Friday, April 28, 2006

A classic pump&dump action

We suspected that Bank of America was doing pump&dump on INTC, now we got proof.

1) Pump and dump INTEL.
2) Downgrade and scoop AMD.

Intel should ask SEC for criminal securities fraud investigation. Such pump&dump may have cost Intel close to $1 billion extra for its stock repurchase program in 1Q06.

AMD should also ask SEC for criminal securities fraud investigation.

Intel Exec does guerilla benchmarketing

Tom Kilroy, Intel's VP of the Digital Enterprise Group, which just saw its revenue dropping by $1 billion year over year, presented to financial analysts some wonderful graphs showing a 3GHZ Woodcrest server outperforming a 2.4GHZ Opteron server by 37%. If true, this would indicate that Woodcrest has a 9.6% advantage over Opteron at the same clock (1.37 /3 * 2.4 -1).

The problem? He used an application called SunGard ACR for the benchmarketing, instead of the industry standard stuff such as the SPEC marks. Was SunGard a server daemon? Or, was it a SuperPi type, where you don't exercise much I/O bandwidth? Mystified, I tried to google "SunGard ACR" to get some information, but the thing is quite elusive. Google doesn't give much hint on what the heck SunGard ACR is. In fact, when I googled "SunGard ACR", the first entry popped up was an Intel commissioned benchmark PDF* showing an Intel configured Xeon/Bensley server with 10K RPM disk faster than an Intel configured Opteron with a 7200RPM disk and a little known motherboard. Intel deserves some serious laugh.

It seems that Tom Kilroy is also a master in "guerilla benchmarketing". Fire a shot and seek cover, hit and run.

Previously, when another Intel VP accidentally used a common benchmark tool, the Clovertown (double Conroe) got fragged by an Athlon 64 at the same clock.

* Interestingly, the PDF was removed after we posted this article, a cached version is here. No, now even the cached version is gone.

IBM claims HP and SUN have Opteron advantage

In this ZDNet interview, Bill Zeitler, senior vice president of IBM's systems and technology group, claimed that Opteron systems bring in more revenue per unit. He said HP's gains were not due to high units, but higher unit revenue from Opteron servers, IBM's own Opteron blades also bring in much higher unit revenue than Xeon blades.

"If you look at HP's results over the last few wasn't their unit growth that was causing them to improve, it was their average unit revenue, and the average unit was improving because they had more AMD content than they had had previously."

"In our own case, the average unit revenues on an AMD blade are much higher than the same kind of Intel blade because the performance is better. And because the performance is better, people put more I/O (input-output components) and more memory and other things on them."

Since AMD has 22.1% of server unit share but 27% of the revenue share, an Opteron server brings in 30% more revenue than a Xeon server.

It makes sense. If you have a Xeon, adding more I/O cards and memory will choke its front side bus.

Thursday, April 27, 2006

Analysts lash out at Intel CEO

Paul Otellini, Intel CEO, claims that Intel lost market share not because of AMD's superiority, but because of Intel's chipset shortages.

Needham's Charlie Glavin doesn't buy this cheap excuse:

"Baloney. I was hearing from the guys who will be buying the chips for the (computer) boxes. They said on a performance basis that AMD's products are better".

Prudential's Mark Lupacis shares a similar view. In a research report, Lupacis dismisses most of Intel's stated goals as wishful thinking. Specifically, Lupacis considers Intel's expectation to grow revenue by 18% in 2H06 "unrealistic", and Intel's gross margin will be negatively impacted by additional costs and larger die sizes. Moreover, Lupacis has discredited Intel's notion that it lost market share only because of chipset shortage.

Lupacis points out that history can't be used as guide, he reckons:

"AMD is clearly a different animal than it was 5 or 10 years ago. The company is no longer limited by capacity and only focused on supplying the low-end of the desktop market. AMD's portfolio is broad, arguably superior in the server and desktop markets, and for the first time ever,competitive in the notebook area.

...For years, the MPU was the gating factor in PC performance, and Intel’s manufacturing muscle focused solely on the MPU led to its market dominance.Today, it is not just the MPU, but how efficiently it interacts with peripheral chips like DRAM. As a result, we think that the market has shifted from a manufacturing driven one to an innovation driven one.Considering that AMD was the first with 64-bit memory addressing in the desktop, the first with a true dual-core product, and the first with an integrated memory controller hub, it appears that AMD is beating Intel to the innovation punch. Intel’s massive manufacturing muscle, on which it leaned so heavily in the past, is not as big of a differentiator anymore.
By early‘07, we expect AMD will have a true quad core solution for both the desktop and server markets, which will look a lot better than Intel's solutions, two dual core chips glued together in a single package. In addition, we expect AMD's lower-cost Turion notebook MPU to continue to gain share for the next several quarters and anticipate a new mobile architecture from AMD in mid-07 that will test Intel's dominance at the high end of the notebook market."

Lupacis expects INTEL to continue losing market share to AMD, leading to a duopoly where AMD controls 30-40% market share. Lupacis rates INTC at underweight with a target price of $15, based on projected 2007 EPS of $0.96.

A more hapless Intel rushes unproven Woodcrest

We have yet to see Dempsey in the wild and Sossaman 32 in action. Now, Intel claims it will ship Woodcrest in June, just one month away. With AMD taking 27% of server market, Intel is reacting in a very frightened and disordered fashion.

Wednesday, April 26, 2006

DELL to pay the ultimate price

AMD now firmly commands the high end of the server and desktop. DELL, a pure Intel vendor, is thus pushed to the ultra low end. Dell is selling Pentium 4 PCes with 19inch LCD for $399, Celeron PC with 19inch LCD for $349, and Intel notebooks at $399. It will be very hard for DELL to make revenue and profit numbers doing such ultra low end stuff. In comparison, HP sells future proof PCes without monitors for $599.

