Friday, December 30, 2005

INTEL's 65nm Pentium 4 Presler disapponting

AnandTech did a set of benchamrks on INTEL's Pentium EE 955 Presler processor build on a 65nm process. Presler is a shrunk version of the Pentium 4. The good news is, the dies size of the processor got reduced from 206mm^2 to 162 mm^2. But there are two pieces of bad news

1) The Pentium EE 955 performance continue to be sub standard. The Presler performs substantially slower than AMD's entry level Athlon 64 3800+ in quite a few benchmarks.

2) The heat production continues to be very high, which means INTEL's 65 nm process is no magic cure for its heat problems. We have seen similiar situation from the Yonah. At 2.1GHZ, the 65 nm Yonah produces about 50 watts.

INTEL can't even fool business week

I was utterly surprised by this business week article, it wrote, and allow me to quote:

"But here's a secret that few outside the chip world know. While Intel execs have said the Pentium M was "built from the ground up" to suit their needs, it actually is a heavily modified version of the Pentium III chip Intel jettisoned back in 2000 in favor of the Pentium 4. "How do you make the Pentium 4 better? Use the Pentium III," scoff execs at rival Advanced Micro Devices.

PLAYING CATCH-UP. Adds former Intel chief chip architect Bob Colwell, who helped design the Pentium Pro, the original basis of the Pentium III: "We wrote a list of 20 to 30 items we would do to improve the PIII, but just left that list lying around. To their credit, the guys in Israel picked it up. But this is no radically new chip."

It seems that INTEL can't hide any more, it's getting smoked out in the Courts and getting exposed by the media on technology. Intel is leaping ahead, but in the opposite direction. Pentium M is 1995 technology. Merom is 1995 technology with copied elements of AMD64.

Read more of this article, very telling indeed.

Thursday, December 29, 2005

Eric and Paul conversation

Paul: Eric, I heard about your proposal, frankly, I was disappointed. "Core Duel"? What does that mean? We just lost a dual core duel against AMD. Actually, we did not answer the duel, but AMD shamelessly crowned itself as the champ, even more ridiculously, the media applauded. The worst thing was that I couldn't say anything to refute the Frenchman. No need to remind people of that episode. Google it and everyone can see it.

Eric: Sorree sir, I meant "Core Duo", C-O-R-E D-U-O. Our dual die CPUs are like two singers in a duet, they share the bandwidth.

Paul: They do?

Eric: Yes. The two cores in our next gen CPU share a FSB, just like two singers share the air as a medium for sound waves.

Paul: That's good analogy indeed, some deep physics too, and "Core Duo" does have a ring to it.

Eric: Thank you, sir. I tried a lot of word combinations on an expensive grid computer and finally nailed it.

Paul: Did you use a Xeon cluster for that?

Eric: No, I didn't. I was afraid the cluster would cause a power outage in our headquaters. I tried some 3rd party service...

Paul: I hope you didn't spend a fortune on it.

Eric: Well, in fact, it's kind of cheap. SUN is doing this $1 per CPU hour computing service with a whole bunch of Opterons, I used 10,000 CPU hours, the SUN folks were bad at accounting and only charged me $100 bucks.

Paul: Ha ha. That's sweet, no wonder those guys are not profitable..."Core Duo", hmm, it's new and refreshing indeed.

Eric: Yes. I believe it will get people's attentions, again.

Paul: What if we do quad core later. You will call it "Core Quadro"?

Eric: Exactly, you are so good...

Paul: Of course, I have been doing marketing for years.

Eric: We also need to change the INTEL logo.

Paul: What? That logo was created by Dr. Moore himself and has been with us for almost 40 years.

Eric: Forgive me. You may not believe this, but in Korea we do. I hired a Fengshui master to check out the AMD logo and INTEL's. The master said it looks like the arrow on the AMD logo shoots down the "e" on INTEL-- "e" could mean "emotions", "earnings" or even "existence"...

Paul: Fengshui? That's superstition.

Eric: Actually, it has scienctific foundation, it's like the Butterfly effect, you see, everyone of us looks at the logo with the fallen "e", that creates a mental effect and ....

Paul: I see where you are going, are you sure about this stuff?

Eric: Absolutely.

Paul: OK. Fine. How do we escape the arrow?

Eric: Let me show you ....

Paul: Not bad, not bad at all, this oval is like an impregnable fortress... What about "Intel Inside", we spent billions on this, our lawyers threatened a lot of people who use the "inside" word, it's like we own this word in the English dictionary, and it worths 36 billion dollars -- which I hope I can put on our balance sheet... Why change it?

Eric: Same reason. The master said that "inside" renders a picture of confinement, and leads to contraction of our markets. We have to get rid of it. But if you insist on not changing it, it's ok, in any case, it is a 36 billion brand...

Paul: No, no, no. I am open minded. We need to try everything we can to change our situation--our current situation is bad... we are getting squeezed inside, by Opterons at the top and Semprons at the bottom, we need to break out.

Eric: I understand, there is a Chinese saying, "desperate man try desperate things". I suggest we change "Intel Inside" to "Leap Ahead".

Paul: Why not "leap forward"?

Eric: "Leap forward" is just going forward, we are always going forward, aren't we? Our Paxville is 50% better then previous Xeons, but that's not good enough. Also, "leap forward" reminds people of the "great leap forward" by Chairman Mao.

Paul: What was that?

Eric: As neighbours, we know Chinese history pretty well. Chairman Mao created this "leap forward" program to catch up with UK in 3 years and surpass US in 10 years.

Paul: I see, I recall watching about it on history channel.

Eric: Since we are behind, we need to leap ahead.

Paul: I see, ahead, ahead of ...

Eric: AMD...

Paul: Good. Now we marketing has defined the job...

Eric: it's up to the engineering to execute it.

Paul: That 's our core advantage, marketing is job 1, and manufacturing is job 2...

Eric: ...Craig said manufacturing was job 1.

Paul: I switched the priorities around.

Eric: So, design is job 3.

Paul: Exactly. I couldn't help laughing when AMD promoted a chip designer to President and COO...

Eric: That's dumb, but the way they did it attracted a lot of top geeks from IBM, SUN and DEC, they got titles equal to Corporate VP...

Paul: That's why we recently promoted some chip folks to senior VPs. Let me call them... what are their names?

INTEL brand change invites questions

INTEL decided to change its logo by removing the dropped "e", change "Pentium" to "Core" and replace "Intel inside" with " leap ahead". Will this rebranding relabelling generate more attention?

Yes. Of course! Now everyone will ask a burning question:
Why is INTEL doing this after so many years of marvelous success with "Intel", "Pentium" and "Intel Inside"?

The answer lies in the words "Leap Ahead"!

Because INTEL is desperately behind!

If the people of the world didn't know this, they now know it and they will check the benchmarks to see how Cores are doing.

I think INTEL is doing huge free advertisement for AMD.

Why should INTEL cut discounts for DELL?

I suggested that INTEL should eliminate rebates and discounts based on volume and AMD quota and establish a uniform pricing scheme to encourage wider adoption of INTEL CPUs. There is an additional reason why INTEL should cut rebates for DELL.

DELL's main business is in corporate, a market where INTEL is preferred. However, DELL uses the threat of going AMD in consumer to get better and better deals and pay less and less for INTEL CPUs. DELL is draining INTEL's life blood.

If INTEL cut rebates, those corporate customers will continue to pay whatever higher prices for their INTEL desktops. The end result is more revenue and profit for INTEL.

INQ reported that INTEL is terminating the "Intel Inside" program. Maybe INTEL will do what I suggested here. We will see soon. But I really don't think INTEL has an alternative. Keeping DELL happy is hurting INTEL itself.

I will buy an x86 powered Mac

I do think Apple's switch to x86 has great potential, running standard components and processors, and with UNIX security, the MacOS will be a great substitute for Windows.

I believe Apple is going to support AMD64 eventually. However, even if it is an INTEL box, I will buy one just to try it out, as long as it does not use the dreaded 200 watt Pentium 4.

Microsoft better run for its money.

INTEL users still hope for 4GHZ

Sadly, one year after INTEL cancelled the 4GHZ Pentium 4, some INTEL users at dell.com still hope for a 4GHZ Pentium.

Right now, INTEL's Pentium 4 processors are basically sold at overclocked frequencies. Their heat production is around 130-200 watts, and they have to use massive heatsinks and employ a technique called thermal throttling to reduce CPU speed when overheating is detected. In comparison, AMD CPUs consume up to 100 watts less. According to the Tomshardware testing, a system with the AMD Athlon 64 X2 4800+ consumes a total of 203 watts at max load, a system with INTEL's next generation Pentium EE 955 (Presler) CPU made at a 65nm process consumes 289 watts at 3.46GHZ. The Presler 3.46GHZ CPU consumes 86 watts more, which indicates that the Presler 3.46GHZ produces almost 200 watts of heat( X2 4800+ is 110 watts, 110+ 86= 196). When clocked to 4GHZ, the Presler system consumes 307 watts.

When INTEL released the Paxville Xeon CPU, it has to redesign the platform to support heavier heatsinks. When Tomshardware tried to do a stress test on dual core Pentium D, the extreme heat from the INTEL CPU burnt several motherbards in a row. According to a white paper by rackable, heat affects overall system reliability and service life substantially. While a hot processor may run well for months, it can cause system to fail miserably afterwards.

