AMD "Steamroller/Excavator" -info, speculations and experience

[Yeroon]

you are wrong, Richland is in old roadmap still for FM2 and with Piledriver cores - it is "Trinity 2.0" platform in some roadmap. Everything on 28nm node (Kaveri/bdver3) is delayed to 2014, GloFo 28nm is not ready for mass production. Steamroller is 28nm architecture, is delayed too.

Proof on the delay to 2014? All I ever saw was speculation based on the upcoming layoffs.

Yeroon

[Darakian]

No, i think no....Im meaning, we will see maybe 8320/8350 in 95W version later and new FX-8370 (maybe 8390) in 2013

Why not? They've done dumb shrinks before. 22nm might help them hit 95w.

[Smartidiot89]

Why not? They've done dumb shrinks before. 22nm might help them hit 95w.

What's a "dumb shrink"? They don't simply press the shrink button when a process is finished and working chips comes out the oven. There are enormous challenges to it and lots of things needs to be remade - it's costly and extremely time consuming, think not of several months rather several quarters. Just shrinking to a new process without doing any sorts of improvements doesn't make sense.

[trueblue]

http://www.donanimhaber.com/islemci/...gorunmuyor.htm

I recall one of the AMD executives (i can't find the quote anymore) making a statement during an interview and saying something to the effect that the "piledriver core of 2013 will not be the same piledriver core of the initial release".

Maybe they have further enhancements that will occur in 2013 to piledriver before shifting to SR cores in 2014.

[informal]

I think it's just a clock bump. With SR finished there is no point in tweaking PD core and I doubt they can get much more than what we have now anyway. PD is a solid bump in performance Vs Bulldozer,I just don't think it can hold up to Haswell next year,that's all.

[tbone8ty]

Is Radeon 2.0 = GCN?

[Darakian]

What's a "dumb shrink"? They don't simply press the shrink button when a process is finished and working chips comes out the oven. There are enormous challenges to it and lots of things needs to be remade - it's costly and extremely time consuming, think not of several months rather several quarters. Just shrinking to a new process without doing any sorts of improvements doesn't make sense.

A dumb shrink is the term given to shrinking a core without giving a redesign, ex. newcastle to winchester in the athlon 64 days. As I recall the only benefit to doing a dumb shrink is to lower the power consumption.

[EniGmA1987]

A dumb shrink is the term given to shrinking a core without giving a redesign, ex. newcastle to winchester in the athlon 64 days. As I recall the only benefit to doing a dumb shrink is to lower the power consumption.

And to get things tested so the new architecture can be designed better. Its why new node + new arch atr the same time almost always have problems of some sort. Just look at Phenom I and Bulldozer. All fixed with subsequent core tweaks because of too many issues from trying too much at once. Steamroller looks great on paper, I just hope it executes well since it is a shink and somewhat of a core re-design again. Granted, at least its not a 100% architecture change so it has more hope than Bulldozer did.

[demonkevy666]

everywhere I've looked says richland is 28nm APU of piledriver. it's trinity on half node (yes AMD is doing half node with APU's)
Kaveri is steamroller apu on 28nm. Steamroller was already stated to be in tapped out and sampling off 28nm.

richland it self looks to be taken off some older road maps it was on them with Krishna and Wichita. Those are no long their and Brazo 2.0 in both their places.

[devguy]

And to get things tested so the new architecture can be designed better. Its why new node + new arch atr the same time almost always have problems of some sort. Just look at Phenom I and Bulldozer. All fixed with subsequent core tweaks because of too many issues from trying too much at once. Steamroller looks great on paper, I just hope it executes well since it is a shink and somewhat of a core re-design again. Granted, at least its not a 100% architecture change so it has more hope than Bulldozer did.

In the case of Phenom I, it was not their first 65nm processor. Remember Brisbane Athlon 64's?

[sdlvx]

This makes no sense. How can AMD release bdver3 GCC support which implies massive changes to decoders and stuff, and then cancel it? Wouldn't GCC support for Steamroller mean that it's already done? Are they just going to throw the chip away?

Also, what about rumors of PS4 using steamroller? I don't understand. Does Sony need so many Steamroller CPUs that AMD can't provide enough for everyone and has to use a tweaked 32nm PD 2.0 for desktop?

