AMD Piledriver rumours ... and expert conjecture

[Reynod]

We have had several requests for a sticky on AMD's yet to be released Piledriver architecture ... so here it is.

I want to make a few things clear though.

Post a question relevant to the topic, or information about the topic, or it will be deleted.

Post any negative personal comments about another user ... and they will be deleted.

Post flame baiting comments about the blue, red and green team and they will be deleted.

Enjoy ...

 

[dogman_1234]

Global Foundry's fabs are currently at 32 nanometre nodes. It would be expensive to transistion to 28. If I were AMD in search of a new node, I would jump to 18 nanometres.

 

[metallifux]

where are AMD's GPU's produced?

 

[$hawn]

where are AMD's GPU's produced?

TSMC, even the Brazo's are made here.

 

[metallifux]

TSMC, even the Brazo's are made here.

do they have the ability to manufacture CPU's there?

 

[$hawn]

do they have the ability to manufacture CPU's there?

Yup...they do but its mostly ARM based stuff....like Apples upcoming A6 for example....although AMD's Zacate is the normal x86 type CPU

 

[sonoran]

do they have the ability to manufacture CPU's there?

I'm sure they could, but the entire CPU+GPU would have to be designed with the TSMC 28nm design rules. You can't just take any design and toss it on any production process.

 

[fazers_on_stun]

Uh,oh...

http://www.pcmag.com/article2/0,2817,2395447,00.asp

And http://www.tomshardware.com/news/q3-2011-earnings-call-bobcat-brazos-fusion,13854.html

Citigroup analyst Glen Yeung picked the omission up and asked "it sounds like what fell short must have been desktop. And I wonder as I think about the execution issue: One, do you sense that there is a longer term impact from that? Did you just turn some customers off forever because you screwed up on the execution side?"

Read somewhat evaded the question, but eventually noted: "In the desktop space, there's a little less pressure, but we had to choose where we did our manufacturing capacity in order to support our notebook growth and to make sure that we try to deliver on the commitment that we made to our customers. And they felt some of that pain in the third quarter because we weren't able to execute as cleanly as we would like. In the notebook space, we're making progress. In the desktop space, I think we know how to manufacture in that space. We just need to be able to move more of the wafers in that direction."

After a lengthy answer that did not answer the analyst's question and left us speculating, Read apologized and said that he "got fired up on that one." In aggregate, it appears that the ongoing manufacturing and yield issues at Globalfoundries have especially hit AMD's desktop products. AMD is waiting to get access to higher margin 32 nm products, but yields aren't where AMD is expecting them to be. Additionally, the transition to 32 nm forced the company to compromise on 45 nm production capacity as well.

CFO Seifert noted that the AMD clearly has some headwinds in the current quarter as well: "45 nm supply is still going to be not where we want it to be because we continue to trade off capacity towards 32 nm. We also will see some ramp-up costs from a 28-nanometer technology perspective."

While carefully describing the 32 production as a challenge and a scenario in which AMD makes improvements step-by-step, Read at one point said that he was "disappointed" by the production yields in this space. "And that occurred over a sustained period of time," he said, which was a clear shot in the direction of its manufacturing subsidiary Globalfoundries. AMD can be much more aggressive toward Globalfoundries, by the way, as its share in the company has dropped to 9.6 percent. Furthermore, AMD just lost its final member on the Globalfoundries board. As a customer, your language can be different from the language you use as an owner.

So unless GF does get their 32nm yield problems sorted out, I think AMD may continue to prioritize mobile & server, and thus PD might get pushed back. I've seen a few articles recently stating Q2, instead of Q1.

Also, I hope this puts an end to the "AMD and GF are doing just fine together" and "AMD has half the BoD chairs at GF" spin from our AMD community reporter 😛.

 

[dogman_1234]

AMD needs to jump ship with GF. Wasn't it Ruiz who got AMD to go with GF?

I still think AMD needs to invest in 18 nm nodes, and quickly.

 

[dogman_1234]

http://www.maximumpc.com/article/news/amd_says_trinity_coming_very_early_next_year

Same story. What is with this VLIW4 GPU?

