[ Bloomberg ] AMD Facing Bleak Future

[cbn]

I would take a more powerful iGPU less Big Core CPU any day, but the direction AMD's headed in right now, the only way we get an iGPU less CPU is if they give us the server die, kinda like Intel's Extreme series.

I would be interested in that too.

But I guess the question is whether the smaller core (ie, cat core) x86 APUs can replace the big core APUs for mobile? I'm guessing they probably could which could help the big core line-up move more upscale.

 

[Jacky60]

We all need Samsung to buy AMD, it's a win win win scenario!

 

[mrmt]

Funny that the article is lining up Samsung as the next AMD, when AMD are getting closer and closer to them. I wouldn't be surprised to see the next semicustom deal come from Samsung- customised x86 part fabbed in their own foundries.

I would. The WSA doesn't have such provision, plus there isn't really anything AMD can do with custom x86 that Samsung wouldn't be able to do with ARM.

 

[poohbear]

*Yawn* we've been hearing the same thing the past 20 years.

 

[mrmt]

I was aware of that with product releases but according to some press releases they sounded like they weren't done with high performance desktop.

In other posts I pointed just that, they corporate communication doesn't match their actions. In fact, they *are* done in high end desktop but they refuse to communicate this. The result is that they still sell FX processors on the market, but take a lot of brand erosion because people start to see AMD committed with obsolete products.

Hopefully the Zen Big cores are big enough, but I wonder do we really need APUs with large cores (when desktop users don't really care about the iGPU size too much)?

I wouldn't have much hope for Zen. It shares a lot of IP blocks and a platform with its ARM counterpart. Given that AMD strategy is built around ARM, I think this will generate constraints for the x86 team, on top of whatever resource limitation they already have.

 

[NTMBK]

I would. The WSA doesn't have such provision, plus there isn't really anything AMD can do with custom x86 that Samsung wouldn't be able to do with ARM.

The main thing it could do is provide x86. x86 still has a lot of value. Samsung has used AMD APUs in its notebooks before, using parts built in their own fabs would be the next logical step.

As for the WSA, it hasn't prohibited them from building parts at other foundries in the past. I doubt it would prevent them from doing it again in the future.

 

[NTMBK]

I wouldn't have much hope for Zen. It shares a lot of IP blocks and a platform with its ARM counterpart. Given that AMD strategy is built around ARM, I think this will generate constraints for the x86 team, on top of whatever resource limitation they already have.

It's quite funny seeing everybody's take on Zen/K12. Over on the S|A forums you have Juan, who is convinced that K12 will be limited because it is forced to share parts with the "inefficient" x86 part- his views, not mine- while here we have people who are convinced that Zen will be limited because it has to share parts with K12.

How about we wait until we actually learn some more details about both parts before making sweeping generalisations? We know essentially nothing about either core, or the SKUs they will be going into. We don't even know if it will be big core or little core, CMT, SMT or SMP, single socket or multisocket, CPU or APU... Either pinning our hopes on it or writing it off entirely just seems laughable at this point.

 

[mrmt]

The main thing it could do is provide x86. x86 still has a lot of value. Samsung has used AMD APUs in its notebooks before, using parts built in their own fabs would be the next logical step.

But that's not what Samsung needs. Samsung needs *competitive* x86 parts, something that AMD isn't really able to deliver by themselves since a few years ago. Building a custom core could solve this problem, if adequately funded, but Samsung market share isn't big enough to make this a viable venture and, and we would be 4 years down the road, because whatever AMD is developing now is the result of the resource-choked pipeline of the Rory era, probably not enough for Samsung needs.

As for the WSA, it hasn't prohibited them from building parts at other foundries in the past. I doubt it would prevent them from doing it again in the future.

The WSA does prohibit AMD from manufacturing chips in other foundries, that's why AMD must get a waiver from GLF every time they want to use TSMC. With GLF stance getting tougher (they forced AMD to backport their 28nm designs to its process and are bringing even GPUs there), I really doubt they would allow AMD to use Samsung as a foundry.

 

[mrmt]

It's quite funny seeing everybody's take on Zen/K12. Over on the S|A forums you have Juan, who is convinced that K12 will be limited because it is forced to share parts with the "inefficient" x86 part- his views, not mine- while here we have people who are convinced that Zen will be limited because it has to share parts with K12.