In fact, to solve the heat production problem, INTEL has decided to abandon clock speed altogether. In 2006, all those 3.x GHZ Pentium 4 chips will be phased out, and will be replaced by more honest CPUs with shorter pipelines and lower frequencies. While the Pentium 4 has 31 pipeline stages to jack up frequency, the Merom core is expected to have a 14 stage pipeline and 2GHZ frequency.

Wednesday, December 28, 2005

Will Bill Gates bow to the dude?

INQ reported that DELL is doing a trial of pre-installing FireFox on PCes. This is good for FireFox, because it saves people the labor of downloading a 5MB file, a 1 minute task on a DSL line.

However, what are the dude's intentions here? Will Bill Gates bend over backwards to cut Windows XP prices so the dude can make more money selling it?

Wall Street Crooks Busted

The fool knows better. Let me quoted it:
[In fact, in every single year of the new millennium, stocks rated "sell" by Wall Street have outperformed stocks rated "buy" or "hold." From 2000 to 2004, stocks that the Street has been telling you to sell have risen 19% per annum on average. Meanwhile, the "buys" and "holds" have risen just 7%.

So is Wall Street just stupid? No. On the contrary, the analysts working on the Street are pretty bright guys -- but that doesn't necessarily work to your advantage.

You see, just because a sell-side analyst believes a stock is a dud, that doesn't necessarily mean he's going to clue you in to this fact. Think back to that 2002 TV ad from Charles Schwab, where the pinstriped broker tells his analyst: "Let's put some lipstick on this pig!" That's how the game was played pre-millennium. Analysts might have had private reservations -- heck, they might have had a private case of the giggles -- about the prospects for a stock. But that certainly didn't mean they were going to rate it a "sell."

Remember, these people have commissions to earn. And they don't make commissions by telling people who don't own a stock that they shouldn't own it. They make commissions by telling people who don't own a stock that they need to buy it right now.
And of course, if their private belief that the company is doomed turns out to be right, there's plenty of time down the road to make a second commission by opining that the circumstances have now changed, and it's time to sell that stock they told you to buy three months ago.]

Joe Osha, UBS and the gang rated INTC "buy,buy,buy" with target prices as high as $37, and INTC keeps going down, as institutions perform heavy dumping of the stock.

Joe Osha and the gang keeps downgrading AMD, and the stock keeps going up.

The other stock the Wall Street keeps pumping is DELL. But, folks, mark my word, DELL is the next Enron. A screw driver company with 80 billion market cap crashing onto $5 billion assets sooner or later will shock a lot of folks yet again.

But the fool has explicitly warned you, those Wall Street analysts are basically crooks who will steal your money.

Saturday, December 24, 2005

INTEL Bensley is only good for 2 way computing

After I wrote this article about Bensley copying Athlon MP, I realized that there is a major problem for INTEL next generation Bensley platform (Dempsey+Blackford): how can INTEL do four way SMP on this chipset?

You can see from the diagram that there are 2 buses off the Blackford chipset, these two buses must be correlated to provide cache coherence for the two CPUs. In a shared FSB, each CPU snoop on the same bus to maintain cache coherence, basically the CPU listens to the bus to see what other CPU is doing to memory. In the digram made by acehardware.com for Athlon MP there was the "snoop bus" for a separate channel, indicating the two buses "snoop" on each other to have cache coherence.

For the same reason, the two buses on the Blackford chipset can't be truly independent, they must carry cache coherence information, or the whole thing will be broken. For instance, suppose Dempsey CPU1 on bus 1 modified its cached copy of memory location X, this information must be propagated via bus1 to the Blackford chipset then thru bus 2 to Dempsey CPU 2.

If one compares the INTEL Blackford chipset to the IBM Hurricane chipset, they look very similar to each other: two FSBs off the chipset. The difference is, on the Hurricane, each FSB can have two CPUs attached, there are three scalability ports that can be connected to other Hurricane chipsets to form larger SMPs. We don't see such scalability ports on the Blackford chipset. ( I expect INTEL to copy the design of the Hurricane chipset.)

In the Hurriacane diagram, two CPUs share a 667MHZ FSB, which makes each CPU having an average of a small but acceptable 330MHZ bandwidth. However, since Dempsey is dual core, if I hang two Dempseys to one of the buses off the Blackford, I get 4 CPU cores competing for the single bus -- deja vu all over again -- we know Xeon with shared FSB scales badly to 4P.

It is like AMD introduced multi-core for the sole purpose of destroying Xeon scalability---INTEL doubles the number of buses? No problem, it has to double the cores also, and again INTEL gets the same core/bus ratio. It 's very doubtful that INTEL can put four buses on the chipset though.

With Direct Connect Architecture, AMD's Opteron 8xx can do glueless 8 way SMP without any chipset. AMD is readying glueless 32P computing with Direct Connect 2.0.

INTEL Bensley (Dempsey + Blackford) copies Athlon MP

A four way Opteron system has a D4 symmetry, a four way Xeon system is not symmetric at all -- Sharikou


INQ had this insightful article which compared the architecture of the next generation INTEL Bensley Platform based on the Blackford chipset with the 1998 Alpha EV6. You can see from the block diagram, basically, INTEL Bensley/Dempsey/Blackford architecture will be identical to Alpha EV6, with a chipset which has two independent buses.

We know DEC Alpha folks have been working at AMD, Dirk Meyer, an Alpha architect, is now AMD's COO and head of the CPU business. Alpha's lead architect is also at AMD. So it should not be surprising that AMD used this same EV6 architecture prior to Opteron (Opteron is pretty much Alpha EV7 for x86).

Indeed, look at the Athlon MP's block diagram (scroll down to the bottom of the linked page, you will see the diagrams) at acehardware.com and compare it to Xeon MP and Bensley platform, you see Athlon MP had pretty much the same architecture as Alpha EV6.

The 2001 Athlon MP had the dual FSB architecture. And that will be INTEL's next generation Bensley/Dempsey/Blackford.


Left: Xeon with shared FSB; Right: the Athlon MP with dual FSB





Acehardware.com wrote back in 2001:
"The AMD-760MP chipset has two 2.1 GB/s connections, one to each processor. In other words, the equivalent of a 4.2 GB/s shared bus. The two processors of a Dual Pentium III Xeon must share a 1.066 GB/s bus, while the two Pentium 4 Xeons have to share a 3.2 GB/s bus. "

As we can see from this benchmark, the discontinued Athlon MP 2GHZ/512KB cache was quite competitive against Xeons (3GHZ) and low end Opterons (1.6GHZ). AMD discontinued Athlon MP because Direct Connect Architecture is much much better.

The conclusion is unequivocally this: INTEL's Bensley platform based on the Blackford chipset is no more than an imitation of Athlon MP, which in turn was based on Alpha EV6. Athlon MP was for 2P servers, the Bensley/Dempsey/Blackford will also be 2P only, so both Athlon MP and Bensley are in for the same market. We should also notice that IBM's Hurricane chipset is also similar to the discontinued AMD-760MP, but with ports for scaling up.

In other words, INTEL's 2006 Bensley platform will be basically the same as 1998's technology by design, with dual FSBs. Of course, the Bensley platform improved on Athlon MP by having higher frequencies on the CPU and the buses.

And, we know that INTEL's Merom will be a Pentium III (1999) with some Pentium IV charateristics added.

The AMD64 (Opteron, Athlon64, Turion 64, Sempron 64) architecture shines because it has embedded memory controllers to reduce latency and 1GHZ HyperTransport links for I/O and communications between CPUs. There is no FSB in AMD64. Opteron can scale linearly up to 8P SMP without any chipset, this is called glueless 8 way SMP. The next generation of Opteron can scale to 32P glueless SMP.

Below is a diagram for Opteron published on vanshardware.com in 2002, as you can see the four opterons were connected to each other directly (gluelessly) with 800MHZ cHTT links (new HTTs are 1GHZ), each Opteron has its own memory, there was no NB chipset involved--remember the year, it was 2002. 2006 Bensley copies 2001 Athlon MP, 5 years behind on that, and we won't see something similar to Opteron from INTEL until 2009.

Thursday, December 22, 2005

Hector calling Malaysia

Based on this news.

Recorded Announcement: Welcome to AMD Malaysia, where customer-centric innovation is our prime directive...

X: AMD Malaysia, X speaking...

Hector: What is going on? Some big customers are calling me, they need more X2s. Hire 500 temps and get them done!

X: Sir. We are running out of plastic.

Hector: what??

X: Sir. We planned a max of 20% increase of shipment over Q3, with 5% cushion. You see, actually, Mr. Rivet only projected 7-13% increase. But we are now over 25%.

Hector: what?!

X: I am calling the suppliers to get more, but they are asking for double price, up to 50 cents each.

Hector: damn. That's robbery, plain and simple! Pay them whatever money they want. Wait, do we have dies?

X: Fortunately, FAB36 dies are coming over now, they are loading a 747 Cargo plane. we have no shortage of that.

Hector: Thank gawd, thanks, bye....Get Mr. Rivet on the phone...

Rivet: Sir, we will have a blow out Q4, beating estimates huge.

Hector: I know, but Wall Street crooks will say we have a history of spotty execution and failed to capture the 40% Q/Q growth potential... or is Henri Richard doing his job too well?

Rivet: Well, I heard his dual core duel video generated a lot of hits. SUN borrowed the idea and is now challenging DELL for a duel.

Hector: The dude is weak and scared, he won't show up.

Rivet: I thought so too. How do we handle the the crooks on the street?

Hector: We can't control their mouths, they can always find ways to say bad things. But let's send out a message..