*) Intel is slashing prices across the board. Now, anyone can get Intel dirt cheap. This neutralizes advantages DELL had over competitors in terms of special Intel pricing.

*) Intel's hyping of the Conroe will create larger Osborne effect on current Intel CPUs as we get closer to the Conroe launch date. This will affect all DELL PC/Server products.

*) Once Conroe is launched in Q3, the limited availability of the Conroe chip will stop DELL's sales cold as people wait for Intel to ramp up production. The projected Conroe volume is 10% in 3Q06, 20% in 4Q06 and 40% in 1Q07.

*) 68% of DELL's business comes from US corporate market. Intel's hyping of the vPro will hurt DELL's sales immediately as vPro stickers won't be available till 3Q06.

*) DELL's direct sale business model is in question. If it's so good, why is DELL's market share of US consumer market so small? (14% of Dell's 2005 revenue)

*) DELL is already losing unit share.

*) AMD has readied next generation chips to frag Conroe.

*) AMD, HP and NVIDIA are working together to penetrate the corporate client market.

*) DELL's balance sheet is terrible. DELL has only $4.1 billion net asset but $60 billion market cap. DELL's valuation is at 15x bookvalue. DELL can't endure three bad quarters.

*) The collapse of DELL will also lead to a quicker collapse of Intel.

Tuesday, April 25, 2006

Intel pushing obsolete technologies to US government

Intel has an online document titled "Government PC Buyer Request for Proposal Guidance". The document contains tables that "include statements that can be included directly in RFPs to clearly justify the addition of a feature as a requirement". Intel claimed that the statements are written in vendor-neutral language. Let's take some examples:

*) Mobile-optimized dual-core processor should have shared cache
*) Chipset should support 667 MHz FSB and should be integrated, and should also support known future operating systems

The two are clearly specific to Intel's outdated FSB architecture.

AMD should propose something more reasonable, such as

*) Mobile processors must support 64 bit Microsoft OSes
*) System should have more than 12.8GB/s of total memory and I/O bandwidth
*) Processor should have integrated memory controller
*) Desktop processor must consume less than 95 watts maximum

Some ideas on utilizing HT 3.0

HT 3.0 is very exciting stuff. HT 3.0 is hot pluggable and link length can be 1 meter long.

I envision in the near future, AMD PCes will come with a cache coherent HT 3.0 port, much like a i1394 port.

You can then buy specific acceleration chips to insert into the port. For instance, there could be a chip for DOOM, you buy the game, it can play fine. But there is also an acceleration chip which you can insert into the HT 3.0 port. The CPU recognizes the chip and delegates some computation, such as fragging a monster, to the dedicated hardware. Take another more realistic case: video encoding. Video encoding is very CPU intensive if done in software, if one can get a MPEG HT 3.0 chip that can be plugged into the PC that performs realtime MPEG encoding, it will be a huge boost. Starting from this idea, for each video format, one can have a specific HT 3.0 chip for it. Or look at PhotoShop, you can buy a chip that does the expensive image processing algorithms. The artists will love it. For your web server, you can get an XML parser chip... In principle, any piece of software can have a dedicated booster chip, as long as there is a market for it.

The beauty of the HT 3.0 cards is that you can remove them from one PC to another with ease. So you use them on your desktop or you can take your booster chips on the road and boost your Turion 64 notebook.

Now, suppose you have two AMD64 PCes both with HT 3.0 ports, you can then buy a HT 3.0 cable to link them together, and immediately you get a SMP boxes with double CPU power and double memory....

That should be a lot of good business for everyone.

Monday, April 24, 2006

Opteron grabs 22.1% of x86 server market

Citing statistics by Mercury Research, AMD reported that its Opteron processor has taken 22.1% of the global x86 server market in 1Q06. This is a 35% increase from AMD's 4Q05 server market share of 16.4%. In revenue terms, Opteron now has 27% of the server market.

Excluding those Dell PowerEdge server with Intel Celeron inside, Opteron's share might be even higher.

HT 3.0 Spec Released

AMD led HyperTransport Consortium just released the HT 3.0 standard. The new features include:

1) 2.6GHZ clock, up to 20.8GB/s (166.4 Gb/s) per link bandwidth
2) AC mode allows links up to 1 meter length for chassis to chassis connections.
3) Auto link splitting between 2x8 or 1x16 links
4) Hot plugging

The extended HT link length allows one to connect multiple individual Opteron servers into higher order SMP boxes. Fujitsu-Siemens have a 8 way Opteron server which is made of four tighly packed 2P Opteron blades connected via HT. Now, one can have Opteron servers farther away forming larger SMP machines. The additional latency due to 1 meter wire length is less than 4ns. A 32P SMP machine can be built by linking 8 four way 1U servers. Since you can hot plug the HT links, you can even add and remove servers (CPUs) on the fly. This is like a cluster, but you are building a true SMP box, running a single instance of a standard SMP OS within a single memory space. This will revolutionize server computing! SUN and IBM better hurry and do more Opteron.

HT 3.0 is 100% backward compatible with previous generations of HT, which means an old NF4 chipset can be hooked onto a newer Opteron with HT 3.0, even though it's less efficient.

I expect AMD's Rev F Opterons to utilize HT 3.0, and Broadcom to supply a lot of the server chipsets (HT bridges with a whole bunch of features integrated).

Currently, an Opteron CPU has three 1GHZ HT links, all of which can be coherent (Opteron 8xx). A coherent HT link can be used for I/O or IPC, but a non-coherent link can only be used for I/O. A next generation socket 1207 Opteron may have more HT links, coupled with the increase in bandwidth, the next generation Opteron servers will be super strong. Even assmuming a Rev F Opteron has three HT links and one DDR2 800MHZ memory controller, the total bandwidth for a 1P Opteron will be 3* 20.8 + 12.8 = 75.2 GB/s. The total bandwidth for a 2P opteron will be 4*20.8 + 2 * 12.8 = 108.8 GB/s.

