AMD CPU speculation... and expert conjecture

[fazers_on_stun]

1] there are trinity updates for 2013 so its not entirely without revision

2] if the goal is to allow better technology to be available to you then hold out for it, from what I last read AMD has 1 more year with the now defective GLOFO, after that they can move away, why put SR on strained GF SoC when the possiblity of TSMC after 2013 may be a realistic possiblity, this bodes better as TSMC have a better process than GF and of course Radeon GPU's will be on the same process, possibly both on 22nm which again is much better than what AMD have with GF.

3] Why the rush, certainly better time and endeavor spent on R&D is more valuable to AMD right now than just releasing for the sake of it. Jaguar and Richland for 2013 to re-evaluate the APU platforms, Trinity 2.0 should be aptly suited to deal with the efforts of HD5000

Sure, but those updates are rather minor compared to the iGPU update for Steamy that AMD originally had slated for a year from now.

As for GF vs. TSMC, it's nice for AMD to have choices. But TSMC also uses strained silicon and not SOI at 28nm. AFAIK AMD was the only CPU/GPU customer anywhere that required SOI - the promised savings by spreading out the capex costs over multiple customers never did materialize; hence AMD switching to strained silicon.

At least one financial analyst agrees with AMD's delaying Steamy and concentrating on the ultra-mobile end: http://seekingalpha.com/article/982931-amd-leaked-roadmap-analysis-prices-margins-and-competitiveness?source=msn

AMD's 2013 roadmap is sensible. On the mid to high end side, we see a logical evolution of the current product line. At the right prices and with the right OEM deals, AMD should be able to position themselves quite well. The exciting stuff happens with the brand new "essential" APU with the "Jaguar" cores in the "Kabini" APU. These cheap chips - coupled with "good enough" performance - could be just what AMD needs to move into newer, low power segments as well as take advantage of the budget ultra-thin notebook category that is likely to spring up.

AMD isn't dead yet, and it doesn't need to "beat" Intel to remain viable - it just needs to pick and choose battles that it can reasonably win. It seems that on the PC client side at least, AMD is making the right moves. Execution and delivery will be key here, which have traditionally been AMD's weak points.

 

[viridiancrystal]

If steamy is released late 2014, it better be the new wheel. Broadwell is scheduled to be out then. Half the transistor size, tri-gates, most likely better process. Assuming Intel can pull 15% for Haswell and 10% for Broadwell (cpu per/watt). Jumps in gpu performance will no doubt be notable. AMD is going to be in two worlds of hurt, considering they are already in one.

 

[Chad Boga]

I have stated several times some possible solutions, but you seem to just want to ignore all of them. Even my initial statement was based on Ivy bridge release data, well before we had any clue what haswell was about.

So you jumped the gun with your wild prognostications, who would have thunk it.

the only wishful thinking around here is you wishing everyone that doesn't worship Intel is a complete moron.

I'm simply of the belief that people who make outrageous statements on scant evidence and obviously no understanding of the history of CPU manufacturing processes, be challenged on those crazy statements.

Imagine the horror of amd's 125W cpu on 14nm.

Why does it have to be a 125W cpu? Intel went from 130W to 95W to 77W, no reason to suspect that AMD can't lower their TDP of future CPU's.

And AMD's biggest challenge for 14nm is more likely to be an economic one, than technical one.

 

[sarinaide]

Sure, but those updates are rather minor compared to the iGPU update for Steamy that AMD originally had slated for a year from now.

As for GF vs. TSMC, it's nice for AMD to have choices. But TSMC also uses strained silicon and not SOI at 28nm. AFAIK AMD was the only CPU/GPU customer anywhere that required SOI - the promised savings by spreading out the capex costs over multiple customers never did materialize; hence AMD switching to strained silicon.

At least one financial analyst agrees with AMD's delaying Steamy and concentrating on the ultra-mobile end: http://seekingalpha.com/article/982931-amd-leaked-roadmap-analysis-prices-margins-and-competitiveness?source=msn

AMD's 2013 roadmap is sensible. On the mid to high end side, we see a logical evolution of the current product line. At the right prices and with the right OEM deals, AMD should be able to position themselves quite well. The exciting stuff happens with the brand new "essential" APU with the "Jaguar" cores in the "Kabini" APU. These cheap chips - coupled with "good enough" performance - could be just what AMD needs to move into newer, low power segments as well as take advantage of the budget ultra-thin notebook category that is likely to spring up.

