buy", Joe Osha says "we are buyers". Merrill Lynch dumped large portion of its INTC holdings at huge loss. Joe Osha rates AMD at "sell", Merill Lynch buying AMD like crazy.

    AMD: time for the kill

    AMD's tech analyst meeting yesterday was huge. The most impressive one was AMD's rapidly expanding capacity and its quadcore roadmap with next gen core. AMD is planning to add another 5K wspm to FAB36 and convert FAB30 to 20K wspm at 300mm. The conversion of FAB30 is a brilliant move, as it cuts the time. The total 45000wspm capacity will be enough for 80% of the world's x86 processor consumption. Also, AMD stated that Chartered FAB7 will be a major supplier.

    In the near term, by the end of July, with FAB36 ramping and Chartered FAB7 producing one million units, AMD should have enough capacity for 30% of the CPU market. This represents a 50% increase of AMD's capacity over 1Q06. What will AMD do with this extra capacity?

    Sell them.

    As I previously pointed out, Intel will be most vulnerable from 3Q06 to 4Q06, during the transition period from Netburst to Conroe. Right now, Intel is stuck with $7 billion worth of legacy chips ($3.5 billion inventory, 49% gross margin), those are Netburst based chips that are hot and slow. Moreover, Intel has squandered most of its cash. Intel is financially very weak. It can't continue to feed 103K employees and will start mass layoffs as I predicted last year. Intel is hoping to convert this huge pile of legacy Pentium 4 chips into cash to fund the production of its Conroe CPUs. Due to Intel's manufacturing inefficiency, it won't be able to switch to Conroe production quickly. Its projected units for Conroe is 35% in 1Q07. Intel must dump its legacy inventory, or it will have to write it off later. Intel's price for the Pentium D 945 3.4GHZ dual core CPU is expected to be $163.

    If AMD can succeed in preventing Intel from clearing the legacy inventory at a good price (above $120), Intel will find itself unable to fund Conroe capacity expansion. To achieve this strategic goal, AMD should prepare to sacrifice margins on desktop CPUs. While maintaining prices on Opterons and Turion X2s, AMD should introduce a low priced dual core and single core processors to stop the Pentium and Celeron dumping. There should be a 2.2GHZ, dual core, 2x 256KB cache AMD64 chip, priced at $150. There should also be a 1.8GHZ, 128KB L2 Sempron at $50 to halt Intel's Celeron flood in the 3rd world. In 1Q06, AMD's ASP was about $95, with rapidly increasing capacity, AMD can lower the ASP while maintaining revenue growth for 3Q06 and 4Q06.

    By the end of 4Q06, AMD will have 50% of FAB36 converted to 65nm and 65nm parts shipping in volume. AMD quad-cores should be out by then too. AMD can then start a trade-in program for Sempron AM2 users to upgrade to 65nm dual core chips.

    Even with massive cuts on desktop CPUs, AMD can still profit nicely from Opteron, Turion 64 X2 and 4x4.

    AMD MUST make 4x4 available for the general market by dropping the requirement for FX CPUs. Why not give this awesome 4x4 power to the non-gamers? Personally, I like to have a 4x4 equiped with two energy efficient AM2 X2 3800+ at 35 watts each. Others may just start with one CPU, reserving the right to add another one. AMD 4x4 gives AMD64 systems much longer life. In this sense, 4x4 for desktop is a disruptive technology. Those CPUs with 4x4 stickers can sell with a 20% premium.

    AMD's added capacity comes online with almost perfect timing. As long as AMD can sell all of its capacity, Intel is bound to collapse due to is weak financials and huge overhead cost.

    AMD replaces Intel in UMPC project

    Intel got kicked out by Samsung and Microsoft.

    Thursday, June 01, 2006

    Turion 64 X2 notebooks in stock at

    The world's only dual core 64 bit technology, the Turion 64 X2, is available in the form of HP DV2000z and Compaq V3000Z. I wonder why there are people buying 32 bit notebooks today.

    Intel's Mooly Eden said two 32 bit cores were better than one 64 bit core. How about two 32 bit core against two 64 bit cores?

    BTW: look at this wall socket PC that can even run off power over ethernet.

    AMD does 4x4 desktop computing

    4x4 is a long term strategic solution to pin down Intel at half the performance

    Two dual core CPUs in desktop, 4x4 computing. This is also great for cheap servers, Google should definitely look into it. There will be two banks of memory, two CPUs connect via ccHT gluelessly as in a 2P Opteron server. The two processors may be sitting next to each other, one heatsink is needed... This reduces motherboard cost substantially.

    I hope it can mixed CPUs, say one dual core and one quad-core, or a dual-core and an FPGA co-processor. Or, you can have one 4GHZ single core CPU for single threaded games, and one dual core for multithreaded ones. I also wonder about the rumored inverse threading.

    4x4 is a long term strategic ccHT solution to pin down Intel permanently at 50% of AMD's performance. As Intel moves to quadcore, so does AMD, 4x4 will have 8 cores. AMD can keep this 2x advantage until Intel develops something similar to ccHT and IMC.

    AMD is so smart.

    I can hear Intel folks crying about AMD changing the rules, "unfair" they complain.

    Intel's thinking is outdated. While the whole computing industry is looking to increase throughput, Intel is trying to increase complexity and squeeze last bit of single thread performance. Intel already ran into a wall with GHZ, but they never learnt the lesson. The AMD 4x4 technology will give AMD 80% higher performance on 80% of the benchmarks.

    Do you want to buy an Intel Extreme processor for $999, or two 35 watt Athlon 64 X2 3800+ AM2s for $600? Do you want 100% performance improvment or alleged 20%? I think the choice is clear.

    AMD should introduce a "4x4 capable" label for its boxed AM2 CPUs, much like the "HDTV capable" sticker on TVs. When people see the sticker, they know the 4x4 potential of the processor. The rest are all low end.

    More on AMD's plans is here.