Rivet: what message...?

Hector: Let's hold an analyst meeting to show some ass kicking stuff, our next gen mass destruction weapons.

Rivet: got it, I will allocate 5 million dollars for that event.

Paul and Craig conversation

Paul: Craig, are you all right? I heard a smashing sound in your office..what is wrong with that chair?

Craig: I am fine. This mexican, he dodged my bullet.

Paul: I told you. Dumping flash is not a long term solution. Now, he IPOed the sucker. Let's move on.

Craig: No. We should punish him more. He still got 40% of it, worth a billion dollars of stock. If we keep doing it, NOR flash will keep falling, and his 40% will worth 0.

Paul: But, we are having a chipset shortage. Remember, chipsets are made in the same 200 mm FABs.

Craig: so?

Paul: We are losing a lot of money on NOR, and we can't sell a CPU without a chipset. It's like shooting ourselves twice in the foot.

Craig: We still got 13 billion cash, enough to crush him 10 times.

Paul: Craig, you may not know, our cash is running out, we had to issue 1.4 billion convertible debt...

Craig: How come?

Paul: We had this 25 billion stock buy back program, you told me to hold the price at all costs. But at 50 million shares volume, that's 1 billion dollars a day....

Craig: So what? Let's finish him first. Then we can make all the moeny we want.

Paul: It is not that simple these days. His lawyer has called to arrange video depositions of both of us, we will be under oath...

Craig: ... will Mike tell them anything?

Paul: Not sure. We can't count on the dude not telling the story, he only cares about his own money.

Craig: You wouldn't tell what I just said, would you?

Paul: Of course not! consider our conversation privileged...

Craig: OK. Forget about NOR, let him have his 40% stock. What's the situation and what are our options?

Paul: We got serious problems on all fronts.

Craig: Really? Start from the smallest one...

Paul: This OLPC project is very dangerous. He is giving every child of the world a laptop.

Craig: I called it a $100 gadget.

Paul: Yeah, I know. But imagine every child of the world growing up with his chip. It's like kids growing up with McDonalds.

Craig: That's attachment, I never thought that mexican is good at this...

Paul: Excatly. It seems he had been planning all these stuff against us for years while we were sleeping. We underestimated him..

Craig: Definitely...arguably.. he is smart, allright? Why can't we put out something similar to destroy him?

Paul: His chip is exteremely low cost. It integrated graphics card, memory controller, USB and other stuff all onto the CPU itself. If using our chip, the PC would need several other chips, and there is no way for us to push the price down to $100. Also, his chip is low power.

Craig: Our centrino is also low power.

Paul: Ours is 25 watts. His is 0.9 watt and does not need a heatsink.

Craig: what about servers?

Paul: He is going to sell an Opteron server for every class room that has 50 OLPC laptops...

Craig: gadgets!

Paul: ..one opteron for every 50 OLPC gadgets.

Craig: we should dump the hell of Xeons and kill the Opterons.

Paul: But we can't make Xeons cheaper than Pentium IVs, or people will just buy Xeons for desktop. Desktop is still 60% of our business.

Craig: I see, let's hike prices on mobiles to make up the loss in servers.

Paul: But he also got this Turion notebook chip, if we hike mobile price, he will eat our notebook share...

Craig: what are the other options?

Paul: we have none ....

Wednesday, December 21, 2005

32 bit x86 computing has its places

You may not need the extreme performance of 64bit all the time. A 32 bit x86 PC the size of a credit card? You bet!

"The CM-i686 comprises an AMD Geode CPU at 266MHz with FPU, MMX and integrated graphics controller, SDRAM, Flash Disk, 100Mb Ethernet, USB, sound, serial ports, infrared interface, general purpose I/O, real-time clock, PS/2 keyboard and mouse controller, parallel port and interfaces to floppy and hard disks....

The CM-i686 has integrated ISA, PCI and LPC buses for interface to off-board expansions."

Monday, December 19, 2005

What if you buy a INTEL 32 bit Centrino or Yonah?

INTEL claimed that need for 64 bits is not there, no one needs more than 4GB on notebooks, Pentium M 32 and Yonah 32 are just fine. Wrong!

Modern CPUs are all 64 bit. The PowerPC for Apple Macs, the GameCube, the PlayStation 2, the Xbox 360, the coming PS3, the SUN UltraSparc, the IBM Power5 and the AMD Opteron, Athlon64, Turion 64 and Sempron 64 are all 64 bits.

In Sun Java VM and Microsoft C# (.NET CLI), the long integers are 64 bits. On 32 bit machines like Yonah and Sossaman, such 64 bit integers have to be emulated using two 32 bit ones. There is no such performance degradation on native 64 bit CPUs such as AMD64.

The PC industry is moving to 64 bit, as INTEL slowly followed AMD's lead into x86_64. Microsoft has released Windows XP x64 and Windows Vista will be a major push for pervasive 64 bit computing. Bill Gates made this a major thing.

32 bit will be obsolete by the end of 2006.

For AMD64 users, they can download any Windows program, 32 bit or 64 bit, and run them equally fine. This is because AMD64 can run 64 bit and 32 bit programs simultaneously. I am doing this right now with my Windows XP x64 on Athlon 64, all my old 32 bit applications run without change.

For Yonah and Sossaman users, what they will encounter will be this: after hours of download, when they click the program exe to run, it pops a mysterious message saying: "Sorry, this is not a valid Windows program".

Our poor fellow is going to call support, put on hold for 30 minutes, then get an explanation if he is lucky:"The app you downloaded was 64 bit and your Yonah32 can't run it". If he is unlucky, the support guy will have no clue and will ask the poor guy to reformat hard drives and reinstall Windows 32 bit version.

To use a metaphor, AMD64 is like a DVD-ROM drive, it can play both DVDs and CDs. The 32 bit Yonah and Centrino is like a CD-ROM drive, it can only play CDs. Buying Yonah or Centrino is like buying a CD-ROM drive when DVD is getting popular.

Why do you listen to Wall Street Analysts?

Cody Acree, an analyst at Stifel Nicolaus, cut his ratings on AMD, the reason?

Acree said AMD's downgrade isn't due to what he called 'any impending problem," but that investors are getting a better grasp of what are the No. 2 chipmaker's strengths.

"AMD has done an impressive job of changing its image from a value-choice, lower-end supplier, with little real threat to Intel, to a technology leader that is setting the performance bar for Intel to try to duplicate".

I don't get it. AMD is leading and more people get to know it. Isn't that great news given AMD's small percentage of market share and the 80% pie to eat?

Instead of spewing hot air in random direction like a headless fly, I suggest Mr. Cody Acree do some simple exercise of reasoning:

1) Is AMD brand and tech leadership getting more recognition in the enterprise market? (hint: 38% four way server market in Q3 05).

2) Is AMD getting more traction in consumer space? (Hint: 67.7% market of retail desktop and 31.5% notebook in Oct 2005)

3) Does INTEL have products in the next six months to compete against AMD?

4) How is AMD capacity going? Is FAB36 cranking? (hint: here)

5) Can AMD sell its products?

6) What will be the revenue and earnings numbers?

7) Finally, what will be the target price?

Looking at Cody Acree's track record, he had zero accuracy on AMD.

SUN should do a stock buy back

INTEL has spent 40 billion dollars and bought a tiny fraction of its stocks back, people have frowned upon this futile effort. DELL spends all its earnings to buy back stocks, with little good effect at all, except probably for those execs with a lot of options.

The situation is different for SUN. SUN's stock price and market cap is so low, it can literally buy back 50% of itself. I heard that SUN's offices are substantially empty, I suggest SUN sell some of its real estate and use the cash gain to buy back stocks. SUN has 3.41 billion outstanding shares and is priced at $4, it can easily buy back 1 billion shares. Buying back will boost stock price and make employee options more valuable, which will be a major incentive for those top talents to stay at SUN.

SUN should stop wasting money on buying more companies. I think SUN has got enough technology to beat everyone else, the problem is not that SUN does not have technology, the only problem is that SUN must use the technology to make money. HP has no processor, no OS and no enterprise software stack, but is making billions on x86 servers. Solaris 10 + T1 + Galaxy is a winning combination. It all depends on execution.

Sunday, December 18, 2005

What do you get for $70K these days

Go to here, and try configure. I got 16 Opteron 880 cores, 128GB ram, 2.7TB 10K SCSI. Try dell.com, see you can anything closer to this.

IWILL, H8501 Octal Opteron™ SCSI Series 5U Rack Server
$71172.53

IWILL, H8501 8-way Opteron™ 5U Rack Server Platform, DDR400 ECC, Video, 4x GbE LAN, Socket 940
AMD, 8 x Dual-Core Opteron™ 880 2.4GHz, HT 1000MHz, 2MB (2x1MB) L2 cache, Socket 940, Retail
KINGSTON, 128GB (32 x 4GB) PC2700 DDR 333MHz CL2.5 SDRAM 184-pin DIMM, ECC Registered, Quad Rank
VIDEO, Integrated ATI® Rage™ XL SVGA 8MB PCI Video Controller
SUPERMICRO, 5-drive Hot-swappable SCSI mobile rack, SCA Ultra320/160, SAF-TE, 3x5.25" drive bays, Black
LSI LOGIC, MegaRAID SCSI 320-1, Single-channel LSI53C1020 Ultra320 SCSI controller, Retail
MAXTOR, Atlas 10K IV 36.7GB, Ultra320 SCSI, 8MB Cache, 80-pin
9 x SEAGATE, Cheetah 300GB, Ultra320 SCSI, 10K RPM, 8MB Cache, 80-pin
SERVICE, Standard Service (3 Years Limited Parts, Lifetime Labor Warranty)

HP beat IBM at its own game

It is kind of funny. IBM piled up $1.827 million dollars , with a eServer xSeries 366 machine with 4 Xeon MP 7040 dual core CPUs, it achieved a TPC-C score of 221,017 tpmC. IBM was using IBM's DB2 UDB 8.2 and Windows x64. HP had a $0.48 million DL585 machine with 4 Opteron 880 dual core CPUs, the score was 202,551. HP was using Microsoft SQL Server 2005 (64bit).