In comparison, a 1P Woodcrest will have a total bandwidth of 10.2GB/s with a 1333MHZ FSB (64 bit), not enough to pump DDR2 800MHZ (128 bit).

SUN is forever young!

Jonathan Schwartz has ascended to the CEO post at Sun Microsystems.

SUN is young and kicking. The shift of strategic focus from pure R&D to growth is the right move at the right time. Sun has enough technologies to choke a horse, but it desperately needs to ship products that customers want. AMD's customer-centric innovation doctrine is a good recipe of success to borrow. SUN needs products, discipline and execution.

Grow, baby, grow!

Sunday, April 23, 2006

VIIV bashed by the Post

Intel ran out of ideas in bundling business.

The idea of mandating Intel or any other company's Ethernet hardware is a joke: Ethernet ports are commodity parts. Quick Resume doesn't actually put a Viiv computer in any sleep mode; it just turns off the display and speakers while barely affecting the rest of the machine.

Read more at the Washington Post. It was quite entertaining but may infuriate Intel fanboys.

Friday, April 21, 2006

AMD and NetLogic doing something interesting

As reported here, AMD and NetLogic will announce a design monday. I guess it will be a NetLogic co-processor for security stuff while Opteron chews network traffic data via 2 HTT links, total 16GB/s of bandwidth.

CISCO, JNPR and FDRY are customers of NetLogic.

Broadcom to be more AMD

From Broadcom's 1Q06 CC transcript:

"Our share with AMD is really going to take off when they roll out their next generation of servers... At that time we expect that our revenue we derive from AMD platforms will exceed our revenue we get from Intel platforms."

"we have a substantial set of design wins in the next generation array of platforms being introduced."

Since the launch of Opteron in 2003, AMD has made virtually no major changes to the design (dual core was designed from the beginning). However, AMD has tripled its design team during the period. We should expect AMD to show some big things very soon.

Thursday, April 20, 2006

DELL may pay the ultimate price

All AMD vendors are gaining on Dell. I projected an Enron time in 18 months about 3 months ago, the clock is ticking. Dell's balance sheet is so terrible, I wonder how Michael Dell can sleep at night. The key here is unit share loss for Dell. Even at the insultingly low prices (Pentium D 820 with 17inch LCD for $469 free shipping), DELL is losing unit share, indicating that people simply don't want Intel's slow and hot chips. Dell and Intel are killing each other.

I proposed that SGI (an Itanic victim) should go Opteron. Now it's the time.

In other news, the class action suit against Intel's illegal monpolistic abuses started in earnest. Anyone who indirectly purchased an Intel CPU since June 29, 2001 can join the suit. I find myself qualifies. I intend to ask for a portion of my money back from Intel.

AMD: Intel is desperate

AMD spokesperson, Damon Muzny (<--Click the link to view the full interview):

" Intel's complex and inefficient architecture with large "compensation caches" made 65nm necessary to bring die sizes back down on Pentium. And this is just one aspect of the big picture. So when you hear the competition crowing about starting a new, smaller design process sooner, understanding the big picture gives a more clear view of what is really going on.

Here's another good example. Intel has been trying to take the focus of their current product portfolio with Conroe demonstration stunts. So while I can't really comment about a competitor's product probably which won't be available for 6 or 9 months, I will say this much about their tactics: it's not like their typical "AMD who?" approach. Clearly we've forced our competitor to respond us. Every time they stress future products they are simply admitting their current products aren't competitive so customers shouldn't buy them. That's just an irrational and desperate move for them. Their claims of future architectural competitiveness largely assume AMD is standing still, and we're not. In fact, as they work to catch up to us, we'll move ahead with our own innovations. So the big picture is Intel in crisis-mode, reacting to AMD's leadership. It shouldn't come as a surprise.

Muzny also indicated that AMD will transition 50% of wafer starts at FAB36 to 65nm by early 2007.

Wednesday, April 19, 2006

Intel is an ailing animal

INTEL 1Q06 highlights: Revenue dropped 15% from 4Q05. Revenue will drop another 10% in 2Q06. Inventory will go up. Gross margin will be 49%, down 10%.

All Intel can hope now is that AMD will screw up. Looking at its 1Q06 report, it's very very sick.

In just one quarter: Server and desktop revenue dropped by $1 billion (from $4.929B to $3.892 B), inventory is piling up to the neck and is snow balling to even bigger in 2Q06; Cash is reduced to $7.8 billion, debt is at $2.6 billion; Current ratio dropped from 2.3 to 1.83 .

Compared to AMD, Intel is declining at extreme speed. AMD has cash of $2.6 billion, debt $0.6 billion. Current ratio 2.16 .

More alarmingly, Intel has 103,000 employees and AMD only has 10% of that.

Intel may try to bite, but its teeth are rotten and loose.

I feel sorry for Intel. All AMD needs to do now is unleashing its hounds.

Read AMD's comments about Intel's here.

Highlights of INTEL's 1Q06

GAAP Results (including the effects of share-based compensation)
Q1 2006 vs. Q1 2005 vs. Q4 2005
Revenue $8.9 billion -5% -12%
Operating Income $1.7 billion -44% -49%
Net Income $1.3 billion -38% -45%
EPS 23 cents -34% -43%


Digital Enterprise Group
Microprocessor revenue 3,892 4,929 4,944


*I noticed that Joe Osha didn't get a chance to ask question during Intel's CC, funny.

AMD revenue share of the x86 market reached 18%

Intel just reported its 1Q06 results, total revenue dropped 5% and EPS dropped 34% from 1Q05. Total revenue from microprocessors was $6.2 billion down from $7.329 billion . This means AMD has grown to 18% in revenue terms, from 15% in 4Q05.