 

[JAYDEEJOHN]

For every dollar spent on GF from their partners, thats one less theyd have to spend, and wouldnt have been able to do whats already occured, so this is the reasons why, where their partners have more monies than Intel or TSMC will ever have, and have and are willing to make these huge commitments.
As thos occurs, of course AMDs shared value declines, but also, with any success from GF, so too will their monies increase due to share count.
PD will benefit from this down the road, and down the road, the other fab goes on line, so, if GF becomes too busy, theyll make more money, both AMD and GF, so AMD is in a good position, its just transitions to 32nm, new arch and waiting on the NY fab.

 

[sonoran]

For every dollar spent on GF from their partners, thats one less theyd have to spend, and wouldnt have been able to do whats already occured, so this is the reasons why, where their partners have more monies than Intel or TSMC will ever have, and have and are willing to make these huge commitments.
As thos occurs, of course AMDs shared value declines, but also, with any success from GF, so too will their monies increase due to share count.
PD will benefit from this down the road, and down the road, the other fab goes on line, so, if GF becomes too busy, theyll make more money, both AMD and GF, so AMD is in a good position, its just transitions to 32nm, new arch and waiting on the NY fab.

The real question is how much money can GF expect to make from its high power, high performance, leading edge process - currently used only by AMD? Unless they're making money with it, they won't have much incentive to invest the time and money to stay on the bleeding edge. The kicker is that GF doesn't seem to be a publicly traded company, so they can be as secretive as they want to be about what they're doing, because they don't have to publish any financials.

 

[we1shcake]

heres what i think needs to happen
1. decrease cache latencies especially the l2
2. cut l3 down massively or even remove it altogether (its not helping much atm)
3. use the spare die area and tdp to add another module or two, giving you a ten/twelve core cpu, in my opinion if your gonna focus on multithreaded do it right
4. find a way to decrease the transistor count of each module ( i have no idea if this is even possible) allowing the cpu to be sold at a lower price and with less power consumption
i think piledriver will increase ipc to the point of phenom 2 or thereabouts, otherwise amd wouldnt be swapping llano out for it (i hope)

 

[JAYDEEJOHN]

The real question is how much money can GF expect to make from its high power, high performance, leading edge process - currently used only by AMD? Unless they're making money with it, they won't have much incentive to invest the time and money to stay on the bleeding edge. The kicker is that GF doesn't seem to be a publicly traded company, so they can be as secretive as they want to be about what they're doing, because they don't have to publish any financials.

Remember, todays cutting edge is tomorrow's everybody elses process.
While the majority of customers may be using 45/90 nm or whatever, they still have to have that cutting edge as some will need it and use it, and tho it may be much cheaper to have an old process to start up, its still not perfected until it actually is

 

[FALC0N]

Interesting. Since BD IPC was actually lower than the prior core used on Llano, they must be banking on a significant core frequency bump on Trinity-- either from Turbo 3 or just better power/thermals.

Actually, if you consider a module the real "core", and many of us do, Bulldozer increased IPC over phenom by 20% or so. A really strong bump for one generation.

Bulldozer wasn't all bad. It just needs refinement.

 

[FALC0N]

Is the 10% boost really any improvement?

You tell me. Sandy Bridge is only 10% faster than westmere.

 

[fazers_on_stun]

The real question is how much money can GF expect to make from its high power, high performance, leading edge process - currently used only by AMD? Unless they're making money with it, they won't have much incentive to invest the time and money to stay on the bleeding edge. The kicker is that GF doesn't seem to be a publicly traded company, so they can be as secretive as they want to be about what they're doing, because they don't have to publish any financials.

+1. IIRC there was some report (perhaps just speculation) earlier this year that GF was losing a lot of money, but ATIC has a ton of cash to pump into the business for the long haul. While there is commonality between the different processes the fabs use (which is why GF's 45nm line also suffered production problems due to the 32nm problems), juggling 3 different nodes - 45nm SOI, 32nm SOI and 28nm strained silicon - in one fab is probably pretty difficult..

 

[BaronMatrix]

And http://www.tomshardware.com/news/q3-2011-earnings-call-bobcat-brazos-fusion,13854.html

Citigroup analyst Glen Yeung picked the omission up and asked "it sounds like what fell short must have been desktop. And I wonder as I think about the execution issue: One, do you sense that there is a longer term impact from that? Did you just turn some customers off forever because you screwed up on the execution side?"