Today, I stumbled onto this new TPC-C report by HP submitted on Dec 5, 2005. It's again the same DL585 with 4 Opteron 880s, the cost is $0.47 million, but with a TPC-C score of 236,054 tpmC. This beats the $1.827 million IBM xSeries 366 by 6.8% in TPC-C performance.

The difference? HP was using IBM DB2 UDB v8.2 in the new TPC-C test!

I think HP is smart, they don't have any religion. Unlike SUN, which always uses its own compiler and Solaris 10 in tests, HP choose whatever fastest.

By the way, the $1.8 million IBM Xeon machine won't be available until March 31, 2006, one day before April fool's day. The HP DL585 had been available since December 5, 2005.

INTEL Dual die Pentium Presler XE should be good enough

For frying two eggs simutaneously. Temperature is at 90C.

Saturday, December 17, 2005

Why did AnandTech handicap the Opteron?

Jason Clark & Ross Whitehead at AnandTech has conducted a benchmark on the future INTEL Bensley platform with the Dempsey CPU and Blackford chipset. However, one eagle eye on AMD board spotted something unusual, Anand's engineers setup the machines with 8GB ram and then ran the tests under 32 bit Windows under PAE mode. The following was from their setup:
"Windows 2003 was configured with /3GB and /PAE switches in the boot.ini to support the 8GB of memory used for our tests. SQL Server Enterprise was set to use AWE extensions, and a maximum memory limit was set at 6144MB."

This is werid. Both the Opteron and Dempsey are 64 bit chips. We know that runing Opteron under 32 bit in PAE mode is much slower than native 64 bit mode. We know the Opteron runs up to 170% faster under 64 bit mode. In most newer benchmarks published by HP and others, Windows Server x64 and 64 bit SQL server were used. PAE mode is for suckers, Linus said PAE is basically like DOS extender and is slow. AMD only added PAE mode as a compatibility measure.

And the test AnandTech ran was some kind of custom script, only two tests. Why didn't they run some standards tests? Such as the Apache bench, the MySQL bench, and the Unix bench Anandtech had been using?

Even with both hands tied to the back, the Opteron 280 tied with the future Dempsey while consuming about half the power.

I hope this Anandtech article was not paid by INTEL and I would like to see some realistic tests. Having 4 Opteron 280 cores and 8GB ram running 32 bit is like running Pentium IV in DOS only mode.

AnandTech had a benchmark testing AMD64 and INTEL EM64T under both 32 and 64 bit mode. The conclusion was INTEL runs slower under 64 bit than 32 bit mode, while AMD64 is faster in 64 bit mode. If this is still the case with Dempsey, then Anandtech may run each chip in its best light: INTEL in 32 bit and AMD in 64 bit.

INTEL looks to India for salvation

The Whitefield project is for a next generation multi-core INTEL Xeon processor with CSI interconnect. CSI is supposed to compete against HyperTransport. A financial audit scandal in India revealed a lot of information about the Whitefield project, which was terminated by INTEL as the project lead Srinivas Raman exited the company.

Still, INTEL's hope to compete against AMD lies with the 600 Indian engineers. While INTEL's middle-eastern team has successfully adopted Pentium III into the Pentium M, it has apparently failed to add iAMD64 instructions to that architecture after 3 years, as INTEL's next generation Yonah mobile and Sossaman server CPUs will be 32 bit only.

A new Xeon project was assigned to the third world CPU design center at Whitefield, competing against top designers from IBM, SUN, DEC and AMD.

Friday, December 16, 2005

AMD attracted a lot of super talents

AMD's CTO Phil Hester was CEO of Newisys, and formerly one of the 15 member IBM Corporate Technology Council and led the RS/6000 development. We heard during the recent AMD analyst meeting that some INTEL Architects are working at AMD, including the one who designed the INTEL Xscale processor.

Today, I stumbled on this AMD page for hiring information, which seems to be introduction of new grand masters in one of AMD's design team:

AMD SENIOR FELLOW, Former Chief Engineer of PowerPC 601 and Power4: I came to AMD after spending two years doing research into technology scalable computer architecture at UT-Austin. Prior to that, I was a distinguished engineer at IBM in Austin where I was the chief engineer on the Power4 chip. For me, AMD represents a dedicated team of people committed to bringing the right "common sense" products to market. The technology developed at AMD is clearly some of the best in the industry, and I believe that AMD will grow rapidly because of this. I am leading a next generation design, and look forward to extending this lead.

AMD CORPORATE FELLOW, Former IBM Fellow, and Chief Architect of Power, PowerPC, xSeries 440,445, and CTO of Newisys: I came to AMD after 5 years at Newisys building enterprise class Opteron based servers and scalable Opteron servers and 30 years at IBM - mostly at Watson Research - working on many architecture innovations from the 801 (first RISC machine) to the large scale scalable x86 systems. I was appointed and IBM Fellow in 1994. I view AMD as the next logical step in my career, where I now have the opportunity to help set the direction for future of industry standard servers as AMD takes Opteron to the next level. I have found my job and AMD to be a very exciting place to work, full of very talented people, all of us working to change the server landscape.

AMD FELLOW, Former Chief Architect of UltraSparc III: After leading the UltraSparc III architecture for seven years I spent a couple of years in Sun Laboratories working on the system architecture for a DARPA Peta-scale supercomputer. I was attracted to AMD by their bold move in defining the AMD64 instruction set and then delivering an impressive implementation of the ISA in the form of the Opteron processors. When I talked to the impressive team of talent that AMD assembled to design their processors I was convinced that this was a team I had to be part of and that would change the future of microprocessors.

AMD CORPORATE FELLOW, Former Lead Architect of Alpha Architecture: Alchemy, the startup where I was CTO, was acquired by AMD in Feb, 2002. Previously, I had major technical roles on development of VAX, Alpha, ARM architectures and microprocessors. I am currently leading mobile systems team for AMD and am excited about the future generations of mobile design where AMD can bring real value and battery life leadership to Windows based mobile systems. For me the AMD vision of providing value to the customer, which is what the customer wants and needs, versus the "current trend of the year" is very rewarding. The MTS technical ladder at AMD goes up to Corporate Fellow which is the equiv. of a Corp VP. This shows that AMD values and rewards Sr. technical contributors without having to branch into management.

Why Former INTEL Employees Hate INTEL?

I have some friends who were former INTEL employees (they got Ph. D degrees). They hate INTEL so much, whenever I bring up AMD stuff, they applaud, and say stuff like INTEL is crap. One thing I heard was that INTEL's peer reveiew system and management was very harsh on any human being.

Today, INTEL announced that it will cut stock options for employees, except "[t]he most senior employees". I bet that "the most senior" is a tiny percentage, and most of INTEL's engineering workforce will work harder and get nothing. Any way, I don't think INTEL stock options worth that much if you consider the impending AMD flood of superior server, desktop and mobile CPUs out of FAB36 and Chartered. The 150 million notebooks off the OLPC program will also eat INTEL market alive, when people find the 0.9 watt 500MHZ AMD CPUs does everything an INTEL Yonah can do: 32 bit x86.

In stark contrast, former AMD employees seem have a strong emotional attachment to AMD. For instance, NVIDIA CEO Jensen Huang was a chip designer at AMD many years ago, today, when he talks about AMD, you can sense he still has some feeling towards it. "What is good for AMD is good for NVIDIA", that's true for business, as NVDIA has 90% of the AMD64 chipset market, but there is more, there is a human factor which is more than money.

Judging from JFTC findings and AMD's allegations in its lawsuit, INTEL is a dirty business, where greed overweighs everything else.

INTEL users may need to beg Microsoft for multi-core qualification

According to Microsoft, INTEL's multi-chip CPUs won't be considered multi-core and won't enjoy the benefit of counting as one CPU. This means buyers of INTEL multi-die techniology not only get slower chip with a FSB but also need to pay dual or quad license fees.


From: Sales <sales@microsoft.com>
To: vince
Subject: RE:'RTCProd=010-132-843' RE: License still ambiguous
Date: Fri, 16 Dec 2005 10:45:08 -0800 (14:45 AST)


Hello Vince,


I was able to find out information about licensing regarding your question about multi-core processors. To answer your question, yes each die that resides on a multi-chip processor (or module) is considered an individual processor. Although the processor as a whole plugs into a single socket, it will not reduce the number of licenses required for our software.

As far as updating our online documentation, I am unsure as to when any changes would be made in consideration to these newer generation of multi-core/die processors.

Thank you,


Zahn
Microsoft Sales Information

DELL XPS 400 trashed, fragile build, poor performance

HardOCP did a review on the Dell Dimension XPS 400, DELL's attempt to get away from the $299 cheap PC business. The result? Once again, DELL proves that it better stick to the cheap section of the market.