Sharing IP blocks between the chips are surely going to constrain both products, one more than the other. He just thinks the more constrained part will be the ARM part, I think quite the opposite, the biggest constraints will go for the x86 part. AMD strategy is ARM, they will focus there. x86 just happens to be an afterthought for them, a money-losing afterthought, especially after 2015.

 

[NTMBK]

But that's not what Samsung needs. Samsung needs *competitive* x86 parts, something that AMD isn't really able to deliver by themselves since a few years ago. Building a custom core could solve this problem, if adequately funded, but Samsung market share isn't big enough to make this a viable venture and, and we would be 4 years down the road, because whatever AMD is developing now is the result of the resource-choked pipeline of the Rory era, probably not enough for Samsung needs.

"Competitive" has many facets. An internally produced semicustom part would reduce cost significantly compared to Intel parts, and could be tuned to provide exactly what Samsung desire instead of having to choose an off-the-shelf part which is closest to what they want.

The WSA does prohibit AMD from manufacturing chips in other foundries, that's why AMD must get a waiver from GLF every time they want to use TSMC. With GLF stance getting tougher (they forced AMD to backport their 28nm designs to its process and are bringing even GPUs there), I really doubt they would allow AMD to use Samsung as a foundry.

Is there any evidence that AMD were forced to port designs to GloFo? Given that they get favourable pricing from GloFo in return for the take-or-pay arrangement, they could simply be choosing to move products over to GloFo in order to make the most of the lower costs and greater availability (instead of competing with Qualcomm for TSMC wafers). Not to mention that the GloFo 28nm process seems slightly more efficient than the TSMC one, providing another incentive to produce e.g. the new mobile GPU for Apple there. And obviously it helps them fulfil their purchase obligation.

 

[Sweepr]

There are cheaper alternatives for Samsung if they intend to develop custom CPU/GPU cores than buying AMD. They have some of the highest clocked big.LITTLE parts and according to rumours they will introduce a custom GPU as soon as next year. They seem to be developing custom CPU cores as well.

 

[NTMBK]

There are cheaper alternatives for Samsung if they intend to develop custom CPU/GPU cores than buying AMD. They have some of the highest clocked big.LITTLE parts and according to rumours they will introduce a custom GPU as soon as next year. They seem to be developing custom CPU cores as well.

But again, they would not be x86 compatible.

And no need for SS to buy AMD just to collaborate with them.

 

[ShintaiDK]

We all need Samsung to buy AMD, it's a win win win scenario!

You mean no more x86 CPUs, consoles dead on the spot and a GPU division we dont know will be killed or continued?

 

[ShintaiDK]

AMD will be gone within 5 years. I am predicting it now.

I doubt they be gone, but they are suffering the VIA outcome. A constant reduction and slow progress into the forgotten lands.

 

[mrmt]

"Competitive" has many facets. An internally produced semicustom part would reduce cost significantly compared to Intel parts, and could be tuned to provide exactly what Samsung desire instead of having to choose an off-the-shelf part which is closest to what they want.



Is there any evidence that AMD were forced to port designs to GloFo? Given that they get favourable pricing from GloFo in return for the take-or-pay arrangement, they could simply be choosing to move products over to GloFo in order to make the most of the lower costs and greater availability (instead of competing with Qualcomm for TSMC wafers). Not to mention that the GloFo 28nm process seems slightly more efficient than the TSMC one, providing another incentive to produce e.g. the new mobile GPU for Apple there. And obviously it helps them fulfil their purchase obligation.

Competitive has indeed many facets, but why do you assume that I'm ingnoring them? AMD for example is in a far worse competitive situation regarding costs than with performance, because the basis of their designs is fundamentally uncompetitive. How semi-custom is supposed to change that? I can't see how, because semi-custom is all about adding a third party IP to a design (low risk approach), not rebuilding fundamental ip blocks in order to reduce costs (that would like fully custom, and a high risk approach). In the end Samsung would have to bear the costs of a brand new, competitive x86 design, but outsourcing this task to a company with such an atrocious track record.

