Tuesday, June 27, 2006

AMD sells chips, Intel sells chips business

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

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

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

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

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

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

48 Comments:

Anonymous Anonymous said...

"Intel's Woodcrest is for low end 2P market only."

It may be lower end than 4p, but 2p is about 85% of the server market and with Intel poised to capture 80% of that 2p market, AMD is going to have its hands full very soon. Couple that with huge cash outlays for new fabs and paying chartered, I wouldn't be surprised if AMD slips back into its customary role of unprofitability. As a matter of fact, I predict AMD will be bankrupt in 6 quarters!

11:16 AM, June 27, 2006  
Blogger Sharikou, Ph. D. said...

It may be lower end than 4p, but 2p is about 85% of the server market and with Intel poised to capture 80% of that 2p market.

Actually, the server market is rapidly moving to 4P. Google uses 4P in its data centers. It's always better to get larger servers as long as performance scales with price. You get 2X the performance at less than 2x price by going Opteron.

couple that with huge cash outlays for new fabs and paying chartered, I wouldn't be surprised if AMD slips back into its customary role of unprofitability.

AMD's average cost per finished CPU is only $40. Remember, AMD has only 10,000 employees, and Intel has 102,000 (Intel dumped 1400 today). AMD can sell millions of CPUs to DELL at rock bottom prices while still making money. With 60% lower ASP and market share loss, Intel is bound to lose and lose. It's simple math.

11:34 AM, June 27, 2006  
Anonymous Anonymous said...

I wonder would you phrase the title on AMD selling its MIPS business?

11:35 AM, June 27, 2006  
Anonymous Anonymous said...

amd just sold their alchemy processor technology they bought in 2002. it seems amd is in deep trouble to do so and needs desperatley more cache for the outstanding price war with intel. i predict amd will be bankrupt in 5 quarter!

11:36 AM, June 27, 2006  
Anonymous Anonymous said...

if Intel Chip is dirt chip, how can AMD fight if itself is not dirt chip? if even you claim AMD wouldn't, how Dell is going to be able to sell AMD machine at higher price than Intel Machine?

be reasonable and use your brain. while you try to bash intel and bring glory to AMD, what you say here, reflecting immature thinking, short sighted view, would itself damaged your lovely AMD brand (assuming you have such influence, which of course, you don't)

11:40 AM, June 27, 2006  
Blogger Sharikou, Ph. D. said...

I wonder would you phrase the title on AMD selling its MIPS business?

AMD has plenty of cash and is busy selling its Opterons and AM2s at very pleasant price points. AMD transfered Alchemy to RAZA as a strategic investment in that startup.

if Intel Chip is dirt chip, how can AMD fight if itself is not dirt chip?

Go to newegg.com or even ebay.com, you see how much Intel and AMD chips are being sold. Intel chips are so dirt chip, but few want them. AMD's X2 3800+ is still selling at $297 and are selling like hotcakes.

Everyone knows that an X2 3800+ frags a Pentium 965 XE.

11:49 AM, June 27, 2006  
Anonymous Anonymous said...

if Intel Chip is dirt chip, how can AMD fight if itself is not dirt chip?

the right people will know, that amd chips are superior to intel chips and will pay whatever price amd charges. so amd will make huge profits, while intel will be bancrupt within 4 quarters from now on!

11:52 AM, June 27, 2006  
Anonymous Anonymous said...

Everyone knows that an X2 3800+ frags a Pentium 965 XE.


even an amd athlon xp can frag this intel dirt!

12:08 PM, June 27, 2006  
Anonymous Anonymous said...

Amd chips are less expensive to produce than Intel's, so AMD ASP might be lower and still generate profit.

Ah Sharikou, what about the news that Intel will do Inverse Hyperthreading too?? ahahahh They are always following AMD, that' the true last five years trend. Anyway, I wonder HOW they could do it without IMC and with that obsolete FSB stuff (and lot of shared cache too!).