Here are the scores:
Purchasing, Presentation and Packaging – 8.5/10
Build Quality - 5/10
Gaming – 7/10
Multimedia and Productivity - 4/10
General Usage - 4/10
Stability – 2/10
Technical Support - 4/10
Value – 5/10
The Bottom Line – 4/10

In summary, XPS 400 is a crappy, fragile, expensive, unstable piece of product from DELL. If you have money, try this, or this, don't spend the money and embarrass yourself with a DELL.

DELL laptops put burn marks on tables and desks

DELL recalls. It has received three reports of damage to table tops and desks. Possibly affected models: Latitude models D410, D505, D510, D600, D610, D800, D810; Inspiron models Inspiron 510M, Inspiron 600M, Inspiron 6000, Inspiron 8600, Inspiron 9200, Inspiron 9300, the XPS Generation 2 notebook; and Precision M20 and M70 mobile workstations.

I cried

As I saw the little child went on her own way, looking back at me one last time. She had been abused and injured deeply by the ChipGoliath, the evil devil, and she will take the time to heal and be strong all by herself. Good luck, Spansion, Good luck, SPSN.

Thursday, December 15, 2005

AMD smoked INTEL out

The latter will take a beating. Japanese court orders public disclosure of documents taken in INTEL KKK raid.

The JFTC noted that as a result of INTEL's misconduct, the combined market share of AMD and a second, much smaller CPU company fell from 24% in 2002 to 11% in 2003.

SUN needs to stablize JAVA

SUN is producing stuff faster than we can digest. We got zillions of buzzwords starting with the letter J, then we got revisions and more buzzwords and equal number of software concepts fading away daily. I guess every SUN senior engineer must be able to claim her J trademark somewhere. It makes me feel dizzy, and I lost interest in learning any of it, not sure which one will be obsolete next 6 months.

My suggestion to SUN: stop labelling every little JAVA library with the big J letter and call it JXXX technology. Just let them be what they are, some type of useful JAVA library and hope someone will find use for it. If you make it JXXX technology, even a curious person like me will feel that there is some steep learning curve and get discouraged, if it is just some library that my standard JAVA skill is good enough, I will be much more comfortable studying it.

Focus on and stablize the core, especially a J2ME standard for high end phones (forget about the low end, no money there) to make creating 3G applications easy.

Furthermore, focus on Solaris 10 usability. I am a UNIX person, I use Linux too, I will switch to Solaris 10 if I can setup ZFS and Containers without reading the manual.

My challenge to SUN: you got smart engineers, you were able to do WABI in a few weeks and beat Bill Gates to the punch, now, if you are really good, make Solaris 10 as easy as Windows Server 2003.

Everybody will sing praises then.

Wednesday, December 14, 2005

INTEL inhales the exhaust of AMD Ferrari

Our Scott Hillis, still employed by Reuters, reports again: "Lewnes dismissed the idea that, in terms of the F1 deal, Intel is inhaling the exhaust of AMD".

It is amusing INTEL always follows AMD's footsteps , "sometimes tripping themselves while doing so" (Henri Richard). No need to mention x86_64, NX bit, dual core, performance per watt... those were old stories. Recently, AMD entered India, INTEL followed.

If INTEL does not want to copy AMD, why doesn't it try something else like horse racing, instead of inhaling the exhaust of AMD Ferrari?

INTEL sells stock to buy stock

We were perplexed by INTEL's annoucement that it is offering $1.4 convertible debt . Now it is clear. INTEL has set the terms of the debt offering:

1) The debt is convertible to INTC stock at a price of $31.53 per share after December 15, 2012.

2) INTEL will use the $1.4 billion debt to buy back stocks and for general operations.

The balance sheet we were looking at was for Q3 2005, which showed INTEL having $10 billion cash. I wonder what will INTEL's Q4 2005 balance sheet look like. We will know soon.

Reuters should fire Scott Hillis right now

Reuters.com - No Spin. No Agenda. Just the Facts. As they happen.

CORRECTED - Intel touts power of new laptop chip

Wed Dec 14, 2005 01:44 PM ET In Dec. 13 SAN FRANCISCO story headlined "Intel touts power of new laptop chip," please read in 12th paragraph ... has one-upped Intel ... instead of ... is looking to one-up Intel next year ... (corrects to show AMD has already released a 64-bit laptop chip).

Original was here.

Reuters.com, like all news media, lives by reputation of being true to the facts.

If Scott Hillis lied, he should be fired, no question about it.

If Scott Hillis hadn't seen or heard the word "Turion 64", which had been all over the place since Q2 (it was even over the sky of San Francisco), then he should also be fired, Reuters can't have such low quality IT reporters spewing misinfo.

If Scott Hillis had seen or heard the word "Turion 64", but did not realize "64" means 64 bit, then he should be fired too, because he is too dumb.

Either way, Scott Hillis should be fired right now.

If Reuters does not fire Scott Hillis, then Reuters should remove its mantra and admit that it is OK to lie or its reports are low quality or some of its reporters are dumb.

Monday, December 12, 2005

INTEL is weird

INTEL has authorized a $25 billion stock buy back plan, using its $10 billion cash, that is $15 billion short in cash, at least for now.

Now it offers to sell $1.4 billion of convertible debt (news: Intel to Offer $1.4 Billion Junior Subordinated Convertible Debentures).

Hmm? Selling stock to buy back more stock? Has INTEL used up all its money in stock buy backs so it needed to borrow for operations?

Buying back stock does not change a company's value (market cap), which should be a multiple of the profit and should be constant when a company buys back shares. However, with reduced number of outstanding shares and constant market cap, the unit share price should increase.

Who will benefit from the increase of the share price?

Option holders come to mind.

Some even argue that INTEL's real purpose is to sell the debt at higher conversion price. Stock buy back is just used to increase the stock price so conversion price can also be set higher.

Sunday, December 11, 2005

Local BestBuy has zero Turion notebook on display

Went to local BestBuy tonight, looked around the notebook section, there were three bulky Athlon 64 notebooks, zero Turion 64.

A salesperson came over and offered help. I asked whether they have Turion notebooks, said no. So I started playing with a SONY Centrino. The salesperson came over, asked if I liked the SONY. I said that I needed 64 bit. I told him I needed to speak with his manager.

The manager came, I asked him whether he knew that AMD filed an anti-trust lawsuit against INTEL, he said yes. I asked him if he knew AMD has alleged that INTEL bought shelf space at BestBuy in an effort to exclude or limit AMD's access to the market. He said yes. I asked him why there was no Turion 64s on display at BestBuy. He said they sold out. I asked him to show me where the machine was, there should be a label if it was sold out. He led to me a spot that was supposed to belong to a Compaq Turion machine. The shelf space was empty, but the label was for a $749 Celeron notebook.

Two other salesperson came over, they explained that the Turions were so hot, they were sold the day they arrived. The manager said they should get another batch in a week.

I found it hard to believe that HP could not keep up with the demand, there doesn't seem to be any Turion shortage. But BestBuy said sold out. I couldn't complain much about that.

So I told the manager that as a consumer, I do not want my choice of CPUs be restricted by any illegal monopolistic scheme. I said I hope there was no conspiracy of any kind, or I can certainly bring a class action lawsuit to the parties involved. So far, only INTEL has been hit with 100 lawsuits, none of the co-conspirators in those class action suits have been named.

INTEL naming getting religious?

INTEL usually use Oregon places for CPU code names. However, the next generation INTEL CPUs seem to have Jewish/Hebrew/Biblical origin.

Yonah: Hebrew, Form of Jonah. Dove. Female

Merom: Eminences, elevations. Non-gerderized Biblical name, female.

Merom Is The Mother Of All Upcoming Processor Designs

INTEL's 2006 outlook mixed

Upside:

1) Apple may start shipping x86 Macs. Although Apple may not do a 100% switch overnite( there are a lot of PowerPC software there), it has the potential to eat a lot of MS Windows market and grow -- to the displeasure of Bill Gates. Apple currently has 2% of the global PC market, or about 4 million units.

2) The VIIV might be a hit.

3) 65nm production already started, 6 months ahead of AMD.

4) DELL dude is still loyal to INTEL.

Downside:

0) More people get to know AMD's lead.

1) Yonah and Sossaman will be 32 bit only, can't run 64 bit applications and not suitable for enterprise computing that needs more than 4GB.

2) Server segment will continue to be non-competitive, as Paxville and Dempsey both lag far behind Opteron in performance.

3) Total transition to new Pentium III based archicture in 2H 2006 will cause mass extinction of P4 architecture and slow customer purchase of P4 and P4 based Xeons including Paxville and Dempsey.

4) Xbox 360 uses IBM chip. Expect the x86 version to die soon, that is 20 million annual INTEL units fading away.

5) The $100 laptop with AMD chip will start shipping in millions, it will definitely eat some celeron market.

6) AMD FAB36 has been in production for 2 months already, finished products will flood the market in Q1 2006.

7) DELL dude will ask more discounts to reward his loyalty.

Saturday, December 10, 2005

AMD trying to do something "nasty" to INTEL?

The INQ reported that AMD is cutting productions of lower speed Athlon64s. At yahoo message board, one poster reckoned that AMD might be doing something "nasty" here: AMD might discontinue all Athlon 64 models below 3700+ rating. Since the highest INTEL chip is 3.6GHZ, this will make AMD's lowest mainstream CPU faster than INTEL's highest. Thus forcing INTEL to compete against Semprons only.