If we look at microprocessor revenue alone. From 4Q05 to 1Q06, Intel's revenue from desktop and server dropped from $4.929 billion to $3.892 billion, Intel's revenue from mobile dropped from $2.4 billion to $2.35 billion. Revene from desktop/server dropped $1 billion in one quarter. Intel claimed that ASP was down only slightly.

For 2Q06, INTEL projected a further revenue drop to $8.0 - $8.6 billion, below seasonal patterns, gross margin will be reduced to 49%. With this kind of revenue and margins, 2Q06 EPS will be 10 cents.

In just one quarter, Intel's cash has been reduced from $11.3 billion to $7.8 billion. Intel has long term debt of $2.6 billion.

Intel's Inventory went up from $3.12 billion to $3.55 billion, this is almost a full quarter of inventory.

Intel entered a death trap

Back in Feburary, I emailed Intel execs reminding them of the "Operation crush" Andy Grove used against Motorola. Essentially, the trick was to pump vaporware like crazy and push competitor off balance.

But the time has changed. In the old days, the microprocessor was a new business, Intel was only selling a few chips a day. Back then, people were still deciding on which instruction set to go, promising a future was extremely important. Today, the world has already settled on AMD64 and have become much more sophisticated thanks to the Internet. Another major difference is Intel today has a huge existing product line to sell --Pentium-M, Pentium 4, Core Duo. Its 4Q05 projections for Conroe shipment is only 25%.

The result of Conroe pumping will be a quick death for Intel. The moment Conroe sells at $250 price, sales of other Pentium stuff will stop cold. Remember, CPU is just part of the computer cost. People would rather pay $150 extra to get a Conroe that can last longer, the electricity savings alone will pay for the extra $150.

Check how the celeron Xbox is selling (after Xbox 360 made a brief appearance), you will understand.

So, expect Intel's ASP to drop below $100, or 50% below current levels. Try figure out the gross margin yourself, and you will see red after deducting the $3 billion a quarter OpEx.

The same issue will plague the OEMs, especially Dell.

Tuesday, April 18, 2006

Mike calls Paul again

Mike: Paul, are you nuts? You pump Conroe like crazy, but only give me 10% what I need.

Paul: Sorry, Mike, our FABs haven't finished the copy-exact process yet. We are building 3 more 65nm FABs, and one 45nm FAB with 450mm wafers. We are the leader in process tech...

Mike: Crap! what about my Xmas sales? How do I tell my customers? They all want Conroe, and I only got 10%.

Paul: No problemo. Just sell the Presler PCes at half the price, lots of bargain hunters out there. We are gonna fry AMD this time, heh heh heh. Their FX60 will drop like a rock, their fat margins will be vaporized. We got $8 billion cash, they got only $2.6b...

Mike: Half the price? I am already selling Pentium D 820 PCes with 17inch LCD for $469, half that will be $230. How am I supposed to make my revenue numbers?

Paul: well, you can try double your units...

Mike: Paul, are you kidding me? Let me be frank with you, Hector showed me his next chip, it's running at 3.8GHZ and about 60% faster....

Paul: ------- (sound of heavy object falling on the floor)

Mike: Are you allright? hello, hello, hello...

Monday, April 17, 2006

Mike calling Paul

Mike: Paul, our business is not going well, your chips are a tough sale. We are going to have a bad quarter..

Paul: Mike, sorry, our business is in bad shape too, inventory is piling up. Let's tough it out together this time, we did it before.

Mike: Paul, that's your problem. We and our share holders can't pay the price because your chips are crap... I am gonna call Hector...

Paul: No, please, don't leave us, we love you, love you so much, we always want you to be happy. You are our only hope. Please stay. Anything you want, name it.

Mike: 50% of your margins, anything less, bye.

Paul: how about 45%.

Mike: I said 50

Paul: You are going to kill our margins..

Mike: your margin is already finished...

Paul: Please....just a small favour, i beg you, the crooks want my head, if you leave, they are gonna kill me... It is not even my fault...

Mike: 49%

Paul: OK. Thank you so much. XOXOXO

Sunday, April 16, 2006

Joe Osha wants heads roll at Intel

Doug Freedman said similar things before, now Joe Osha isn't convinced that Paul Ottellini is the right man for the job. But again, I think it's way too late to change horse. Intel is still brilliant in sales, the IDF guerilla benchmarketing was a masterpiece. Intel's mistakes dated back in 2000 when it went Netbust. What Intel has is an engineering problem. There is no chance that Mooly Eden's team can match AMD's Grand Masters. The gap is too big. Right now, the Israeli regiment owns Intel. The guys in the silicon valley must feel useless now. Bob Colwell, the P6 architect, is probably the only one who has the technical authority to command the troops.

Thursday, April 13, 2006

Clovertown scores revealed

Clovertown (2GHZ, 65nm, 2x4MB L2, FB-DIMM) slower than Athlon 64 (2GHZ, socket 939, 512KB L2, DDR)

Intel showed off Clovertown quad-core server CPUs running on the Bensley platform with FB-DIMM memory at Spring IDF Taipei. Clovertown is basically two 65nm Conroe CPUs stacked together, with total of 8MB L2 cache. This page contained the benchmark scores for a 2P Clovertown. The reported clockspeed was 2GHZ (could be 2.13GHZ, same as Conroe E6400).

For single threaded test, the 2GHZ Clovertown got a Cinebench 9.5* score of 362. Daniel J. Casaletto, Intel Vice President, Digital Enterprise Group Director, Microprocessor Architecture and Planning, was running the demo. For 2P 8 cores, the score scaled to 1723, or 4.7x. Adding 7 cores led to 3.7x more performance. I think this is quite poor, you get only about half a core's worth when you add a core.

FSB (Front Side Bus) is an Intel bottleneck. Eight Conroe cores fighting for a 1066MHZ FSB is not a pretty sight: in 2P Clovertown each Conroe core gets only 133MHZ bandwidth or 1GB/s, not much better than a 80486 (Well, you may say there is the dual bus, so it should be 266MHZ, but 8 cores of cache coherence traffic must be considered). Fortunately, Cinebench doesn't put a lot of load on the bus, it spends most of the time digesting the data.