12:09 PM, June 27, 2006  
Anonymous Anonymous said...

This article [link below] claims that an Athlon FX model 62 CPU has an improvement of only 2 to 3 percent in 64-bit mode. It contradicts your previous statements of much higher benefits for the K8 architecture in 64-bit mode. The artcile is also quick to note that in the desktop market the Conroe processors offer a significant performance advantage over Athlon 64 offerings, they fail to mention the power consumption and overclocking advantages, however. I agree to the overwhelming evidence of Opteron superiority in the P4 server market, still..
As the Conroe is ramping up and will hit the market several months earlyer than Athlon 64 at 65 nm process, Intel will have a superior product for all that time. Now, talking about a price war.. isn't it a real blessing for AMD and a sure necessity at the same time. It looks like this price war is specifically engineered by AMD and just shakin' the poor Intel for a ride along the way. Who's the big brother now?

http://www.hkepc.com/bbs/itnews.php?
tid=621654&starttime=0&endtime=0

12:21 PM, June 27, 2006  
Blogger Sharikou, Ph. D. said...

Ah Sharikou, what about the news that Intel will do Inverse Hyperthreading too??

That's why you saw those pretty high superpi numbers with Conroe.

12:28 PM, June 27, 2006  
Blogger Sharikou, Ph. D. said...

Now, talking about a price war.. isn't it a real blessing for AMD and a sure necessity at the same time. It looks like this price war is specifically engineered by AMD and just shakin' the poor Intel for a ride along the way.

The price war will devastate Intel. Intel's ASP was $150, AMD was below $100. Intel needs to cut revenue by 40% to compete against AMD..

12:31 PM, June 27, 2006  
Anonymous Anonymous said...

Hey Sharikou, good title you have here. It made my first laugh for the day!

12:31 PM, June 27, 2006  
Anonymous Anonymous said...

"While Intel generated cash by dumping businesses, AMD just sold 8064 Opteron 8xx CPUs to be used in 2016 4P Opteron servers. That's some nice profit there."

That's really no big deal. Even if they sell $1000 a chip, it's only $8M, less than 1% of their quarterly revenue and 3% net income. Do they get such deal every quarter? Probably not even every year.

12:41 PM, June 27, 2006  
Anonymous Anonymous said...

Ah Sharikou, what about the news that Intel will do Inverse Hyperthreading too??

That's why you saw those pretty high superpi numbers with Conroe.


so intel has already implemented inverse hyperthreading in woodcrest?

12:48 PM, June 27, 2006  
Anonymous Anonymous said...

"This article [link below] claims that an Athlon FX model 62 CPU has an improvement of only 2 to 3 percent in 64-bit mode."

The link you supplied only had claims but no evidence. I wouldn't give a shxt to that. Also I won't be surprised such 'internal' info comes from someone with excessive P-XE 965 stock.

FYI, different apps have different performance boosts under 64-bit environemnt. How these apps are optimized for 64-bit is also a big factor.

12:50 PM, June 27, 2006  
Blogger Sharikou, Ph. D. said...

Do they get such deal every quarter? Probably not even every year.

Wrong. AMD didn't issue a PR for that order. It was a piece of cake for AMD. Google builds 30000 4P Opteron servers per month.

12:58 PM, June 27, 2006  
Anonymous Anonymous said...

"2p is about 85%" gezz give me a break this is Intel spin for me to buy a 2P wishing that for once I could buy something from Intel that would see me thru my companys growth nooo not again sorry thats why 4p is now the new market, I hate buying new servers every damm year because of lack of performance do it once, do right,
4P is right end of story

1:00 PM, June 27, 2006  
Anonymous Anonymous said...

FYI, different apps have different performance boosts under 64-bit environemnt. How these apps are optimized for 64-bit is also a big factor.

even intel has support for 64bits for over a year now in its desktop processors. what's the difference of intel and amd technology? does intel only use 63bits?