It does make a lot of sense. AMD's yield is so high and chips are so uniform, they aritifically lower the clocks of some of their CPUs and sell them as low end. You can clearly see this, as the lower frequency AMD chips also have lower power consumption.

AMD will soon implement the new process technology jointly developed with IBM that can enhanced performance by another 40%. Unless INTEL can do something quick, it will find itself at the ultra low end of the spectrum.

Friday, December 09, 2005

Dr. Barret lost his usual grace

INTEl chairman Craig Barret derided the OLPC as a gadget, not a fully functional PC.

Dr. Barret said: "Mr. Negroponte has called it a $100 laptop -- I think a more realistic title should be 'the $100 gadget'," "The problem is that gadgets have not been successful." And he added, "We work in the are of low cost affordable PCs, but full function PCs, Not handheld devices and not gadgets."

The OLPC runs AMD's low power x86 processor and Linux, it has features that even surpass the normal laptop PCes. Unlike AMD's main stream CPUs, this AMD CPU is 32bit, but it certainly can do anything an INTEL 32bit Yonah or Sossaman CPU can.

Also, Dr. Barret forgot that INTEL is a big player in handeld devices, its StrongArm based Xscale processors are used in many handhelds. Of course, we know the guy who architected the INTEL handheld chip is working at AMD now.

INTEL's predicament

AMD's FAB36 has been in production since Oct 14, 2005, finished chips should hit the market soon. No doubt, there will be an oversupply of CPUs in the market, even Joe Osha is worried about the excess capacity. What will customers buy? I think price/performance will be a deciding factor. AMD is used to selling Sempron 64s at $50 or less. No matter how you look at it, INTEL will lose market share and units.

However, what really matters to a company is not units or market share, it's revenue and profits. If INTEL can sell 100 chips for 10 billion, it will be super rich.

So, facing the inevitable decline of units and market share, the only way INTEL can maintain revenue and profit growth is to hike CPU prices.

AMD's capacity is limited and can only supply 25% of the market next year, INTEL has the say on the rest 75%. Furthermore, corporations tend to prefer INTEL brand. INTEL shouldn't have much problem increase the ASPs by 20%, to negate the effect of unit decrease (ASP increase can be done by simply lowering production of low end models).

But there is a problem: INTEL has made it a top priority to keep DELL happy. When the DELL dude makes noise about AMD, INTEL folks cave in and give more discounts, or in other words, selling at lower ASPs. And to prevent others such as HP and Lenovo from being too disgruntled, INTEL must give them some albeit less discounts also.

So, INTEL is facing a dilemma here. On one hand, market share drop is an ineveitability and it needs to hike ASP to maintain revenue growth. On the other hand, feeding DELL is a futile effort to retain market share and will result in lower ASP. (With DELL being the favourite darling, others get pushed over to AMD).

The two are in direct conflict.

Clearly, keep feeding DELL with lower priced products coupled with losing overall market share will lead to revenue and profit short fall.

It seems that the only way out is for INTEL to drop the most favoured treatment for DELL and increase prices cross the board, and make prices the same for everyone, making it possible for every one (instead of just DELL) to profit in the INTEL space and create a more balanced INTEL market.

My suggestions to INTEL:

0) Face the reality that AMD will take 25% share.

1) Abolish the volume based or AMD quota based rebate system.

2) Establish an uniform pricing scheme, set the same equal global price for everyone, just like McDonalds sell cheeseburgers at the same price. Get rid of all discounts. Such as an uniform pricing system will make small whitebox players competitive against big guys like DELL and foster a broader market presence.

3) Reduce the production of low end chips, this will lead to higher ASP.

Taking the pulse of the FAB tech

AMD FAB36 started production on October 14, 2005, approximately 20,000 12inch wafers have already been processed by now, and we should expect to see a flood of 10 million Athlon 64 in January 2006. Buckle up for the impact.

Meanwhile, Chartered FAB7 has AMD APM system installed and is ready to ramp up. AMD is also starting to choose a site for a new fab.

On semi-conductor technology, AMD+IBM has something that can increase performance by 40% once again, using a bunch of exotic methods. AMD and IBM will use the new methods immediately on current 90nm production and subsequent 65nm production.

INTEL also made some major semi-conductor break through, using a new material called InSb. "The first commercial products based on this technology are not expected until at least 2015".

Thursday, December 08, 2005

INTEL 2005 Q5 Growth 4% to 6%

Joe Osha said :"We are buyers", UBS said INTEL frogleaps... In the last few days, the folks on the street have been pumping up INTEL like crazy, expecting huge upside. In Q3 2005, INTEL revenue was $9.96 billion. Today, INTEL gave Q4, 2005 estimate at $10.4 to $10.6 billion. Revenue growth 4% to 6%. Seems to be substantially below seasonal pattern.

INTEL is facing strong competition on server front. INTEL CFO Andy Bryant said "Server orders are tough to win, and it's hand-to-hand combat in some places". Analysts are disappointed.

AMD projected 7 to 13% growth. Overall, the PC market is doing very well.

INTEL has authroized a $25 billion stock buy back plan, using its $10 billion cash.

AMD's new 300mm FAB36 has entered production and finished products are expected in Q1 2006. AMD also plans to rapidly ramp production at Chartered FAB7 and is planning another new fab. INTEL declined to give outlook for 2006.

Wednesday, December 07, 2005

How can AMD play the DELL game for fun and profit

Every 12 months or so, the DELL dude plays the same game: "We are actively evaluating AMD, it looks like a yes". Then the INTEL folks reluctantly give more discounts, until the DELL dude says: "It looked like a yes, but it turned out to be a no. But we will keep evaluating AMD".

I bet Dr. Ruiz is getting very mad being played by a dude with only a high school diploma like this.

This year, it was no different. Apparently, INTEL decided that they could give no more. Then the dude made it very real: there were news that 3 Taiwanese companies Honhai, Quanta and Asus making AMD servers, desktops and notebooks for DELL, and could flood the market any time when DELL said yes.

Of course, we know the result of this game...

Now, how can AMD counter this and stop being played like a fool?

The common wisdom is that AMD should punish DELL by refusing to deal. But I find that this is not the best approach.

AMD's competitor is INTEL, not DELL.

Here is my proposal for AMD

1) Actively and publicly engages DELL. AMD should issue regular press releases saying stuff like just sent a quad-core Opteron box for DELL to test(though later the box was found missing), sent dual core Turions to DELL(no real chips, best are sent to HP), established a dedicated team(no real engineers, janitors with regular pizza parties are good enough) working with DELL, etc, etc. DELL dude might be attracted. At least, DELL dude won't say no no no, it will be happy to play along so it can get more deals from INTEL.

2) The threat of DELL entering Opteron market will drive HP nuts, and AMD can get more HP desktop and notebook business in exchange of a promise. To achieve max effect, AMD should always send the best future products to HP (such as quad-core, dual core turion at 15 watts, etc), but issue press releases for engaging DELL (which gets to see no real goodies).

3) INTEL is deeply worried, and decides to sell iAMD64 Xeons and 32 bit Yonah to DELL at even deeper discounts.

4) HP found DELL's Xeon boxes much cheaper than HP's and very hard to compete against on pricing, so it does more Opterons and prices Opterons based on benchmarks.

5) When DELL dude is ready to sign up for dirt cheap Opterons, AMD suddenly says "it now looks like a no".

The result, DELL gets cheaper INTEL CPUs, and HP does more AMD business. The cycle repeats itself over and over.

This will be a win-win-win situation for AMD, DELL and HP, and INTEL will be the only loser.

Both AMD and DELL are draining INTEL

It is amusing to see DELL playing AMD card once again, and I am sure DELL has gotten some big discounts from INTEL again. After listening to DELL CFO Jim Schneider's talk at Lehman Brothers Global Technology Conference, I concluded that DELL actually has a lot of room to grow in desktop and notebook space, despite AMD eating INTEL's lunch. Why?

AMD's capacity is limited, and INTEL will have at least 65% of the CPU market next year. DELL now has only 18% of the PC market. With most favoured CPU pricing from INTEL and bargaining power on other components, DELL can eat a bigger portion of the INTEL pie by beating others like HP, Acer and whitebox INTEL makers.

For INTEL, the situation is like this

1) AMD is eating INTEL's lunch in server, desktop and mobile, there is no way to stop that, AMD has better technology period.

2) DELL keeps doing a bigger portion of the INTEL market, but at lower and lower ASPs paid to INTEL.

Jim said: "INTEL has been very competitive in terms of pricing with us".

Hector said: "DELL is a tough *** to get, but we will get the dude eventually".

Otellini said: "Keeping Dell happy is certainly one of our most important jobs".

Hurd said: Our INTEL PC business has a near zero margin...

Acer said: Our INTEL PC business has a negative margin....

Lenovo said: It is unfair for INTEL to give DELL better prices, we shall sue...

The universe is in perfect harmony....

Tuesday, December 06, 2005

If I were Hector Ruiz

1) I would have announced Spansion IPO plan back in June 2004, so INTEL would have lost the interest to dump flash, and I could have sold Spansion at higher price.

2) I would have sued INTEL in April 2003, on the same day Opteron was launched., but given the timing I chose, I would sue without publicity.

3) Then I would have 30% of the server market by now.

4) Then the trial would have started when I released the dual core Opteron in 2005, and I could collect 6 billion dollars in 2006 to build two more fabs.