Let's pay more attention to this photo here, which shows the 2P Clovertown in action and is quite exciting. Look at the upper left corner, it reads Cinebench 64 Bit Edition. Finally, we can see Intel got 64 bit working, it's running the 64 bit version of Cinebench! My congratulations to the Israeli engineers for getting AMD64 figured out. Welcome to the exciting world of pervasive 64bit computing! Now, Microsoft will allow the world+dog to go 64 bit.

A reader kindly provided us the CINBENCH 64bit Edition result for a 2GHZ socket 939 setup. The spec: Athlon 64, Socket 939, 90nm, 2GHZ, 512KB L2. The score: 370. <--click to view screen capture.

On my old Athlon 64 2800+ (1.8GHZ, Socket 754, 130nm, 0.5 MB L2 cache), I got a 64 bit Cinebench 9.5 score of 294. My ClawHammer is a bit slower than Conroe CORE, but only a little. If you consider my CPU is only 1.8GHZ and only uses single channel DDR, and my old PC only has integrated S3 UniChrome graphics which eats some memory, it's quite good. I managed to overclock it to 1.9GHZ and got a score of 312. I expect the three year old ClawHammer to get a score 0f 294*2/1.8= 327 at 2GHZ, within 10% of the future Conroe. So I feel I made a good investment buying AMD* - I paid less than $100 for the CPU+MB.

Clock for clock, the performance of Intel CORE (Merom/Conroe) is very close to socket 939 Athlon 64 . Mooly Eden definitely over-exaggerated Conroe performance.

So far, Intel is trying to follow AMD's footsteps. Conroe is still a few months away, and AMD is a moving target. Dirk Meyer said AMD will soon release a higher capability product besides changing from DDR to DDR2. With the new process technologies jointly developed with IBM, AMD can boost clockspeed by an upward of 40%.

The Conroe performance analysis is here. I pointed out that when working set is larger than Conroe's unified cache (4MB), Conroe performs slower than Athlon64. Conroe only shines in simple and single threaded tests where the whole working set sits inside the 4MB unified cache. The Cinebench 9.5 needs over 150MB to run. Clovertown's 8MB cache has some positive benefit, as Cinebench works on the scene top down, only a slice of the scene is being worked on at any moment of time . But still, the effect of the 2x4MB cache is much smaller than the cases where the working set sits inside the cache all the time.

In other news, AMD subpoenas Microsoft in its anti-trust lawsuit against Intel. From the documents it's seeking, AMD is clearly suspecting that Intel hindered the development of Windows for AMD64.

* Cinebench is essentially a CPU/Memroy performance test, it can be downloaded from .

* I bought the CPU+MB (with S3 IGP) combo for $79 at Frys, nowadays, such deals are impossible to find.

*I am interested in seeing some Clovertown and Sempron socket 939 comparisons. If you have such a machine running Windows x64, please submit your results in the comments. Don't under estimate AMD desktop CPUs, check out this Athlon 64 and Xeon comparison.

There are several people keep posting 32 bit benchmarks for Athlon 64. Please note Intel was doing a 64 bit Cinebench 9.5 . That's why Intel got a high score of 362 at 2GHZ. As I can see from posts by users on the internet, a Conroe at 2.4GHZ gets about the same score. So, please read the benmark condition: 64 bit edition of Cinebench 9.5.

AMD has a lot to learn from Intel

AMD is scheduling a tech analyst meeting in June. What is the point? Call up a bunch of financial analysts to look at some chips, then what? The world+dog is still wondering if AMD has something -- people need to run their benchmarks and talk about numbers.

Intel put two boxes at IDF and the whole world cheered. Henri Richard must admit that that was brilliant guerilla benchmarketing.

I projected another round of GHZ war when Intel announced going Pentium-III last year. AMD should prepare for that.

Intel is peddling its Pentium D805 dual core for $120, AMD needs to react to that with something like a dual core Sempron.

AMD management fails to think big

Frankly, AMD's goal of achieving 30% market share by 2008 is lame.

It's not even self-consistent.

AMD is ramping up FAB36 to 17000wspm, it also needs Chartered FAB7, which means FAB30 is also running at full speed. With this kind of capacity, it should be able to supply 55% of the market, unless its yield is only 25%, or nobody wants the AMD64 CPUs.

Even INQ is doubting AMD's purpose of building another FAB. If you are only planning to get 30% of the market, then those new FABs are a waste. The world only needs about 250 million x86 CPUs and is not going to switch to 64 core, 3600 mm^2 die size ones in 2008.

For both 4Q05 and 1Q06, AMD reported shortages on packaging material. This is clear indication of management's failure to forecast demand. It's a chicken and egg situation, unless AMD tells customers that it can supply 1000 truck loads, customers will think AMD is too small to provide that.

Dell may be a screw driver company, but it at least has an ambition to be number one. Kevin Rollins can talk about $100 billion revenue and how Dell is going to rule the PC market all day long , even though Dell has only 17% market share today. That's forward thinking, that's vision. If Dell sets a goal of $100 billion, and it achieves $75, that's very good progress, nobody will complain. Even when Dell set a $100 billion goal but only achieved $60, people still like that positive attitude. Even Google is talking about $100 billion revenue. If you set a small goal and achieve it, nobody cares.

Imagine you are fighting a battle, and your general tells you that the top goal is only to defeat a small piece of your enemy and survival is more important.

AMD's tiny goal fails to inspire.

Also, AMD issued a guidance that was inconsistent with rest of the story. It stated that it's ramping production day and night, increasing ASP, lowering inventory, and market is strong, yet it guided flat to down for 2Q06 with an extra week. That doesn't compute.