1:08 PM, June 27, 2006  
Anonymous Anonymous said...

no, Intel can run 64 bits, but it runs it slower. AMD gets a performance boost with 64 bit programs, but Intel doesn't.

3:19 PM, June 27, 2006  
Anonymous Anonymous said...

no, Intel can run 64 bits, but it runs it slower. AMD gets a performance boost with 64 bit programs, but Intel doesn't.

please answer the question. why is intel slower than amd in 64bits? isn't 64bits == 64bits?

3:32 PM, June 27, 2006  
Anonymous Anonymous said...

"even intel has support for 64bits for over a year now in its desktop processors. what's the difference of intel and amd technology? does intel only use 63bits?"

What the hack are you talking about? The problem was never that some chip has less number of bits, but what apps were optimized for it, and how much benefit it could get.

Just google around and you'll find big difference on different CPU running different apps.

BTW, please don't try to make quick-wit (63bits???) comments by showing yourself a tech idiot. That's really sad and unnecessary.

3:32 PM, June 27, 2006  
Anonymous Anonymous said...

"Google builds 30000 4P Opteron servers per month."

Got a link for that? I challenge you to prove that statement.

3:56 PM, June 27, 2006  
Anonymous Anonymous said...

I hate buying new servers every damm year because of lack of performance do it once, do right,
4P is right end of story


i totaly agree with you. therefor we don't buy 4p x86 intel/amd shitboxes, but go ibm system z9 mainframe. don't go for second best 64bit computing.

4:03 PM, June 27, 2006  
Anonymous Anonymous said...

"2p is about 85%" gezz give me a break this is Intel spin for me to buy a 2P wishing that for once I could buy something from Intel that would see me thru my companys growth nooo not again sorry thats why 4p is now the new market, I hate buying new servers every damm year because of lack of performance do it once, do right,
4P is right end of story


next year will bring 4 cores per cpu. i see no reason why the 4 cpu market should increase dramatically.

4:08 PM, June 27, 2006  
Blogger Sharikou, Ph. D. said...

"Google builds 30000 4P Opteron servers per month."

Got a link for that? I challenge you to prove that statement.


I updated the main article, check the link there.

5:03 PM, June 27, 2006  
Anonymous Anonymous said...

Just wanted to point out two things:

"Actually, the server market is rapidly moving to 4P. Google uses 4P in its data centers. It's always better to get larger servers as long as performance scales with price. You get 2X the performance at less than 2x price by going Opteron."

Google is dumping AMD and moving back to Intel so that might not be a very good example.

"That's why you saw those pretty high superpi numbers with Conroe. "

The Super Pi scores have been run with 1 core disabled and the results are unchanged so inverse threading has not been enabled previously (it was only introduced in the most recent Bad-Axe BIOS update) and has nothing to do with the Conroe performance we've seen to date.

5:09 PM, June 27, 2006  
Blogger Sharikou, Ph. D. said...

Google is dumping AMD and moving back to Intel so that might not be a very good example.

Dude, Charlie at INQ was teasing you Intel fanboys.

You sing on this:

http://www.theinquirer.net/?article=32617
'
Then Charlie wrote this:

http://www.theinquirer.net/?article=32656

5:24 PM, June 27, 2006  
Anonymous Anonymous said...

why did intel sell its communication chip business? because it doesn't need it anymore. the concept is pretty much the same when AMD sold their alchemy chip division to RAZA.

that's weird, you claimed intel sold its Xscale division because it needs money, but you didn't say anything about AMD's spin off.

double standards.

6:13 PM, June 27, 2006  
Anonymous Anonymous said...

"why did intel sell its communication chip business?"

Because it's losing hundreds of million dollars every quarter, and Intel can't afford to keep it anymore.

If the comm biz made money, Intel will need it, instead of selling it.

8:16 PM, June 27, 2006  
Anonymous Anonymous said...