The SUN also rises--T1000 ruthless

This is shocking, my jaws dropped when I saw the numbers. I thought SUN would sell the UltraSparc T1 (Niagra) 8 core server at a starting price of $10,000. No. SUN is doing something unprecedented here on price/performance. The T1000 is not sold at $8000, not $5000, but starts from $2995 for 6 cores and 2GB ram! With 8 cores of UltraSparc power, the T1 will slaughter competitions across the board. INTEL is dead, IBM is dead! AMD better hurry up and do something quick!

What about performance? With only one floating point unit, the T1 is not designed to do heavy lifting like the Opteron. With 4 memory controllers and 8 integer cores and 32 hardware threads, the T1 is essentially a 8 way server chip that can handle massive parallel web and transactional traffic. Look at the benchmarks, the T1 beats popular high performance servers by a wide margin. For instance, the T1 beats all four way servers on SpecWeb2005, including 4 way IBM power5 1.9GHZ; on SAP SD two-Tier benchmark, the T1 beats 4 way Xeon and 4 way Itanium servers from HP; on Lotus Domino iNotes benchmarks, the T1 beats a 8 way IBM Power5 1.5GHZ.

Let's look at some real numbers.

SPECweb2005 --

Sun Fire(TM) T2000 (8 cores, 1 chip), SPECweb2005 = 13854;

DELL PowerEdge 2850, 2x INTEL Paxville Dual core 2.8GHZ, SPECweb2005 = 4850.

SAP SD two-tier --

Sun Fire(TM) T2000 (8 cores, 1 chip): 950 SD Users (Sales & Distribution), Ave. dialog resp. time: 1.91 seconds, Throughput: Fully processed order line items/hour:95,670, Dialog steps/hour: 287,000, SAPS: 4,780

Dell PowerEdge 6850, 4-way SMP, Intel XEON MP, 3.33 GHz,L1 Execution Trace Cache, 1 MB L2 cache, 8 MB L3 cache,16 GB main memory: 540 SD users (Sales & Distribution)Average dialog response time: 1.88 seconds Throughput:Fully processed order line items/hour: 54,670, Dialog steps / hour: 164,000, SAPS: 2,730

HP ProLiant DL385, 2-way SMP, Dual-core AMD Opteron processor Model 275 2.2 GHz, 128 KB L1 cache, 1 MB L2 cache: 983 SD Users (Sales & Distribution), Average dialog response time: 1.98 seconds; Throughput:Fully processed order line items/hour: 98,330Dialog steps/hour: 295,000, SAPS: 4,920

It is clear that one T1 processor can easily beat 4 INTEL Xeon MP processors, but it is behind two dual core AMD Opteron processors (total 4 cores).

Remember this, the T1 is a single CPU that consumes less than 80 watts! The Power5 is a 150 watt monster with 19MB of cache. SUN is indeed years ahead in multi-core and CMT architecture.

Right now, the T1 is no doubt THE highest performing single CPU, at least in integer performance. An Opteron 285SE with two 2.6GHZ cores can't match the throughput of eight 1.2GHZ UltraSparc cores. But Sparc and x64 are two separate markets, in particular, there is no Windows server market for Sparc. AMD is not far behind and still has time. With quad-core Opteron, AMD can probably have a slight lead again, and AMD can do it in six months. The T1 will definitely take Linux market from the 150 watts Xeons, so will Opteron. IBM Power5 looks foolish unless you go beyond 8P, but SUN is making a multi-processor version of Niagra.

The T1 is a perfect fit for webhosting and video serving.

AMD should learn from SUN so it can make better multi-core x86_64 chips, maybe AMD can also make simplied x64 cores with focus on integer performance to reduce die size; SUN should ask AMD to help in manufacturing, AMD's DSLSSOI can increase speed by at least 30%, and AMD's APM 2.0 can improve yields to 95%.

Monday, December 05, 2005

Steve Jobs won't be an idiot

This ZDNET blog analysed the Apple on INTEL situation and concluded that there won't be Intel powered iMacs or iBooks in 2006. The argument makes sense, Apple fans have been enjoying 64 bit computing for a quite a while now, and Steve Jobs surely won't push them back a big step by adopting the 32 bit Dothan or Yonah. Steve Jobs is no idiot.

AMD can save Apple's day, the Turion64 is sexy, cool and powerful. The Athlon 64 FX60 will beat the hell out of any INTEL chip.

INTEL leadership lacks basic wisdom

INTEL lagged behind AMD for 6 years back in 2003 when AMD introduced its Opteron CPU. AMD started to show sign of life, both its CPU business and flash business were making some small profits, Opteron gained 5% server market, AMD folks were getting a bit cocky. At the time, I proposed that AMD should spin off Spansion right away and focus solely on CPU business. Apparently, AMD folks decided to milk the cash cow a bit more.

INTEL folks got so jealous of AMD, they decided to dump flash like hell to punish AMD. Overnight, NOR flash price dropped 40%, and overnight, AMD found it was losing big money on the flash business. AMD's stock fell like a rock.

But in restrospect, INTEL execs were simply not shrewd. What they did pushed AMD near the edge of a cliff, and forced AMD to fight back with all its strength and 1000% more.

Quickly, AMD decided to spin off flash, and pushed dual core out the door, and then launched the deadly lawsuit.

Look forward 3 years, INTEL is in deep doodoo. INTEL is so far behind in servers and desktops, there is zero hope that it can catch up, Paxville was a big disappointment, and Dempsey shows no promise, Yonah is slow and 32 bit... And, AMD's progress is accelerating onto 32P glueless multi-processing. On the flash ground, Spansion, an independent company is going to gun down INTEL's NOR business with its cost effective mirror bit technology.

If INTEL execs were smarter, they would have let AMD folks gain a bit here and there, making small money here and there and make sure AMD folks feel no pressure, so INTEL could buy the time to catch up. When you were far behind, the worst thing you could do is wavering big flags and shouting....

Now, it's too late.

Geode EPIC for Windows CE

I found the title of this article amusing: Geode EPIC dev kits support Windows Embedded.

Geode @ 1.1watt is ruling the EPIC architecture? I am sure it has more performance per watt than Itanics.

Sunday, December 04, 2005

Missing Turion MT or INTEL's mafia hand?

AMD has two kinds of Turion 64 mobile CPUs, the MT version has a max power of 25 watts (2 watts below the 27 watt Centrino/Pentium M), the ML version has a max power of 35 watts. The price difference between MT and ML is only $5. See AMD's pricing list.

You can search "Turion MT" on www.froogle.com, and there are a lot of such CPUs for sale plus a couple of notebooks built with this "Centrino killer". Most puzzling is that HP has no notebooks built with Turion MT, only the Turion ML, which tends to have lower battery life. In the US, Averatec seems to be the only company that uses the Turion MT.

What is going on? Is AMD having problem making Turion MT? Judging from the tiny price differential ($5), it does not seem to be the case. Also, you can buy Turion MT CPUs on the internet.

Or, is INTEL doing its dirty work again, limiting HP's use of the Turion CPU by the threat of cutting off Pentium M supply? If you recall the alleged incident that INTEL withheld CPUs to Compaq and forced it to drop AMD products, this kind of conspiracy theory is not that far fetched.

On server and desktop, INTEL is totally knocked out on performance. The only area INTEL is showing relative strength is mobile, there, the ONLY advantage INTEL has is perceived longer battery life. INTEL mobile CPUs have no 64 bit, no NX-bit, no floating point performance, no nothing. Perceived longer battery life is the only thing left there.

I used the word "perceived battery life" because careful tests do not show INTEL having quantifiable advantage. LaptopLogic has a comprehensive benchmark that compares Turion ML 37 (2GHZ, 1MB cache) and Dothan 760 (2GHZ, 2MB cache), the Turion won most of the tests. The Turion ML 37 won 15 tests, the Dothan 760 won 7 tests. The Turion notebook and INTEL Dothan notebook had almost identical battery life, even though ML instead of MT version was used. Considering that the AMD64 platform does not need a separate NB chip, this can be expected.

INTEL delayed Opteron for two years

AMD was the world's first CPU maker to break the 1GHZ line, the Athlon CPU introduced in 1999 had a lead over INTEL and at one point had over 20% of the market. At the same time, AMD started developing the Opteron, which was code named Hammer.

We all know what happened next: INTEL signed exclusive deals with big Athlon PC makers such as Sony, NEC, Gateway, and put strict AMD quotas on others such as Compaq. Overnight, AMD market share dropped to 15% or less and was largely pushed into the whitebox market, its revenue plunged. AMD had to borrow money to survive, it almost bankrupted. AMD had to cut R&D. The Opteron was delayed again and again, due to lack of resources. Eventually, AMD launched Opteron in April 2003.

Some people expected Opteron to gain substantial market share due to its 2x performance lead over Xeon, but it did not happen. Until June 2005, Opteron had only 5% server market share, and did not show any sign of fast growth.

Everything changed after June 2005. In one quarter, Opteron revenue jumped 89%.

What happened? AMD filed the anti-trust lawsuit against INTEL. Immediately, we see almost 180% degree turn from many major vendors, from OEMs to retailers and distributors. They all started to use AMD for one reason or another.

For instance, sometime after the lawsuit, I went to curcuitcity.com and searched "turion", every time, the site sent back a programmed HTTP error, searching any random string, I got a well formatted web page. I emailed circuitcity.com, and surprisingly, they acknowledged it and fixed it the next day and the search found the Turion notebooks.