As for the WSA, there's devinder kumar affirming that they are obligated to manufacture all their MPUs with GLF. Check here:

http://www.brightsideofnews.com/201...-targets2c-revises-globalfoundries-agreement/

 

[NTMBK]

And yet we have confirmation that AMD will be producing 14nm parts at Samsung: http://techreport.com/news/27149/report-samsung-to-produce-14-nm-chips-for-amd

As for "brand new, competitive x86 design", we have yet to see how Zen will do. Yes, clearly the BD family just isn't competitive, we can all agree on that, but given that we are looking at 14nm parts I expect that we are talking about Zen, not BD.

 

[ShintaiDK]

Zen isnt coming anytime soon is it. And with the R&D allocated, its quite hopeful to put it mildly to expect anything extraordinary.

 

[NTMBK]

Zen isnt coming anytime soon is it. And with the R&D allocated, its quite hopeful to put it mildly to expect anything extraordinary.

Zen is coming in 2016, right around when Samsung/GloFo's 14nm process will be ready- and coincidentally when its next x86 semicustom part is meant to be launching.

 

[mrmt]

And yet we have confirmation that AMD will be producing 14nm parts at Samsung: http://techreport.com/news/27149/report-samsung-to-produce-14-nm-chips-for-amd

It might be possible, especially because the process will be the same of GLF's, so there's no backporting involved. If this report is true then that means GLF process won't be available at the same time frame as of Samsungs's

As for "brand new, competitive x86 design", we have yet to see how Zen will do.

AMD has less than half of the R&D budget they have 5 years ago, they have less R&D money than Nvidia (and they got K1 with that) and far less than Qualcomm, that evidence is far more solid than AMD PR kool-aid.

If AMD is overshooting its R&D budget and going for more than they can handle the project will be delayed or they won't be able to deliver at all. If they don't, then the project will be much more limited in terms of scope than previous one (bigger than cat core, but not much bigger).

 

[ShintaiDK]

Zen is coming in 2016, right around when Samsung/GloFo's 14nm process will be ready- and coincidentally when its next x86 semicustom part is meant to be launching.

That could easily be 2 years from now, plus any delays.

 

[caswow]

there are no amd thread where overall everything is negative.

 

[NTMBK]

It might be possible, especially because the process will be the same of GLF's, so there's no backporting involved. If this report is true then that means GLF process won't be available at the same time frame as of Samsungs's

And where are you getting that from? No mention of that in the report.

AMD has less than half of the R&D budget they have 5 years ago, they have less R&D money than Nvidia (and they got K1 with that) and far less than Qualcomm, that evidence is far more solid than AMD PR kool-aid.

If AMD is overshooting its R&D budget and going for more than they can handle the project will be delayed or they won't be able to deliver at all. If they don't, then the project will be much more limited in terms of scope than previous one (bigger than cat core, but not much bigger).

It doesn't need to be a massive Haswell-killer core to be successful, necessarily. Something not much larger than the cat cores would probably be plenty for laptops- especially since the market is tending towards thin and light devices- and most consumer desktops. Amusingly Intel have made things easier for AMD in that respect, by lowering users' performance expectations with NUCs, 2-in-1s and Ultrabooks. Anyway, we shall have to wait and see. I certainly agree that it will be difficult for AMD and I have my doubts about whether they can succeed, but that is a long way from completely writing it off.

 

[NTMBK]

That could easily be 2 years from now, plus any delays.

I am aware of that- I can also subtract two numbers correctly. :thumbsup:

Yes, AMD certainly has a lean 2015 ahead of them. On the "big core" side they have a Carrizo placeholder for mobile and... that's about it. Still, they should finally get some stacked memory parts out of the door for the GPU market, so there's at least some interesting parts. There's also the 20nm ultramobile SoCs which are meant to be happening in 2015, though they've been kind of quiet of that front. It's a bit weird. I guess 28nm Carrizo-L is aimed at the craptop market, and the 20nm parts are aimed at markets where power consumption matters most (like 2-in-1s)?

 

[cbn]

It's quite funny seeing everybody's take on Zen/K12. Over on the S|A forums you have Juan, who is convinced that K12 will be limited because it is forced to share parts with the "inefficient" x86 part- his views, not mine- while here we have people who are convinced that Zen will be limited because it has to share parts with K12.

Wasn't the the lead designer quoted not too long ago saying that K12 would have a bigger engine than Zen?

 

[ShintaiDK]

Zen is not even taped out yet. It could for the matter of argument still be cancelled.