The link merely says that google buys 30,000 servers a month. Doesn't say anything about them being all AMD. Nice try. But this is what I expect from Shari-fraud.

8:52 PM, June 27, 2006  
Anonymous Anonymous said...

I think all of you should read this...

TG Daily

This seems to clarify what is really happening, not speculating.

8:58 PM, June 27, 2006  
Anonymous Anonymous said...

Despite what some of the headlines said this morning, Marvell didn't purchase "the XScale unit" from Intel today.

Marvell today purchased Intel's license to manufacture XScale technology, specifically for cellular and handheld devices, granted to Intel years ago by ARM Holdings. It's a simple transfer of rights, with a bonus attached that Marvell gets Intel's people who build it for them - up to 1,400 employees, whose paychecks will now be authorized by a building fewer than two miles away from the previous signer.

9:51 PM, June 27, 2006  
Blogger Sharikou, Ph. D. said...

The link merely says that google buys 30,000 servers a month. Doesn't say anything about them being all AMD. Nice try. But this is what I expect from Shari-fraud.

Don't be in denial. Google buys 30,000 four way servers a month. We all know they are opterons. They can't be Xeon Paxville, which is 170 watts each. Try do the math, if they were Intel Xeon Paxville, Google would need 30000*170*4 = 20.4 Mega watt extra every month.

1:16 AM, June 28, 2006  
Blogger Unknown said...

Actually there is only 1 sane reason Google is buying Intel chips for their servers. Intel offered to pay the difference in cost of electricity usage (compared to Opterons) monthly if Google pick up their chips.

Why it is as brilliant as stacking 2 cores on top of each other!!!

2:00 AM, June 28, 2006  
Anonymous Anonymous said...

"They can't be Xeon Paxville, which is 170 watts each. Try do the math, if they were Intel Xeon Paxville, Google would need 30000*170*4 = 20.4 Mega watt extra every month. "

Where on earth did u get 3000 a month? and Google doesn't use paxville, but woodcrest instead.

6:49 AM, June 28, 2006  
Anonymous Anonymous said...

You are really losing it, numbnuts.

30000 servers a month, huh. At..$200? Read your link a little more closely. Does it withstand any kind of application of logic to contend that Google is building _1000_ servers a day, and they're only paying $200 for them?

You are truly not a smart person.

1:11 PM, June 28, 2006  
Blogger Sharikou, Ph. D. said...

30000 servers a month, huh. At..$200?

$200 is for the motherboard. Google is buying MB directly from ODMs. 1000 servers a day is no surprise, google is expanding like crazy everywhere. Hardware is cheap. It has been reported that Google is the 4th largest server maker in the US...

1:20 PM, June 28, 2006  
Anonymous Anonymous said...

I find it humurous how people who have little relevant things to say, try to compensate this with personal attacks or flaming.

Sharikou, I do not buy everything you say, although I think you are correct on many articles that you write. Atleast I respect that you try to back up your claims with some actual facts/proof/logic, unlike some of these people here who like to call you "numbnuts".

You don't need to result to name calling if you don't agree with someone, there are more civilized ways to get your point accross. But I guess we've hit the main problem here: there is no point, atleast not a relevant one.

BTW the core multiplexing feature on the intel motherboard that was recently discovered does not do reverse hyperthreading so it seems, although we won't know until it's really confirmed by Intel I guess.

“We will support the disabling of one core in BIOS in a future BIOS rev.,” said Daniel Snyder, a spokesman for Intel Corp., when asked about the background of the technology. This means that once one core is disabled, the whole 2MB or 4MB cache reservoir on microprocessors that have shared level-two (L2) cache will be usable by one processing engine, which should boost performance in applications that cannot take advantage of two executing cores.

Either the name of the "Core multiplexion" option is deceiving about what it truly does, or it's not this feature after all what is talked about in the above quote. Anyway, you can read more about it at http://www.xbitlabs.com/news/cpu/display/20060627095448.html

2:48 PM, June 28, 2006  
Anonymous Anonymous said...