As another example, before the lawsuit, Supermicro designed a whole slew of Opteron motherboards, they were initially sold under Supermicro brand on some online stores, such as monarchcomputer. INQ reported the availability of such boards. Immediately, these boards were taken off the web site. When weeks later these boards reappeared, they were all marked as "Monarch" motherboards, which did not inspire confidence, because we know Monarch is not a MB maker. Today, 5 months after the lawsuit, you can go to monarchcomputer.com, and you can see the Supermicro Opteron boards under the SuperMicro brand.

Similar phenomena were observed in retail stores. AMD alleged in their lawsuit that INTEL sets quota of shelf space for AMD in retail stores such as bestbuy and circuitcity. Before the lawsuit, when you go to BestBuy or CircuitCity, you saw far more INTEL PCes than AMD PCes. After the lawsuit, you see almost equal number of desktop PCes from both AMD and INTEL. The result: AMD now has 67.7% of the US retail desktop sales and 31.5% of the notebook sales. AMD surpassed INTEL in total PCes (desktop+notebook) sold in US stores.

These facts tell us this: people don't care for the INTEL brand, people are not stupid, when they see 64, they know it is 2 times of 32. Once AMD64 is given a fair chance, it flourishes on technical merits. People have spoken with their wallet.

Saturday, December 03, 2005

INTEL enterprise roadmap hopeless

Tomshardware published an INTEL roadmap which goes as far as 2008. Judging from the information presented, INTEL probably lags even more behind AMD in CPU architecture and multi-core design. When we look at modern CPU designs such as AMD Opteron, Alpha EV7 and Sun UltraSparc T1, they all have one thing in common, they have on die memory controller to reduce memory latency. The UltraSparc T1 has 4 memory controllers, and we expect the socket F Opterons to have at least 2 memory controllers. Opteron and UltraSparc T1 are true multi-core designs as multiple CPU cores reside on a single piece of silicon connected with internal multi-GHZ links. However, in INTEL's dual core design, the two cores don't know they are sitting next to each other on the same die, they go out to the FSB to communicate, which is a very primitive design. The next generations of INTEL server CPUs won't even pretend they are multi-core, instead they will be multi-die--basically INTEL will pack multiple CPU dies into the same package and call it multi-core, these cores don't talk to each other directly any way.

The INTEL Dempsey (Netburst/Xeon) will be dual die with 4MB cache, expected release date is mid-2006.

The INTEL Whitefield will be quad-core with 8 to 16 MB cache and will be released in 2008.

The INTEL Hapertown will be a multi-die CPU with 8 cores and will also be released in 2008.

INTEL's inefficient design requires at least 2 to 4 MB of cache for each CPU core, the doubling of cache negates any gain INTEL made by shrinking to 65nm. In contrast, AMD's Opteron only needs 1MB. We see almost linear performance scaling in 8 core Opteron systems.

The best desktop PC!

This desktop is amazing, for $799 (after $100 rebate), you get (try build it yourself using lowest priced components, you can't get down to this price)

HP Pavilion Desktop PC (A1250N)HP A1250N
• AMD Athlon™ 64 X2 3800+ dual core 64 bit processor, which is faster than Pentium D 840 on some tests
• 250GB 7200RPM hard drive
• 1GB of DDR SDRAM
• Dual optical drives
• LightScribe-enabled
• 9-in-1 memory card reader

No wonder people give it a score of 4.9/5.0 at http://tinyurl.com/7pfas
(I saw one customer complained that the power supply is only 300 watts, are people getting used to Pentium 4's 600 watts?? This PC needs only 300 watt power supply, because the CPU is energy efficient).

Hot chips are harmful to web hosting

A friend of mine is looking for a dedicated server for his web site, I suggest him to choose AMD64 based solution, performance is one reason, heat is another. I told him AMD CPUs produce far less heat and use far less electricity.

My friend said: "But the data center pays for electricity and the electricity is included in the monthly fee".

Well, he was only partially right. Although from a cost perspective, power consumption does not seem to be a direct issue, but heat harms system reliability. Larger power consumption means larger current, and larger heat dissipation on all system components, such as capacitors. Long term exposure will reduce the mean time to failure of capacitors and hard drives and even fans. For example, at higher temperatures, the lubricant of the mechanical parts will evaporate (molecules escape into the air), and the parts will eventually burn out due to friction. When that happens, the host will replace the hardware for free, but there will be down time and data loss for sure. You won't experience problems initially, but after a year or so, you may see the effect.

A web host takes their total cost and averages it to set a price. Choosing reliable and low power solutions help reduce the total cost of a web host, and the benefit can be eventually passed back to the customers.

Friday, December 02, 2005

INTEL is mad?

Rahul Sood, president and CTO of VoodooPC, wrote some interesting stories on how INTEL played his company like a toy, and how INTEL set quotas for him to reduce the use of AMD chips. I think AMD lawyers should get his testimony for the lawsuit against INTEL. He is a tech guru with a sense of justice, jury will definitely like such a guy.

His newest comments made to http://www.internetnews.com was equally sharp:

"I think Intel is mad and will come out with some compelling products. The Pentium M is fantastic and Yonah will be really good, but it would be a mistake for them to think AMD is going to sit back and wait, because what they're working on for next year is phenomenal."

"I don't see AMD wanting as much volume as Intel. I think they're positioning as a premium alternative to Intel, and, in many ways, they already are."

Not so quick to jump to conclusions, Mr. Sood, read this first before believing in any of INTEL's furture stuff. Numbers don't lie, perceptions can be deceiving.

Thursday, December 01, 2005

AMD vs INTEL lawsuit should be tried in "People's Court"

Representatives from various media and press sent a letetr to the judge presiding over the AMD vs INTEL anti trust lawsuit, urging the judge not to grant any requests to seal the trial information. They argue that the public deserves to know what really happened, as INTEL's actions afffected almost everyone.

It makes perfect sense. We paid INTEL good money for its CPUs, we have the right to know how INTEL abused its monopoly power and deprived us the right of choice if INTEL indeed did it as AMD alleged. The case is not just for AMD but for the advancement of technology and civilization. If INTEL did use dirty tricks to push its chips to consumers, it should be severely punished, so an example can be set to deter similar acts by others in the future. Any company that willfully thwarted the advance of civilization must be punished to the maximum allowed by the law.

SUN needs AMD's help

Judge from this article, SUN is probably quite ahead in multi-core design, its UltraSparc T1 has 8 cores and consumes less than 80 watts. Quite amazing. But SUN should not be too cocky.

The problem is, each of the Niagra core runs only at 1.2GHZ, so it's like a 9.6GHZ chip in some sense. AMD is not too far behind. AMD can probably beat UltraSparc T1 with quad-cores. First, AMD needs at least two memory controller per quad-core CPU, I believe socket F's 1207 pins are partly for an extra memory controller, then AMD has to control the power consumption. It is not a major problem, the 2.2GHZ Turion MT-40 has a maximum thermal output of 25 watts. Packing 4 such cores, it's within 100 watts. AMD may also get some heat reduction from its 65nm process. So AMD is very close to introduce quad-core at 2.2GHZ.

SUN is disadvantaged because it does not have its own FAB and has no experience in manufacturing, it is at the mercy of others such as Texas Intruments when it comes to clock speed. AMD is the master in terms of semiconductor process, and it knows how to tweak things and make faster chips. Maybe SUN can contract its UltraSparc T1 production to AMD. AMD's DSLSSOI process can increase frequency by 30%, that will make the Niagra running at 1.6GHZ, a much more competitive situation against Power5.

INTEL's chest beating shows lack of confidence

FAB Tech analysed INTEL CEO Paul Otellini's recent declaration that: "Intel is committed to widening its lead in advanced semiconductor manufacturing,", that "Our manufacturing network is a strategic asset of unmatched scope and scale that gives Intel the ability to provide customers with leading-edge products inhigh volume. Today's announcement of a second 45nm high volume factory reaffirms that Intel platforms will contain the most advanced and innovative technology in the world for years to come."

On the surface, Paul is upbeat. Under the surface, we see signs of weakness.

INTEL is not fundamentally a manufacturing company. CPUs win by design and architecture. Right now, AMD64's Direct Connect Architecture is years ahead, INTEL won't have it until 2009. AMD is also working on Direct Connect Architecture 2.0, which will take it to the next high ground in 2007. INTEL, on the other hand, is basically going back to Pentium III.

Both INTEL and AMD buy the same kind of FAB equipment from companies like Applied Materials, KLA-Tencor, etc. INTEL has more money to spend on more cutting edge and unproven FAB technology, AMD plays safe. But AMD is usually only 6 months late in buying the newest tools.

The difference between INTEL and AMD is the silicon wafer and the process. INTEL uses bulk silion wafer and does strained silicon (SS). AMD uses expensive SOI wafer and does Dual Stress Linear Strained Silicon On Insulator (DSLSSOI), a technology jointly developed by AMD and IBM to improve performance and reduce heat. In terms of software, AMD's APM 2.0 is years ahead of the industry, it can control the process down to single transistor on each wafer. That's why AMD's yield is so high and why AMD can competete against INTEL's five 300mm FABs with only one 200mm FAB (FAB30). Many big FABs are starting to license AMD's APM software.

Moving to 65nm then 45nm will reduce feature length, however, due to INTEL's outdated FSB architecture, INTEL has to keep doubling cache sizes to get reasonable performance, the result is that INTEL did not get any die size reduction when going to 90nm or 65nm. In contrast, AMD CPUs, with Direct Connect Architecture, do not need big caches, consequently, AMD CPUs have reduced die size when moving to 90nm, allowing AMD to make more CPUs off the same wafer.