The AMD Execution Thread [2007 - 2017]

[Arun]

Done and done. Anyway, this discussion is horrifyingly low-quality IMO, so I'll try to change that a bit by pointing out some much needed facts. Here are the PDFs with information about NVIDIA and AMD's Q307 financials:
http://www.amd.com/us-en/assets/content_type/DownloadableAssets/Q307Financials.pdf
http://media.corporate-ir.net/media_files/irol/11/116466/10Q_Q308.pdf

On page 3 of the AMD PDF, and page 26 of the NVIDIA PDF, you'll find the per-business unit data. Operating expenses can be obtained by subtracting operating income to (revenue * gross margins). Based on a variety of data, I will estimate NVIDIA's GPU (non-Quadro) margins at 42% and AMD's at 36%. This is hopefully not too far off, but either way it doesn't change the arguement.

This results in estimated GPU operating expenses of $94M for AMD and $90M for NVIDIA. However, it should also be noted that the definitions fo the GPU business units for the two companies are quite different. AMD says it "includes graphics, video and multimedia products developed for use in desktop and notebook computers, including home media PCs, professional workstations and servers."

That means it includes FireGL, FireMV, TV Wonder, Theater and embedded graphics sales. Quadro is a separate business unit for NV, they don't have an equivalent to TV Wonder/Theater and I believe embedded (although really small) is included in the 'All Others' category, but I could be wrong there. It's also hard to guess what's the proportion of R&D compared to marketing/administrative expenses, but I suspect the difference isn't huge.

So what does this tell us? That despite all odds and despite misreading and/or misunderstanding a variety of data, LordEC911's claim regarding GPU R&D likely isn't too far off. Conclusion: You should all try to be a bit less confident next time around, kthxbye!

P.S.: As far as I can tell (because ATI's business units were grouped quite differently pre-acquisition), NVIDIA's GPU operating expenses have grown much faster than AMD's and they were probably outspent by ATI in the G7x vs R5xx generation, for example. If someone really wanted me to, I could try to extrapolate some more precise numbers there, but it's probably not worth the time/effort to be honest.

 

[silent_guy]

Done and done. Anyway, this discussion is horrifyingly low-quality IMO, ...

Come on, once in a while you have to allow us to have some good old fun. You know, like Dumb and Dumber.

 

[kyetech]

On a side note heres what I think about AMDs future.

I think by the time we hit 22nm process the ATI purchase will become the smartest thing they ever did. You may think this is stupid but I think AMD / INTEL could put Nvidia out of business within 8 years.

Nvidia are flying right now, but unless they get to grips with producing x86 capable SOCs they will eventually lose the entire mid / entry level market to Intel and AMD. When 22nm allows 4 billion trannies only the highest end are going to need discrete GPUs. Everyone else will be happy with one of the above monsters as a single integrated do all chip. I mean there is a serious amount of computation 4 billion trannies can do. And I aint talking about cross dressing type ! ;-)

Ive not been through this thread so have no idea whats been said. Perhaps Ill just get laughed and flamed. But just for fun, heres my predication: hehe.

in 8 years from now Nvidia will be out of business and bought up in bits by the other two. AMD will be booming and have 35+% market share and rising.

 

[IdaGno]

in 8 years from now Nvidia will be out of business and bought up in bits by the other two. AMD will be booming and have 35+% market share and rising.

donwanna derail the thread, but...assuming you are correct re nVidia's long-term viability, or perhaps more correctly, lack thereof, & assuming further that they are not bought by Intel & that they are restrained from buying AMD - the x86 license issue - I wonder if nVidia might find a home for itself in purchasing VIA?

 

[Arun]

I'm at a loss as to how a company with $2B in cash would have a problem acquiring a technology that costs less than $200M to develop, and probably not more than 1.5-2.5x that to acquire. I guess if you were Doug Freedman and you had no clue about the industry and thought AMD was the only option, then that'd be excusable - but you're not Doug Freedman, right?

I'd honestly be a lot more worried about NVIDIA's recent MCP suckiness than x86. In a platform company, what matters is your weakest link... And MCP can't seem to get a chip to market in less than 6-12 months late, nor do they seem able to create roadmaps that make any sense whatsoever. Sigh...

 

[MulciberXP]

...

 

[kyetech]

I see, So if x86 knowledge only costs 200m to develop and therefore aquire from scratch, how much would it cost a company to develop GPU knowledge from scratch?

 

[ShaidarHaran]

I see, So if x86 knowledge only costs 200m to develop and therefore aquire from scratch, how much would it cost a company to develop GPU knowledge from scratch?

Quite a bit more, I'd imagine. x86 is a known-quantity. GPU architecture is unique to each vendor and even across different families from the same vendor.

 

[Arun]

I see, So if x86 knowledge only costs 200m to develop and therefore aquire from scratch, how much would it cost a company to develop GPU knowledge from scratch?

I think it's worth pointing out the 'assuming you can acquire AAA engineers with experience in the field' assumption I didn't mention. Stexar was a step in that direction but substantially too small, so NVIDIA's x86 strategy today is mostly tied around potential acquisition targets.

As for GPU architectures, the primary difference is that perf/cost$ matters a lot more there than for x86, because die sizes are inherently higher. Certainly a project with the perf/mm² and perf/watt of Conroe would cost a lot more than $200M to make from scratch and would take many years (if you could even find the right talent), but the kind of x86 core you need for a single-chip CPU+GPU+MCP SoC is quite different from that, especially due to the commoditization of the market.

Anyway, let's stop discussing that in this thread - the only reason I was talking of it here is because I wanted to make very clear that NVIDIA/Jen-Hsun won't acquire AMD. They just won't and it doesn't make any sense whatsoever when you look at the full picture.