Sharikou,

Im one of AMD supporter and has been using AMD since AMD K6-2 and I Would love to see AMD capture +- 50% of the CPU market. However i dont want to see Intel die the way u want/predict them to be because then nobody will keep AMD honest.

My question is, How do u see its going to benefit us user if Intel die?

*Forgive me for my bad english.

-MyBoy-

3:06 PM, June 28, 2006  
Anonymous Anonymous said...

"BTW the core multiplexing feature on the intel motherboard that was recently discovered does not do reverse hyperthreading so it seems, although we won't know until it's really confirmed by Intel I guess."

Indeed! But the rumor about reverse hyperthreading is so prevalent, that both AMD and Intel side of fans want to have a say about it. At some point this is like fans arguing which soccer team plays the best. Well, all of them can't fly, that I know...

5:19 PM, June 28, 2006  
Anonymous Anonymous said...

"Either the name of the "Core multiplexion" option is deceiving about what it truly does, or it's not this feature after all what is talked about in the above quote."

Just to be sure, there were two settings in the bios, one was core multiplexing and the other was to disable a core.

It is not deceiving, you may have just misread the forums article Here
Also here is a link showing the bios and the two options.

6:19 PM, June 28, 2006  
Anonymous Anonymous said...

"Just to be sure, there were two settings in the bios, one was core multiplexing and the other was to disable a core."

Indeed! That is such a wonderful Blue Crystal! Why would you believe that the "multiplexed" dual cores would perform better? Multiplexing is different from cooperating, and very, very different from CMT (chip multi-threading). That forum and that link on digit-life are totally clueless.

The "Core Multiplexing" could be a way to save power consumption, that only one core is active while the other is put to sleep. A new thread could be assigned (multiplexed) to either core in order to avoid L1 cache thrashing. That's my guess, anyway. It's definitely not reverse hyperthreading though.

7:56 AM, June 29, 2006  
Anonymous Anonymous said...

only one core is active while the other is put to sleep

I have read that Turion X2 can already disable one core, if necessary, in power save mode.

10:46 AM, June 29, 2006  
Anonymous Anonymous said...

"The "Core Multiplexing" could be a way to save power consumption, that only one core is active while the other is put to sleep."

Hey Edward, everytime I post a reply to one of your comments we end up on the wrong foot.

So to keep it simple I would like you to look here
just so we are on the same page.

Thanks.
Thanks.

1:46 PM, June 30, 2006  
Anonymous Anonymous said...

"Hey Edward, everytime I post a reply to one of your comments we end up on the wrong foot."

Maybe that's because you didn't read well?

See my next sentence (right after what you quoted) in my previous comment that says A new thread could be assigned (multiplexed) to either core in order to avoid L1 cache thrashing. That IS multiplexing the cores (ie. demultiplexing the threads). Don't need to bring up a wiki page on 'multiplexing' like anyone's technical challenged here.

While there is no definite knowing what core multiplexing does, the 'consensus' is that it basically halt one core for single-threaded apps, and by this way double the L2 cache for the active core.

5:56 AM, July 02, 2006  
Anonymous Anonymous said...

Edward it is really hard to discuss anything with you. In hein sight, the link might have been inappropriate, and a little misleading to my point.

From what I have read it would seem to take a single threaded application and split it between the two cores, then combine the two threads afterwards back into a single thread.

This would then speed up the single application, right?

Isn't that what AMD is supposed to do?

As I have said before this is not my field.

Thanks.

4:15 PM, July 02, 2006  
Anonymous Anonymous said...

"Don't be in denial. Google buys 30,000 four way servers a month. We all know they are opterons"

I read the updated original article. I think you extrapolated the opteron part. The article does not mention anything about the processor mix. Way to go "journalist"

5:28 PM, July 07, 2006  

Post a Comment

<< Home