[zorg]

The value of AMD is their GPU tech. Their GCN design is 2 or 3 years ahead the competition. Briefly: it's Larrabee done right.

[silent_guy]

Originally Posted by zorg

The value of AMD is their GPU tech. Their GCN design is 2 or 3 years ahead the competition. Briefly: it's Larrabee done right.

If we assume, for the sake of argument, that all of this is true, then it's no wonder that they're only valued as low as they are. Because not a lot of people seem to care about the compute abilities of GPUs, especially those of AMD.

[Alexko]

Originally Posted by silent_guy

If we assume, for the sake of argument, that all of this is true, then it's no wonder that they're only valued as low as they are. Because not a lot of people seem to care about the compute abilities of GPUs, especially those of AMD.

To be fair, GCN-based FirePros were only released a couple of days ago. Earlier VLIW cards were much less compute-friendly, so this is kind of new for AMD.

[dkanter]

Actually, I'd say the largest intangible asset that AMD has is the ability to effectively design and validate x86 CPUs. There is only one other company in existence with that ability.

And no, ARM will never buy AMD. ARM makes IP. AMD makes chips. ARM doesn't wish to compete with its customers. Samsung would make much more sense.

DK

[Kaotik]

Originally Posted by dkanter

Actually, I'd say the largest intangible asset that AMD has is the ability to effectively design and validate x86 CPUs. There is only one other company in existence with that ability.

And what's the 3rd one that can do high performing GPUs besides AMD and nVidia?

[3dilettante]

Did the most recent settlement give clarity on how readily the cross-licensing agreement can carry over in the event of a purchase? The last description indicated it was no longer an automatic death sentence, but it would be a multibillion dollar maybe.

I suppose the whole package involved in producing leading-edge (or mostly so) x86 would probably not survive in the hypothetical case that AMD goes under and various players come in to buy up what's left.
Patent pools can be bought when picking a company's bones clean, but there's not much use to them if the people making use of them aren't there.

[imaxx]

There is something I miss. If all this was true and AMD would be at loss, why they did a rescission with their (ex) foundry?

Even now that it is done, it still continues to make their chips, so they had no reason to pay that horrid amount of money for nothing, since they can already do GPUs with TMSC and GF is improving.

Unless their CEO is drunk 24/7, it means they got a incoming good reason (contract) to do that.

[aaronspink]

Originally Posted by Kaotik

And what's the 3rd one that can do high performing GPUs besides AMD and nVidia?

You are assuming that is a relevant market outside of gamers. All data points to there being sufficient number of manufacturers of graphics IP in the tablet and smart phone areas that ATI/Nvidia bring relatively low economic value from that area.

[Alexko]

Originally Posted by aaronspink

You are assuming that is a relevant market outside of gamers. All data points to there being sufficient number of manufacturers of graphics IP in the tablet and smart phone areas that ATI/Nvidia bring relatively low economic value from that area.

Actually, I can't really see anything that NVIDIA brings to the table that ARM, IMG or Qualcomm don't.

[Ryan Smith]

Originally Posted by Alexko

To be fair, GCN-based FirePros were only released a couple of days ago. Earlier VLIW cards were much less compute-friendly, so this is kind of new for AMD.

Though it's unfortunate that AMD has canceled the FireStream series given that they finally have a solid compute architecture in GCN.

[CarstenS]

Are you sure that there won't be a FirePro "C" (for compute, or maybe "S" for servers compared to "w" for Workstation) anytime soon?

[Ryan Smith]

Originally Posted by CarstenS

Are you sure that there won't be a FirePro "C" (for compute, or maybe "S" for servers compared to "w" for Workstation) anytime soon?

I'm sure as press can be. If AMD is working on such a product then they aren't ready to announce it.

What we're being told is that FireStream has been eliminated entirely, and that the new FirePro cards are going to serve as both AMD's pro graphics product and their dedicated compute product. When I asked AMD about that, reinforced that position by noting that there's at least one compute cluster being constructed using the new cards.

[Alexko]

Originally Posted by Ryan Smith

I'm sure as press can be. If AMD is working on such a product then they aren't ready to announce it.

What we're being told is that FireStream has been eliminated entirely, and that the new FirePro cards are going to serve as both AMD's pro graphics product and their dedicated compute product. When I asked AMD about that, reinforced that position by noting that there's at least one compute cluster being constructed using the new cards.