[tbone8ty]

Is Radeon 2.0 = GCN?

hmm whats radeon 2.0?

[Olivon]

If I well understand, Richland is Trinity + HSA feature and Bolton PCH. Graphic parts is maybe the same.

[vietthanhpro]

Is Radeon 2.0 = GCN?

Radeon 2.0 = VLIW4

[tbone8ty]

explains slide a little more maybe

http://wccftech.com/amd-richland-apu...d-8000-series/

hopefully kabini and tamesh on 28nm do well

[tbone8ty]

Radeon 2.0 = VLIW4

i think Radeon 2.0 = GCN

[Smartidiot89]

i think Radeon 2.0 = GCN

I know Radeon 2.0 = VLIW4 Seriously though, it has been stated quite clearly in earlier press materials that "Radeon 2.0" is in Trinity, which is VLIW4. Don't expect more then clock bumps with Richland...

[PedantOne]

i think Radeon 2.0 = GCN

in official AMD TRinity slides is Radeon Cores 2.0 = VLIW4, Radeon GCN 1.0 = GCN

Yeroon = proof is that leaked roadmap?

[Darakian]

This makes no sense. How can AMD release bdver3 GCC support which implies massive changes to decoders and stuff, and then cancel it? Wouldn't GCC support for Steamroller mean that it's already done? Are they just going to throw the chip away?

Also, what about rumors of PS4 using steamroller? I don't understand. Does Sony need so many Steamroller CPUs that AMD can't provide enough for everyone and has to use a tweaked 32nm PD 2.0 for desktop?

Compiler support would imply that the design is done. Fabbing a working chip and doing so en-mass is the next step. Do you have a link for the compiler support?

[vietthanhpro]

Testing performed by AMD Performance Labs. Calculated compute performance or Theoretical Maximum GFLOPS score for 2013 Kaveri (4C, 8CU) 100w APU, use standard formula of (CPU Cores x freq x 8 FLOPS) + (GPU Cores x freq x 2 FLOPS). The calculated GFLOPS for the 2013 Kaveri (4C, 8CU) 100w APU was 1050. GFLOPs scores for 2011 A-Series “Llano” was 580 and the 2013 A-Series “Trinity” was 819. Scores rounded to the nearest whole number.

Now, Trinity: 4 core x 3.8ghz x 8 + 6 SIMD x 64 x 0,8Ghz x 2 = 736 GFLOPS
--> Richland 28nm: 819 GFLOPS
Richland: 4 core x 4Ghz x8 + 6 SIMD x 64 x 0.9Ghz x 2 = 819.2 GFLOPS

[Yeroon]

in official AMD TRinity slides is Radeon Cores 2.0 = VLIW4, Radeon GCN 1.0 = GCN

Yeroon = proof is that leaked roadmap?

Yeah, slide is good enough for me. Havent seen that one earlier.
It makes sense that no major changes will happen until 28nm, so richland being PD cores and vliw4 doesn't suprise me. I did read about trinity 2.0 in the summer, but was hardly mentioned again, but if trinity 2.0 = richland, that makes a lot of sense.

[Manicdan]

this is getting confusing, whats the actual changes for richland compared to trinity? im reading that its just trinity with CGN.

[Alex-Ro]

Well trinity with GCN would be great for a refresh....Some improvement in memory controller would be welcomed also

[PedantOne]

in very old roadmap was Richland APU, it was Dual Core + VLIW-4 196 SP?s, originally Trinity for notebooks!



But this is "new Richland", There are two possibilies about it:

1. Piledriver CPU + VliW-4 GPU on 32nm node, higher clocks, nothing more
2. Whole Trinity (Piledriver/VLIW-4) on 28nm node

I bet first is right!

[Olivon]

More recently, SemiAccurate?s moles have come back and said that Kaveri has slipped yet again to 2014 if it is still alive. Big if. That would put the chip 18 months late best case, and then you have to ask yourself why not just skip it? Time will tell what happens, but one thing that is not in question is if there will be an updated APU in 2013. That answer is definitely no, Intel has the game all to itself next year.

http://semiaccurate.com/2012/11/06/a...gain-2014-now/