The Search Engine War

AMD + against INTEL +

Wednesday, April 12, 2006

AMD CEO confident on technology leadership

During AMD's 1Q06 earnings conference call, CEO Hector Ruiz was asked how AMD can continue to gain share as Intel becomes more competitive in 2H06, Dr. Ruiz responded:

"[W]e expected, we had planned that our competitor would eventually have to follow and react to what we have done and get better. It will be interesting to see the things that we're going to do later, which will again continue to force them(Intel) to react and figure out what else to do next. We don't intend to in any shape or form give any leeway in our leadership relative to product and technology."

On a similar question, Dirk Meyer, former lead architect of DEC Alpha, AMD Athlon, AMD Opteron, now AMD's President & COO, stated that AMD will introduce "high capability" products later this year.

AMD plans to reveal more of its technology roadmap on June 1, 2006.

AMD reported a 1Q06 EPS of $0.38, beating concensus estimate ($0.29) by 30%. This is an 80% sequential increase from the GAAP EPS of $0.21 for 4Q05.

More importantly, AMD improved it balance sheet dramatically from 4Q05 in just one quarter: Cash went up to $2.63 billion from $1.8 billion, debt was reduced to $0.616 billion down from $1.33 billion. Stock holder equity increased to $4.71 billion, up from $3.35 billion. In comparison, DELL has a lesser stock holder equity of $4.2 billion, Broadcom has a stock holder equity of $3.1 billion. AMD has a current ratio of 2.1, DELL's current ratio is 1.1 .

French government found a way to hack any Xeon

Loïc Duflot, a computer security specialist for the French government's Secretary General for National Defense information technology laboratory, claimed that he found a way to take remote control of any Intel Pentium/Xeon running any operating system. The problem might be related to the bugs (Errata) in the EM64T implementation. has an article on 3GHZ Opteron Spec scores.

Saturday, April 08, 2006

Conroe performance claim being busted

FUD is like ghost movies, you don't get scared by seeing a ghost, you get scared by not seeing one -- Sharikou

Recall Intel's Mooly Eden said Con-roe will be 20% faster than AMD's future chips without even knowing AMD's plans? During the Spring 2006 IDF, Intel setup a Conroe and an Athlon 64 box, then directed benchmarkers such as Anand to push buttons*, but peaking into Windows device manager of the alleged Conroe wasn't allowed.

During the IDF, I emailed various Intel execs, AMD execs and Anand, I pointed out that such a pre-arranged blackbox Intel setup against AMD was unfair and challenged Intel to lend the Conroe box to Anand for a real drill. However, Intel dared not to answer such a simple challenge based on the rules of fair competition. The INQ sharply criticised this kind of guerilla benchmarketing.

In fact, Anand had no way to verify Intel's IDF Conroe setup, the Conroe configuration parameters were provided by Intel. Anand noted that "it looked like Intel had done the unimaginable" with regard to the situation. Nonetheless, Anand assured readers that "there was nothing fishy going on with the benchmarks or the install" based on his trust on Intel's honesty -- which was seriously lacking from past records. Thus we had an interesting situation: Anand relied on Intel's reputation to validate the Conroe setup while Intel relied on Anand's reputation to validate the Conroe scores -- a loop of trust was formed to convince the world + dog.

Now, for the very first time, someone actually got hold of a Conroe chip in their own lab and did some tests. It was a 2.4GHZ Conroe (Link: CPU-Z) against an Athlon 64 overclocked to 2.8GHZ. The overclocked Athlon 64 had a 2.8/2.4 -1 = 16.7% clockspeed advantage.

The following results were obtained by running 32 bit ScienceMark binaries optimized for Intel Pentium:

Molecular Dynamics
A64: 1872.68
Conroe : 2133.38 -- 14% faster

Primordia (Energy calculations for 1 atom)
Athlon64: 1506.83 -- 10% faster
Conroe: 1365.85

Athlon64: 1345.05 -- 26.3% faster
Conroe: 1065.59

Athlon64: 1512.55 -- 21.7% faster
Conroe: 1242.94

The above results were for an Athlon overclocked to 2.8GHZ and a Conroe at 2.4GHZ, with the Athlon having a 16.7% clockspeed advantage. For a direct comparision at the same clockspeed, we normalize the Conroe scores by taking into account the frequency difference. Assuming the best scenario in which Conroe scores scale linearly with clock speed, we multiply the Conroe scores by a factor of 2.8/2.4. Thus, with a 2.8GHZ Conroe, we would have

Molecular Dynamics
Athlon 64 2.8GHZ: 1872.68
Conroe 2.8GHZ : 2133.38 * 2.8/2.4 = 2489 -- 32.9% faster

Primordia (Atom)
Athlon64 2.8GHZ: 1506.83
Conroe 2.8GHZ: 1365.85 * 2.8/2.4 = 1593.49 -- 5.7% faster

Athlon64 2.8GHZ: 1345.05 -- 8.2% faster
Conroe 2.8GHZ: 1065.59 * 2.8/2.4

Athlon64 2.8GHZ: 1512.55 -- 4.3% faster
Conroe 2.8GHZ: 1242.94 * 2.8/2.4 = 1450

ScienceMark is a strictly CPU/memory test, it doesn't involve video or disk I/O, it is basically a raw speed test. The ScienceMark is freely available from for both Windows XP and Windows XP x64.

However, the above results showed a violent CPU performance fluctuation for Conroe, from it being 32% faster to being 8% slower. How can this be explained?

The cause of the Conroe performance fluctuations can't be the types of computation involved. We notice that MolDyn is a floating point computation while the Cipher is an integer computation. However, both MolDyn and Primordia are floating point calaculations on quantum mechanical properties of matter, yet, Conroe's Primodia performance is only 5.7% faster than Athlon 64, a 27% relative performance drop from MolDyn.