Why is that a bad thing? What did FireStream cards offer that FirePros don't?

[CarstenS]

Originally Posted by Alexko

Why is that a bad thing? What did FireStream cards offer that FirePros don't?

That's precisely what I was trying to imply.

[dkanter]

I agree that it doesn't really make sense to have a compute vs. workstation card for AMD. If you look at Nvidia, the ASIC is the same. It's not like you want to change the # of cores, the memory controllers or anything. You want the highest performance, and both should have ECC options.

The difference is the board, since some of the compute cards are mounted in slightly different ways. E.g. some compute cards seem to have really beefy heatsinks to achieve slightly higher density.

Given that AMD isn't starting from huge volumes, why have a dedicated part? It seems far smarter to just release what you have, and when the volumes grow, then you can use that to justify a dedicated SKU for compute.

DK

[Dave Baumann]

FirePro has been used as the compute brand for a couple of generations. You'll note a number of FirePro products with a "P" suffix on the brand string - AFAIK this denotes "passive" (cooler) for server rack mounted application and with limited display outputs (i.e. FirePro V7800P, FirePro V9800P)

[Ryan Smith]

Originally Posted by Alexko

Why is that a bad thing? What did FireStream cards offer that FirePros don't?

Oh don't get me wrong, it's not a particularly bad thing (since as Dave notes AMD traditionally has passive FirePros too); anything you could do on FireStream you could do on FirePro, which has always been a problem for FireStream. It's just unfortunate timing - if anything could drive the success of FireStream, it would have been GCN.

Though now that I thinkof it, it does create a pricing hole for AMD. FirePros are traditionally priced lower than the equivalent Quadro but higher than the equivalent Tesla. The MSRP on the K20 (once it finally launches) will be $800 lower. Having only 2 product lines means that AMD has to pick whether they compete with Tesla pricing or Quadro pricing.

[Kaotik]

Originally Posted by Ryan Smith

Oh don't get me wrong, it's not a particularly bad thing (since as Dave notes AMD traditionally has passive FirePros too); anything you could do on FireStream you could do on FirePro, which has always been a problem for FireStream. It's just unfortunate timing - if anything could drive the success of FireStream, it would have been GCN.

Though now that I thinkof it, it does create a pricing hole for AMD. FirePros are traditionally priced lower than the equivalent Quadro but higher than the equivalent Tesla. The MSRP on the K20 (once it finally launches) will be $800 lower. Having only 2 product lines means that AMD has to pick whether they compete with Tesla pricing or Quadro pricing.

Actually it's $3199


edit:
Funny thing, the slide originates from article you wrote, but still you thought the price is only 1/4th of what it actually is?

[Alexko]

Originally Posted by Kaotik

Actually it's $3199


edit:
Funny thing, the slide originates from article you wrote, but still you thought the price is only 1/4th of what it actually is?

I think Ryan meant $800 cheaper than the FirePro W9000, which costs $4000.

[Ryan Smith]

Originally Posted by Alexko

I think Ryan meant $800 cheaper than the FirePro W9000, which costs $4000.

Correct. $4K for a W9000, $3.2K for a Tesla K20.

[Kaotik]

Whoops, my bad

[Alexko]

Originally Posted by Ryan Smith

Correct. $4K for a W9000, $3.2K for a Tesla K20.

And you have a good point too, I'm not sure how AMD can compete with a cheaper Tesla that's also likely to be more powerful.

And if they drop the price to, say, $2800, they're throwing profit away on the workstation (Quadro) market. I guess they'll have to split the lineup again. How did they handle it for the 40nm generation?

[no-X]

I think W9000 is priced as Quadro 6000's competitor.

[Dave Baumann]

"P" vs non "P" are different products with different target markets, and are sold as such - if you want multiple displays, or framelock or to simply put it in a worksation tower you're not going to be buying the passive board with one displayout. A similar analogy would be Radeon 7970 vs 7970M, do we believe these are the same products with the same target markets?

A simple search on the Newegg from the links above vs their respestive non-P branded products should give you some indication on how they are priced differently.

[Dave Baumann]

Originally Posted by Ryan Smith

Oh don't get me wrong, it's not a particularly bad thing (since as Dave notes AMD traditionally has passive FirePros too);

Not "traditionally", no. As I mentioned, though, the FireStream brand being depricated in favour of FirePro is not necessarily a new one. I don't see that it changes the go-to-market approach though.