AMD isn't dead yet, and it doesn't need to "beat" Intel to remain viable - it just needs to pick and choose battles that it can reasonably win. It seems that on the PC client side at least, AMD is making the right moves. Execution and delivery will be key here, which have traditionally been AMD's weak points.

I know the general feeling is "aaah another delay" but I don't really think it is all that bad news for a number of reasons;

1] Richland and Jaguar will release in 2013, Richland will have a dieshrink along with a far more robust unconfirmed but HD8000 iGPU component, not only is it on a much smaller die than VLIW parts but considerably less power for more functional parts, still expected to maintain the 100w TPD for the highest end but lower load power consumption so in a way it is a release.

2] future technologies available to AMD in 2014 makes Kaveri far more attractive for 2014, considering trinity is rarely under threat for iGPU performance.

Hearsay, rumours, speculation on Steamroller;

- Dieshrink to 22nm process
- Unified socket
- Steamroller architecture, focus on lowering icache misses, misspredicts, tightening latencies and yielding sizable IPC gains over Piledriver.
- DDR4 support
- Dedicated memory channels for iGPU with significant improvements to bandwidth.
- Future Radeon cores, also on a significantly reduced die space.
- L3 Cache
- Greater parallelism, guess that has to do with a lot of work in the next 16 odd months with HSA partners, Microsoft expected to be one of them.

It looks more and more like the better decision to release Steamroller when the pieces position themselves a lot better, Richland and Jaguar should offer impressive midpoint varients so its not like the processor line is dry, what will Richland deliver GCN technology, Zero Core 2.0, more GPGPU Compute, more performance on lower power, improved bandwidth, think Trinity 2.0 should have quantifiable gains of Trinity, if I was to guess a number 20-25% iGPU performance gains, far more effiecient architecture than VLIW4, even that may be less considering it was more like 35% with trinity over llano on the same VLIW architectures.

I haven't seen any further slides on FX going forward, I guess this lends to speculation that FX will be unified, I assume that means AMD will be using dual and quad core variants and the Moar Coars ideology seems to be dead, not that it was really bearing fruits anyways.

 

[-Fran-]

Actually, just because of DDR4 if makes sense to push it a whole year. If DDR4 is not (won't be) ready for when they start pushing samples, they won't squeeze more out of the iGPU; or the whole platform for that matter.

Steamroller was going to be the first with the GPU "fused" into the CPU, right? I'm kind of confused with the "new" roadmap now.

Cheers!

 

[esrever]

28nm is MIA at glofo still and kaveri as well as SR FX was slated for that process. Funny how the 28nm part to be fabbed at TSMC is still on track.

 

[Blandge]

I know the general feeling is "aaah another delay" but I don't really think it is all that bad news for a number of reasons;

1] Richland and Jaguar will release in 2013, Richland will have a dieshrink along with a far more robust unconfirmed but HD8000 iGPU component, not only is it on a much smaller die than VLIW parts but considerably less power for more functional parts, still expected to maintain the 100w TPD for the highest end but lower load power consumption so in a way it is a release.

2] future technologies available to AMD in 2014 makes Kaveri far more attractive for 2014, considering trinity is rarely under threat for iGPU performance.

Hearsay, rumours, speculation on Steamroller;

- Dieshrink to 22nm process
- Unified socket
- Steamroller architecture, focus on lowering icache misses, misspredicts, tightening latencies and yielding sizable IPC gains over Piledriver.
- DDR4 support
- Dedicated memory channels for iGPU with significant improvements to bandwidth.
- Future Radeon cores, also on a significantly reduced die space.
- L3 Cache
- Greater parallelism, guess that has to do with a lot of work in the next 16 odd months with HSA partners, Microsoft expected to be one of them.

It looks more and more like the better decision to release Steamroller when the pieces position themselves a lot better, Richland and Jaguar should offer impressive midpoint varients so its not like the processor line is dry, what will Richland deliver GCN technology, Zero Core 2.0, more GPGPU Compute, more performance on lower power, improved bandwidth, think Trinity 2.0 should have quantifiable gains of Trinity, if I was to guess a number 20-25% iGPU performance gains, far more effiecient architecture than VLIW4, even that may be less considering it was more like 35% with trinity over llano on the same VLIW architectures.

I haven't seen any further slides on FX going forward, I guess this lends to speculation that FX will be unified, I assume that means AMD will be using dual and quad core variants and the Moar Coars ideology seems to be dead, not that it was really bearing fruits anyways.