Read somewhat evaded the question, but eventually noted: "In the desktop space, there's a little less pressure, but we had to choose where we did our manufacturing capacity in order to support our notebook growth and to make sure that we try to deliver on the commitment that we made to our customers. And they felt some of that pain in the third quarter because we weren't able to execute as cleanly as we would like. In the notebook space, we're making progress. In the desktop space, I think we know how to manufacture in that space. We just need to be able to move more of the wafers in that direction."

After a lengthy answer that did not answer the analyst's question and left us speculating, Read apologized and said that he "got fired up on that one." In aggregate, it appears that the ongoing manufacturing and yield issues at Globalfoundries have especially hit AMD's desktop products. AMD is waiting to get access to higher margin 32 nm products, but yields aren't where AMD is expecting them to be. Additionally, the transition to 32 nm forced the company to compromise on 45 nm production capacity as well.

CFO Seifert noted that the AMD clearly has some headwinds in the current quarter as well: "45 nm supply is still going to be not where we want it to be because we continue to trade off capacity towards 32 nm. We also will see some ramp-up costs from a 28-nanometer technology perspective."

While carefully describing the 32 production as a challenge and a scenario in which AMD makes improvements step-by-step, Read at one point said that he was "disappointed" by the production yields in this space. "And that occurred over a sustained period of time," he said, which was a clear shot in the direction of its manufacturing subsidiary Globalfoundries. AMD can be much more aggressive toward Globalfoundries, by the way, as its share in the company has dropped to 9.6 percent. Furthermore, AMD just lost its final member on the Globalfoundries board. As a customer, your language can be different from the language you use as an owner.

So unless GF does get their 32nm yield problems sorted out, I think AMD may continue to prioritize mobile & server, and thus PD might get pushed back. I've seen a few articles recently stating Q2, instead of Q1.

Also, I hope this puts an end to the "AMD and GF are doing just fine together" and "AMD has half the BoD chairs at GF" spin from our AMD community reporter 😛.

Why bring me up? You're kind of a trouble maker. What you're sayign is that when TSMC had its 40nm problems, it meant that the relationship was strained or when TSMC raised prices on AMD and nVidia for what was said to be Apple's A6 that meant things weren't going well. I don't remember word of them losing their Board seats, but I do have to work.

And we still haven't heard of another customer needing as much space as AMD, so again AMD is providing MOST of their income, though they are taping out chips from several designers.

But back on topic, I may join XTreme just to ask someone to do lightly threaded AVX\XOP testing. This will show where they put most of the logic. My assumption is with the 0-1 and 0-2 cores affinity tests we won't see as much of a difference. Sure it won't make it faster on older ISAs but if it was Intel, they could and would PAY for optimization. That makes it difficult for AMD as they can't pay AS MUCH.

I'm still getting my 8150 system though.

 

[archibael]

Actually, if you consider a module the real "core", and many of us do, Bulldozer increased IPC over phenom by 20% or so. A really strong bump for one generation.

Bulldozer wasn't all bad. It just needs refinement.

Not a bad point. However, AMD is still talking about them in terms of cores, and the specs for Trinity make this clear (1 and 2 PD module systems, corresponding to the market space for 2- and 4-core Llano). So you'll go from a 4-core Llano to a 2-module Trinity.

Under your conception of modules as "cores", you go from a 4-core chip to a 2-core chip with 20% higher IPC. Which may indeed be okay for a variety of workloads, but might suffer performance wise in higher threading unless Piledriver gets an IPC bump through architecture refinements or is able to be clocked higher.

 

[gnomio]

Amd roadmap
http://mb.zol.com.cn/240/2405453.html

Will Piledriver be the mainstream release and Komodo (God help Amd with these names lol) the high end market release?

 

[gnomio]

is the end of the AM3+ socket with Bulldozer and Piledriver.?
I believe and I might be wrong but after that it will all be on FM2 or FM3;
can someone help verify.
thanks.

Fm2 will be the socket for the 10 Core Komodo chips according to sources.
Piledriver won't have a chipset change.

 

[jimmysmitty]

That article doesn't say there are BD yield issues. Everyone KEEPS saying it's not BD but Llano. IF they are wasting wafers there are fewer for FX.