 

[MTd2]

AMD fell almost 6.5%. Some say it was caused by the release of Intel's ultra low power chips.

 

[3dilettante]

AMD has demonstrated Shanghai samples and released a few details on its 45nm process.

http://www.amd.com/us-en/assets/content_type/DownloadableAssets/AMD_45nm_Press_Presentation.pdf

For those curious as to how AMD went about increasing its L3 capacity from 2 MiB to 6, look no further.

Whereas Barcelona was a pretty square die with the L3 in the corners, AMD went rectangular and slathered extra L3 on one side.
Montreal appears to have the same amount per-die, so it seems the trend will continue into 2009 as far as the 10h family is concerned.

 

[ShaidarHaran]

Nice find. Good explanation of how immersion lithography works as well. Interesting that they're creating air pockets in the dielectric to achieve lower k values. I wonder if anyone else is doing that (other than IBM, of course).

 

[3dilettante]

I'm curious about how the pores affect mechanical strength and the thermal cycling behavior of the chips.

The material wouldn't be stronger than standard materials, and it would be compounded by the already brittle SOI wafer material.

The bullet point about improved yield was likely put there because observers already know that yields would suffer without special measures.


The lower K value might help AMD somewhat for Shanghai, if Barcelona suffered from issues stemming from its old pipeline hitting signaling or timing issues.
The pressure on a design not optimized for reduced interconnect scaling should be lessened if the capacitance of the interconnect layer improves.

 

[Farhan]

AMD has demonstrated Shanghai samples and released a few details on its 45nm process.

http://www.amd.com/us-en/assets/content_type/DownloadableAssets/AMD_45nm_Press_Presentation.pdf

For those curious as to how AMD went about increasing its L3 capacity from 2 MiB to 6, look no further.

Whereas Barcelona was a pretty square die with the L3 in the corners, AMD went rectangular and slathered extra L3 on one side.
Montreal appears to have the same amount per-die, so it seems the trend will continue into 2009 as far as the 10h family is concerned.

Nice. Looks like their highK/metal gate process is gate first vs gate last for intel.

 

[MTd2]

Either AMD is completely dishonest with people or there is a grand greate huge secret plan:


"At the breakfast in Colonie on Tuesday, AMD officials outlined some significant changes. Plans are now calling for three factories rather than four, and a larger workforce than originally estimated, about 1,400 new jobs instead of 1,200. But again, still no word on when.

"In terms of a commitment date as to when we will firmly and finally answer the question everyone is asking, we still can't do that right now," said AMD Wafer Manufacturing Strategies Director Terry Caudell"

http://capitalnews9.com/content/headlines/111746/amd-plans-moving-forward/Default.aspx

"Advanced Micro Devices Inc. has had talks with Hudson Valley Community College about developing a curriculum tailored for graduating the types of workers the chip maker will need when--and if--it builds a chip plant in Malta."

http://albany.bizjournals.com/albany/stories/2008/03/03/daily21.html?jst=b_ln_hl

The local news are showing this every day. They are messing up with people that have nothing to do with AMD. I hope AMD, despite its huge financial losses, has some great triumph card. If this construction plan fails, it will just be another source of really bad publicity for AMD.

 

[IdaGno]

donwanna derail the thread, but...assuming you are correct re nVidia's long-term viability, or perhaps more correctly, lack thereof, & assuming further that they are not bought by Intel & that they are restrained from buying AMD - the x86 license issue - I wonder if nVidia might find a home for itself in purchasing VIA?

meanwhile...

Monica Chen, Taipei; Joseph Tsai, DIGITIMES [Tuesday 18 March 2008]

Nvidia reportedly was in talks with VIA Technologies about a possible acquisition, but no deal was made due to the high price named by VIA, according to sources at motherboard makers.

http://www.digitimes.com/mobos/a20080318PD212.html

 

[MTd2]

The count down to Ruiz stepping down AMD. Remember this article?

"If Meyer is to take over, the change will likely occur when Ruiz's latest yearly contract expires on April 26, 2008."

http://techreport.com/discussions.x/13632

 

[MTd2]

first-quarter net loss actually narrowed to $358 million, or 59 cents a share, from $611.0 million, or $1.11 a share, a year ago.

http://www.reuters.com/article/rbssTechMediaTelecomNews/idUSWNAS830220080417

 

[dizietsma]

first-quarter net loss actually narrowed to $358 million, or 59 cents a share, from $611.0 million, or $1.11 a share, a year ago.

http://www.reuters.com/article/rbssTechMediaTelecomNews/idUSWNAS830220080417

http://biz.yahoo.com/ap/080417/earns_amd.html?.v=4

"sales across all business segments were lower than it had expected"

But

"AMD's revenue rose 22 percent to $1.51 billion from $1.23 billion" ( year on year)

So they increases sales by 22% but they expected more? Considering their product line up at the moment compared to their two biggest competitors expecting more than 22% seems to me to be wishful thinking!

A 22% increase in sales should mean lots of profit, but they are obviously suffering from bottom selling or else their debt repayments are too high already. They only paid $50m for the aquisition of ATI this quarter (only! how long are those one off costs going to roll on for ? ) so the money must be bleeding from somewhere.

Intel has a price cut soon as well.

They are going to have to sell off some of the family silver soon. Work force cuts are not enough. At this rate they will not be able to pay off the interest on all those bonds they sold unless they sell one of their two FABS.

Asset lite is another way of saying being butt raped FAB wise I think

 

[Arun]

I'll write a news piece to comment on this later today, no time to do it now - there are few interesting tidbits in the INTC and AMD earnings & their associated CC, nothing revolutionary though.