Sounds eerily like BD. A new michroarchitecture and process shrink with powerful features and performance targets. Gets delayed and delayed and delayed, and by the time it is released, none of these features aren't new anymore compared to what Intel is offering. Hope this doesn't happen again :/

 

[fazers_on_stun]

I know the general feeling is "aaah another delay" but I don't really think it is all that bad news for a number of reasons;

1] Richland and Jaguar will release in 2013, Richland will have a dieshrink along with a far more robust unconfirmed but HD8000 iGPU component, not only is it on a much smaller die than VLIW parts but considerably less power for more functional parts, still expected to maintain the 100w TPD for the highest end but lower load power consumption so in a way it is a release.

2] future technologies available to AMD in 2014 makes Kaveri far more attractive for 2014, considering trinity is rarely under threat for iGPU performance.

It looks more and more like the better decision to release Steamroller when the pieces position themselves a lot better, Richland and Jaguar should offer impressive midpoint varients so its not like the processor line is dry, what will Richland deliver GCN technology, Zero Core 2.0, more GPGPU Compute, more performance on lower power, improved bandwidth, think Trinity 2.0 should have quantifiable gains of Trinity, if I was to guess a number 20-25% iGPU performance gains, far more effiecient architecture than VLIW4, even that may be less considering it was more like 35% with trinity over llano on the same VLIW architectures.

Well IMO it's never a good thing to delay a product since the company earns zero income return on the R&D expenses already incurred during the delay.

To repeat from the earlier article: "Execution and delivery will be key here, which have traditionally been AMD's weak points." So if the roadmap rumor is true, AMD again failed to execute and deliver on time another product - Barcelona, Bulldozer and now Steamroller..

The other question is whether AMD can support 9000+ employees on GPUs and lower ASP & margin parts. I'm thinking we'll see another round of layoffs no later than this time next year, if not before. Hopefully their SeaMicro cloud server parts will pay off, as they should earn a pretty good profit on those.

 

[jdwii]

Well IMO it's never a good thing to delay a product since the company earns zero income return on the R&D expenses already incurred during the delay.

To repeat from the earlier article: "Execution and delivery will be key here, which have traditionally been AMD's weak points." So if the roadmap rumor is true, AMD again failed to execute and deliver on time another product - Barcelona, Bulldozer and now Steamroller..

The other question is whether AMD can support 9000+ employees on GPUs and lower ASP & margin parts. I'm thinking we'll see another round of layoffs no later than this time next year, if not before. Hopefully their SeaMicro cloud server parts will pay off, as they should earn a pretty good profit on those.

Their stock wont stop falling, if steamroller is delayed(most likely is) Amd will have to fight another year with inferior products in the server bracket. Kinda want to see them fail(why should we care much when they don't give a crap about their consumers) but then again i don't want to see Intel only make CPU's.

Just wish they would get a real CEO.

 

[blackkstar]

You guys are giving Intel too much credit with Ivy Bridge. The IPC didn't change at all, only thing different is turbo works better: http://www.sisoftware.net/?d=qa&f=cpu_ivb

On top of that, Nehalem vs Sandy Bridge was only about a 15% increase in IPC, and it seemed to hover between 10% and 20%.

I don't expect any IPC improvement from Broadwell. I expect a better working turbo maybe and perhaps more cores thrown at desktop CPUs if they exist, but that's it. Intel has been slowing down immensely.

The difference between P4 and Core 2 Duo was massive, Core 2 Duo to Core i7 Nehalem smaller, Core i7 Nehalem to Sandy Bridge, even smaller, and Sandy Bridge to Ivy Bridge is just a better working turbo. P4 to Core 2 Duo was abandoning Netburst which needed to happen, Core 2 Duo to Nehalem was just putting IMC on die.

The fact that everyone thinks Intel is absolutely destroying with their architecture is absurd. I wouldn't be surprised if Haswell had between 5% and 15% better IPC tops.

 

[sarinaide]

Richland is still expected to have improvements to Piledriver cores along with some tinkering to the IMC, but the more notable one is that it is still expected to feature a new SeaIslands GCN part using much lower wattage. Since we know that AMD is not as religious on massive power drops they will probably see this as the opportunity to throw more transistors into the APU delivering more performance in the same TPD window. I still believe Trinity and Jaguar will be very impressive products but I still can't help but feel a bit disappointed that I have to wait for Q3 2014 to see the next evolutionary AMD product.

As above I think it is the more logical decision with technologies only available in 2014, then the FABS issues etc, it just feels really odd the manner in how this is broken. Anyways of another not sure slides of this will appear on OBR soon enough but looks like FX is scrapped or never intended after Vishera lending more croutons for the soup of thought that AMD are unifying the processors into higher performance APU's.