As far as Trinity:

http://img.donanimhaber.com//images/haber/29938/amdtrinitydetay_2_dh_fx57.jpg


Also the word is that UVD is getting a new feature to challenge QuickSync. Searching for the reference. It's also possible that Trinity is better because the GPU is already 28nm so it's actually the opposite of a shrink and has the simpler VLIW4 arrangement. Also, they can optimize compilers more for LWP (Lightweight Profiling) and XOP\AVX.

Again the problems with X were caused by a perfect storm of SW optimizations - or lack of and Win 7 no understanding modules. Of PD\Trinity can't fix that but I'm pretty certain of the issues I believe are occurring in L1 and L2. FX is the first Shared L2 arch. PD should tweak the Write Through and eviction scheme to increase bandwidth.


It's really difficult to get to in-depth with what they may have done with PD, but I don't think they will get 20% more clock speed at this point. I would think that if they got the first Trinity chips back in June, they have probably finalized max clocks.

BTW, I see everyone started right back in with delete worthy posts. Perhaps I'll come to a SB post and set a good example.

The thing though, is the article did not state any specific architecture technology. It just said that GF was having yield issues with their 32nm process technology. Ergo, the entire lineup of archs from AMD are subject to these yield issues. It would make sense considering that obtaining these brand new Bulldozers is as hard as finding a Dodo in the wild. Almost non-existant for my work while when SB hit, we had them day one.

As for UVD, I will believe it when I see it. QuickSync is quite a feat. Intel really hit nVidia and AMD hard there. I don't think they wont be ble to make something similar, but performance is the only thing I want to see there.

If Trinity has the simpler VLIW4, then its just as I said, based on the HD6K series. That means the only way to improve performance at all is either higher GPU clocks or more SPUs. I am still doubtint Trinity will utilize the new MIMD like the HD7970/7950 will use.

And still, the Windows 7 excuss is over used. It makes no sense for AMD, who has been struggling and not competting for almost 5 years, to design an arch and not work with Microsoft to make sure it will work with 7 to give the best performance. That just insane. I understand designing to make it last but thats different than desigining an arch that can't properly be utilized by the OS that is currently the OS that people are using.

 

[JAYDEEJOHN]

VLIW5 vs VLIW4 comes down to size.
Its around 7% better per area, if I remember correctly, so it will be much better than Llano.
Also, dont give up on UVD enhancements, its the real deal.
The power ratio is/will be much better also for the IGP, minus the size, allows for more shader area/power/perf
It really comes down to how good PDs tweaks will be, for an overall upgrade

 

[gamerk316]

And still, the Windows 7 excuss is over used. It makes no sense for AMD, who has been struggling and not competting for almost 5 years, to design an arch and not work with Microsoft to make sure it will work with 7 to give the best performance. That just insane. I understand designing to make it last but thats different than desigining an arch that can't properly be utilized by the OS that is currently the OS that people are using.

I'll make it even simpler: If AMD is telling the OS that a 4-module BD is really 8 cores, they have no one to blame but themselves if the scheduler assumes all 8 cores are created equal.

A really quick fix could be done via BIOS update: Simply label the second core of a BD module as a logical processor. Now the scheduler will treat those cores simmilar to how they already handle Hyperthreaded cores, helping to avoid the issue.

 

[fazers_on_stun]

Why bring me up?

Because you stated in several posts that "AMD and GF are just fine together". As you can see from Read's statement and finger-pointing, not so..

You're kind of a trouble maker.

Well, geez - your title says "community reporter" -- not 'AMD spin control'. If expecting a modicum of credibility from a CR is making trouble, then so be it.

I don't remember word of them losing their Board seats, but I do have to work.

IOW, you have time to post, but no time to actually check facts first..

At any rate, I would not be surprised to see PD/Trinity pushed back to 2H 2012.

 

[BaronMatrix]

http://www.maximumpc.com/article/news/amd_says_trinity_coming_very_early_next_year

Same story. What is with this VLIW4 GPU?

It stands for Very Long instruction Word 4, meaning there are 4 clusters now instead of 5 previously, which is why 5000 series chips had more SPs (1600 vs 1536). This slows down games slightly but speeds up GPGPU as 4 is a better divisor being a multiple of two.

The GPU is actually a bit smaller also.