As we look deeper in the ScienceMark, we notice that in the default MolDyn benchmark setting, there are only 4 cells with a simple cubic lattice, no more than 32 molecules are involved. The program is basically tracking the momenta and positions of a handful of molecules and computing scattering effects. About 2MB to 4MB memory is needed. The Primodia calculation for a single Ag (silver) atom with 47 electrons needs just a bit more memory than MolDyn. However, both the Cipher and STREAM tests involve a lot more than 4MB.

The reason why Conroe did so well in the MolDyn test is simple: Conroe has a huge 4MB of unified cache, for such single threaded tests that can fit in 4MB*, Conroe can just run off the cache with very high speed. Since cache misses drastically reduce peformance, applications run off cache exhibit unrealistic performance numbers.

However, once you go over the 4MB limit, Conroe is slower than Athlon 64 at the same clock. Both the Cryptography and STREM tests use a lot more than 4MB, larger than Conroe's 4MB cache, and Conroe immediately falls below Athlon 64 on the performance curve.

I can bet on this: if one increases the number of cells in the MolDyn test to 9, thus increases the working set to larger than 4MB, Conroe will perform worse than Athlon 64 at the same clockspeed.

There is another set of results on Conroe and Athlon 64, showing Athlon 64 beating Conroe on WinRAR file compression at the same frequency.

Most games are also cache sensitive, increasing Athlon 64's cache by 512KB, you see up to 8% performance increase in FPS.

I have added a comparison between Clovertown(double Conroe) and Athlon 64 2800+.

The conclusion is: clock for clock, Athlon 64 will beat Conroe in real application environments that require a working set of larger than 4MB, or in other words, larger than Conroe's 4MB cache. This means in any real multi-tasking or server environment the Core architecture will be an underdog. Even worse, for Intel's shared cache architecture, cache thrashing is a distinct possibility under heavy loads.

Most modern applications need a lot more then 4MB. IE needs at least 50MB when viewing a normal web page(with Flash, JS, DHTML, AJAX..); Photo Editing apps need around 40MB; FireFox takes 23MB when I use it to view; DivX grabs 23MB even before I open a video...

Frankly, I am really disappointed by Intel's decisions. This gimmick of using 4MB cache to get unreasonably good scores on the most simplistic tests is cheap from design point of view but expensive for manufacturing. Mooly Eden kept talking about the 4 Meg cache in the technology analyst meeting, and promised to add even more cache, however, the 4MB cache is definitely eating a lot of die area and Intel's limited capacity. It is almost like using Netburst's ridiculous hyperpipeline to pump up GHZ at the expense of power consumption and real performance. I wouldn't accuse Intel of benchmark fraud, but people need to know the 4MB limitation of the Conroe.

So far, Athlon 64 is being tested under 32 bit mode with executables optimized for the Pentium. Athlon 64 gets 10-40% performance improvement running in 64 bit mode, a benchmark under Windows x64 or Windows Vista should show the real strength of AMD64 architecture.

As a test drive, I downloaded the 64 bit version of ScienceMark and ran it on my Athlon 64 2800+(Socket 754, 130nm, 512K L2, at 1.799GHZ stock frequency, with 1GB PC3200 DDR) under Windows XP x64. For the 64 bit MolDyn test, I got a score of 1479.12 ScienceMarks, almost 50% faster than the 32 bit result on the same old PC. I suspect that on a Socket 939 Rev E6 platform with SSE3 support, the 64 bit result will be even better. A reader submit the 64 bit result for a 2GHZ Athlon 64, you can view the result here.

AMD should work with benchmark creators to ensure that application benchmarks have a working set larger than the cache size of Conroe -- 4MB.

AMD's Rev F socket AM2 will be available for system builders on May 15, 2006. At 65nm, using Stress Memorization Technology co-developed with IBM, AMD will be able to increase clockspeed to 4GHZ. AMD is also working on Z-RAM, a SOI based technology that may increase cache density by 500%.

*For those who question this authenticity of this Conroe benchmark, the person who posted the result had shown at least some CPU-Z screen captures indicating the various properties of the Conroe CPU. Anand wasn't even allowed to look at the Windows device manager, all he did was pushing some buttons as directed by Intel IDF staff. All the system specs of the Conroe system was provided by Intel. Anand had no verification of the setup. Also, unlike Anand, who receives a lot of ad money from Intel, this person who posted the Conroe results had nothing to gain financially either way. Clearly, this test has more credibility than Anand's. Anand's failure to mention that he was merely a button pusher and his obvious pumping style made his credibility very much in doubt.

*Intel touted its 1 cycle SSE execution, but the STREAM results weren't impressive. Henri Richard mentioned Conroe is more like K8.

*To verify this, you can download ScienceMark, then run the MolDyn, Primordia, Cipher and STREAM benchmarks on your own PC. You will find that the default MolDyn test uses very little memroy, Primodia uses a bit more, but Cipher and STREAM use a lot more than 4MB. To check this, you launch the ScienceMark program, then launch the dialog box for running MolDyn benchmark, at this point, the simulation hasn's started, two threads are created for this task, using a process viewer program, you note the memory used for the task so far is about 7MB. Then you click at the Run Simulation button, you will notice that another thread is created to run the simulation, now the memory used by whole task is smaller than 11MB for most of the time, meaning the benchmark thread uses less than 4MB and thus can fit in the 4MB cache of a Conroe.

Friday, April 07, 2006

AMD must have the aspiration for #1

You can judge a people's intelligence from their writings alone, that was how people concluded that Richard Nixon had an IQ of 155, Bill Clinton had an IQ of 182, without doing an IQ test.

Even though Joe Osha downgraded AMD to sell at some point, I said repeatedly that he is one of the smartest. I can see that just from his writings, the ability to learn and the ability to correlate discrete pieces of information into a logical and coherent conclusion are essential measures of intelligence.

So, it's no surprise that Joe Osha has come to similar conclusions as mine: 2Q06 will spell the start of ice age for Intel.