 

[griptwister]

You think I should Buy a FX-8350 or wait for the Steamroller? You think I can get away with using my 970 board? I'm so confused with all this stuff... They're canceling AM3+... They're not canceling AM3+, Vishera was only supposed to fit the 990FX boards... FM2 is the new highend soccet... IDK. Can anyone give me something more concrete?

 

[sarinaide]

It was said not after Zambezi came out that AMD wanted to unify sockets by Steamroller, after much confusion caused from rumour, speculation, AMD slides it was believed not that long ago that AM3+ and FM2 were going to be carried over, that said I now doubt this very much;

1] 990/SB950 will be nearly 3 years by Q3 2014 and well behind the technologies available with very little room to improve so, for this AM3+ and Vishera is the last pure AMD CPU.

2] Richland appears to be the final Trinity part available on FM2.

3] Lots of speculation suggests Kaveri will employ dual and single module varients featuring L3 cache, some suggest on die memory, though pretty much most of the speculation tends towards DDR4 support of which the FM2 chipset does not support, Also dedicated DIMM slots toward the iGPU along with unified CPU/iGPU IMC.

To me it looks like a unified possibly a FM3 socket for Steamroller and possibly Excavator after that.

 

[jdwii]

It was said not after Zambezi came out that AMD wanted to unify sockets by Steamroller, after much confusion caused from rumour, speculation, AMD slides it was believed not that long ago that AM3+ and FM2 were going to be carried over, that said I now doubt this very much;

1] 990/SB950 will be nearly 3 years by Q3 2014 and well behind the technologies available with very little room to improve so, for this AM3+ and Vishera is the last pure AMD CPU.

2] Richland appears to be the final Trinity part available on FM2.

3] Lots of speculation suggests Kaveri will employ dual and single module varients featuring L3 cache, some suggest on die memory, though pretty much most of the speculation tends towards DDR4 support of which the FM2 chipset does not support, Also dedicated DIMM slots toward the iGPU along with unified CPU/iGPU IMC.

To me it looks like a unified possibly a FM3 socket for Steamroller and possibly Excavator after that.

Why would they do that makes no since they still make money on the fx parts in the go pure APU they probably will go out of business whats next getting rid of their GPU sector to just focus on APU's :/

 

[sarinaide]

Why would they do that makes no since they still make money on the fx parts in the go pure APU they probably will go out of business whats next getting rid of their GPU sector to just focus on APU's :/

Think about it, FX is a game that AMD don't play anymore ever since the "FUTURE IS FUSION" policy was adopted, FX doesn't fit the bill since it is not fitted for HSA workloads, not only that but FM2 APU's closed the gap to top FX parts, FX has no future and you basically get the impression AMD are disinterested in it hence whey ES samples of PD to retail showed absolutely skint nothing in improvements considering the ES's were out in February.

I would not look at a Kaveri as a lil APU, consider the features I have absolutely no doubts it will embarass the best FX processors today, the catch is they are already talking around $200 for the top end Kaveri APU's, so as they become more high end so does the costs incur.

 

[mayankleoboy1]

http://www.geeks3d.com/20121108/the-butterfly-effect-unity-4/

nice video made using unity4 engine.

 

[fazers_on_stun]

Their stock wont stop falling, if steamroller is delayed(most likely is) Amd will have to fight another year with inferior products in the server bracket. Kinda want to see them fail(why should we care much when they don't give a crap about their consumers) but then again i don't want to see Intel only make CPU's.

Just wish they would get a real CEO.

All stocks fell yesterday, due to Obama's re-election plus the EU ongoing crisis. If Obama and the Republicans can't reach agreement to stave off the sequester going into effect Jan. 1st, you will see the USA going into recession #2 and dragging the world economy down with it..

If Read can successfully transition AMD to an ARM + x86 competitor, then history will smile upon him . Otherwise he'll be another Ruiz, maybe the last one..

 

[bawchicawawa]

What are the benefits of implementing arm into the the revision of trinity?
I'm sure it won't be like tegra 3 with it's low powered 5th core that is used when the others aren't being worked hard.

 

[fazers_on_stun]

You guys are giving Intel too much credit with Ivy Bridge. The IPC didn't change at all, only thing different is turbo works better: http://www.sisoftware.net/?d=qa&f=cpu_ivb

On top of that, Nehalem vs Sandy Bridge was only about a 15% increase in IPC, and it seemed to hover between 10% and 20%.

I don't expect any IPC improvement from Broadwell. I expect a better working turbo maybe and perhaps more cores thrown at desktop CPUs if they exist, but that's it. Intel has been slowing down immensely.

The difference between P4 and Core 2 Duo was massive, Core 2 Duo to Core i7 Nehalem smaller, Core i7 Nehalem to Sandy Bridge, even smaller, and Sandy Bridge to Ivy Bridge is just a better working turbo. P4 to Core 2 Duo was abandoning Netburst which needed to happen, Core 2 Duo to Nehalem was just putting IMC on die.

The fact that everyone thinks Intel is absolutely destroying with their architecture is absurd. I wouldn't be surprised if Haswell had between 5% and 15% better IPC tops.

Think the DT CPUs show around 5% IPC gain from Sandy to Ivy, which is not bad considering Ivy is mostly a die shrink tick with few architectural improvements in the CPU, but a lot of changes in the GPU.

Yeah, most people expect Haswell (tock of Ivy) to show around 10% IPC gains..

 

[de5_Roy]

amd going into arm now is risky imo. they need to have their own customized arm cpu core instead of using the vanilla design. besides, coming up with a decent arm based apu could easily take a couple of years or even more - if they customize it for possible improvement. by that time competitors like qualcomm and samsung will have even more mature chips let alone nvidia. this (arm venture) could end up being a losing game for amd if they don't do it right - launch a product based on ref. arm core and unpolished(for use with arm) radeon gpu and be outdone by the competitors or take time to build something well-performing while existing chips advance further ahead.
i mostly agree with a previous observation - this looks a lot like bulldozer-redux (or reflux), amd improving on previous underwhelming product and hit delay before moving to a smaller fabrication process.

 

[sarinaide]

I see even S/A were confussed, listing Richland initially as Kaveri and then proceded to write a ill informed article recently.

Richland and Kabini still technically PD, can we re-open the PD thread ? Richland should deliver a fair performance increase over Trinity.

 

[de5_Roy]

I see even S/A were confussed, listing Richland initially as Kaveri and then proceded to write a ill informed article recently.

Richland and Kabini still technically PD, can we re-open the PD thread ? Richland should deliver a fair performance increase over Trinity.

really? the kabini(jaguar) diagrams that i saw matches neither steamroller nor piledriver.
http://semiaccurate.com/2012/08/28/amd-let-the-new-cat-out-of-the-bag-with-the-jaguar-core/jaguar_core_diagram/
i don't have in-depth knowledge on these cpu architectures though...

 

[sarinaide]

It was part of the Piledriver family slides once upon a time, but AMD slides change like a power point presentation so I may be wrong. If the Piledriver thread lasted 150 odd pages, I will hate to see how long Steamy's speculation thread is going to be, it looks cast in stone that the soonest Steamroller part will roll out in the second half of 2014.

I think the early assesment will be;

1] Unlikely to beat intel in x86 but will it be within say 5-10% window.

2] How far can AMD revolutionize iGPU performance.

It could be a very exciting thread as more comes out, right now the post Vishera hype will settle down. I think the difference here is most knew what Vishera was going to offer, Steamroller is the unknown quantity.

 

[-Fran-]

Wasn't Jaguar as a whole the successor to Bobcat? Which means an OOO CPU and improvements tweaks?

Cheers!

 

[fazers_on_stun]

It was part of the Piledriver family slides once upon a time, but AMD slides change like a power point presentation so I may be wrong.

Yep - AMD changes roadmaps more often than their underpants

If the Piledriver thread lasted 150 odd pages, I will hate to see how long Steamy's speculation thread is going to be, it looks cast in stone that the soonest Steamroller part will roll out in the second half of 2014.

Well if we wander off-topic like this, my bet is 200 pages easy! . And note that lately my predictions have been on a winning streak! Not to toot my own horn too much..

I think the early assesment will be;

1] Unlikely to beat intel in x86 but will it be within say 5-10% window.

2] How far can AMD revolutionize iGPU performance.

It could be a very exciting thread as more comes out, right now the post Vishera hype will settle down. I think the difference here is most knew what Vishera was going to offer, Steamroller is the unknown quantity.

My bet is that it'll come close to Ivy, but won't beat Haswell in CPU. If still 28nm and not 20nm in 2H 2014, I'd also bet any iGPU improvements might only amount to 30% or less, depending on how big a die and TDP AMD is willing to go with.

If Intel had been willing to make Ivy an actual 95W TDP CPU, instead of 77W, I'd bet that the HD4K would have matched or beaten Trinity's as they could have stuck 24 or maybe even 32 EU's on it..