I previously projected operating losses for Intel from 3Q06 onward, I had revised my projections to include the possibility of a 2Q06 operating loss. I talked about inventory build up here, about NGMA effects here, about AMD's shift to higher end (increase of ASP) here, about Intel's capacity here, about AMD's five generation lead over Intel here, about Intel's Bensley paltform(with Dempsey, Woodcrest) here... The list can continue, basically, Joe Osha now agrees to most of them.

But still, Joe Osha thinks Intel has a lead in mobile. However, if he reads this about the 53 watt Core32 Duo and the coming Turion 64 X2 as well as the delay of Merom, he will change his mind, soon if not now.

Joe Osha summed up the situation for Intel:

Over the longer term, it’s fair to ask whether Intel will ever return to the dominance it once enjoyed... The company’s failure to acknowledge and react to clear cues from the marketplace opened up a competitive opportunity for AMD in servers that it will take years, if ever, to reverse. Even now, our checks suggest a disconcerting lack of urgency at Intel given the problems that the company faces.

In my simple and humble words:

Intel doesn't know it's dead because it doesn't have a brain.

My concern is: AMD's CFO has failed to correctly project demand twice, once in 4Q05, once in 1Q06. Twice, AMD benefited its competitor by handing CPU sales over to Intel.

Can AMD leadership wake up from an underdog mentality and strive to become the #1 global x86 vendor, by the end of 2006 if possible?

If AMD aspires to being #1, the time to plan for it is NOW. Intel is most vulnerable till 4Q06. It's time for the kill.

AMD's statement that it will only double 2005 capacity by 2008 saddens me.

PS: to keep this article at the top, I modified its date to April 19 -- the Intel capitulation day

Another baggie pumps INTC

Bear Stearns research analyst Gurinder Kalra pumps INTEL like crazy. Doing our regular crook check at, not surprisingly, we found Bear Stearns foolishly increased its INTC holdings by 271.56% in the December 31, 2005 reporting period, da baggie probably bought more later, before INTC's crash on Jan 18, 2006.

When the numbers are out, I am sure some retards will get fired for making their bosses into long term under water INTC stock holders - or baggies.

1Q06 notebook sale drops by 35-40%

Digitimes wrote:

Taiwan channel distributors estimate that notebook sales in the first quarter are likely to decline by 35-40% on-year, amid sluggish demand. These players estimate that sales in the first three quarters should drop by 20-30% in the first three quarters of this year.

I told you, Mooly Eden. Only idiots and fanatics will pay premium for 53 watt, 32 bit, expensive Core32 Duo, when Turion 64 X2 is on the way. And, one 64 bit Turion 64 core is better than two 32 bit Core32 Duo cores.

Intel had revenue of $3.01 billion from mobile in 1Q05, a 35% haircut is $1 billion reduction in revenue. This alone would bring Intel's 1Q06 revenue to $9.434 - $1 = $8.434.

My whipser number for Intel's 1Q06 revenue is $7.67 to $8.13.

Deutsche Bank AG to lose $1 billion on INTC

AMD will report 1Q06 results in a few days. Ben Lynch @ Deutsche Bank (DB) downgraded AMD and upgraded INTC again. Doing our regular crook check at, we found DB increased its INTC holdings slightly to 71.262 million shares. DB probably bought even more before INTC's 20% crash after January 18, 2006.

We thought Ben Lynch was smart, but it turned out he is a retard after all. It now seems that DB was selling INTC in September 2005, then Ben Lynch got them to buy more at $26 afterwards. I can imagine Ben's boss shouting at him day and night.

My advice to Ben Lynch: dump INTC at a loss and buy AMD, you can still triple your money.

Wednesday, April 05, 2006

Opteron enters telecom - carrier grade SUN blades

"Today Sun strengthens its leadership in the telecom market -- Sun's largest vertical market -- where Netra servers are used by all of the top 10 network equipment providers worldwide".

Why Opteron? Why? Why? Why?

64 bit, big iron reliability, performane/watt, dedicated memory channels, 3 hyperTransport links for total of 24GB/s bandwidth.

Intel's FSB has a total of 6.4GB/s, not enough bandwidth to handle DDR2 memory, no room left for massive amount of IP traffic.

The SUN shines.

Tuesday, April 04, 2006

Core Duo MacBook Pro Literally Hot

I told you it's a 53 watt 32 bit chip. Mac fans are getting burnt and Steve Jobs now knows it. One Mac user measures the temps of the case and reads 121F. Amazingly, some diehard Mac defenders say being hot is a good thing. Other Mac fans claim that a notebook is not a laptop, so the MacBook is not intended to be put on their lap. Wait for Turion 64 X2, which has max thermal power of only 25 watts and can independently power down each of the cores.

AMD to increase performance by 40%

According to Hans Deppe, vice president and general manager, AMD Saxony LLC & Co. KG, 65nm process at FAB36 is expected to provide a 40 percent improvement in transistor performance. Actually, AMD and IBM already announced this on Dec 6, 2005. Transistor performance is measured by switching speed. It seems that 4GHZ Opterons are not that far away. In December 2004, AMD and IBM announced 24% transistor performance increase with DSLSSOI, and Opteron speed went from 2.4GHZ to 2.8GHZ (3GHZ soon).

AMD has ramped FAB30 to 30,000 wspm, and is ramping FAB36 to 17,000wspm by the end of 2006, reaching 20,000wspm by mid 2007. Chartered FAB7 is on schedule to ship by mid 2006. It seems my previous estimate was a bit too conservative.

Unisys server catches fire -- Intel Inside

Today at the LinuxWorld Expo, a Unisys server was literallty on fire. Smoke triggered fire alarm and firemen rushed in, a Unisys employee giving talk at an Intel booth was disoriented. Benchmarks show 4P Opteron trounces 16P Unisys Xeon. Unisys got lots of watts/performance.