Steamroller core

[Olikan]

slides:

http://www.tomshardware.com/news/AMD-Steamroller-Piledriver-Kaveri-processors,17217.html
http://www.pcper.com/reviews/Processors/AMD-Unveils-Steamroller-Improvements

 

[Ancalagon44]

Steamroller: back to K8 level of performance? Lol. We can only hope.

Why are we seeing Steamroller slides when Vishera isnt even out yet? Are they trying to prepare us for a disappointment by hyping Steamroller?

Sorry I'm just jaded on AMD after they hyped BD up to much and it turned out to be the return of the Pentium 4.

EDIT: Its interesting that the slides seem to focus on increasing single threaded execution,particular integer execution, via faster instruction decoding and especially faster integer execution. Here is an idea AMD - if you didnt remove all of this great stuff from K10 to make BD, maybe you wouldnt need to go through the trouble of adding it back to Steamroller?

 

[joshhedge]

I hope it lives up to the hype. Competition is always welcome. Will be interesting to see how the APUs do.

 

[Idontcare]

Is Steamroller a 32nm CPU?

 

[grimpr]

Looks great, on slides, i bet it will be a Phenom 2 like success for AMD.

 

[sefsefsefsef]

"Major improvements in store handling" and "Dynamic resizing of L2 cache" are the two most interesting parts to me. Do you think they're abandoning "write-through" for the L1-D cache? Also I hope that the dynamic L2 resizing is going to be a performance enhancement option, and not just a power savings technique.

 

[Rezist]

Looks like 28nm TSMC according to anand's article

 

[Haserath]

At a store near you in 2013-2015 or perhaps later judging by AMD's history.

Hopefully it's a late 2013 release at the latest.

 

[AnandThenMan]

Looks like 28nm TSMC according to anand's article

Not correct (I see it has been fixed in the article) it will use GloFo's 28nm bulk process.

 

[SlowSpyder]

Sorry if I missed in in the article, but does anyone know if this is an AM3+ part, or a new socket? I got a long useful life out of my AM2+ (still on it in fact), so I'm just curious.

 

[inf64]

Steamroller: back to K8 level of performance? Lol. We can only hope.

Be serious please. Bulldozer is faster than K8,it's around the level of first gen K10 core (65nm Barcelona). This core is ~10-15% faster than K8 in integer and 1.5-2x faster in SSE workloads. In FP stuff Bulldozer is just crushing K8 (not hard since K8 has 64bit FPU).

On topic of SR core,integer execution looks to be having the major perf. boost while FP coprocessor is sort of a letdown. It's the same old FP unit we have in BD and PD. AMD even states they cut it down further to save power and area (they apparently axed one MMX pipe). Unless they somehow found a way to make that same old FP unit perform 50% better (if it had a major design bug?) then they will stand no chance against Haswell. Haswell will add FMA execution support and then AMD will need 4x the "cores" of intel equivalent Haswell server chip to match its FP throughput . It's obvious why this is not realistic(16 modules SR server chip is not gonna happen).

 

[Haserath]

Sorry if I missed in in the article, but does anyone know if this is an AM3+ part, or a new socket? I got a long useful life out of my AM2+ (still on it in fact), so I'm just curious.

Last I heard, Piledriver was the last AM3+ processor.

On topic of SR core,integer execution looks to be having the major perf. boost while FP coprocessor is sort of a letdown. It's the same old FP unit we have in BD and PD. AMD even states they cut it down further to save power and area (they apparently axed one MMX pipe). Unless they somehow found a way to make that same old FP unit perform 50% better (if it had a major design bug?) then they will stand no chance against Haswell. Haswell will add FMA execution support and then AMD will need 4x the "cores" of intel equivalent Haswell server chip to match its FP throughput . It's obvious why this is not realistic(16 modules SR server chip is not gonna happen).

It says the same throughput at lower power/area.

What should help improve overall performance is the cache/prediction/front end improvements.

FMA also is only a fraction of all FP instructions.

 

[Phynaz]

Going back to independent decoders for each core.

The past is the future, or something like that. Guess it invalidates the major design ideas behind bulldozer.

 

[Jovec]

Unfortunately these are the changes that needed to be included with Piledriver in 2012, not Steamroller launching sometime in 2013.

 

[Schmide]

Arggg WTF AMD that's a front loader dude.

 

[SunnyD]

Wait wait wait... I thought THIS was universally known/admitted to be the problem with Bulldozer - the fact that they moved away from hand-tuned/drawn logic and moved to automated crap which ended up costing them in terms of power, performance AND delays. 😕

(A quick google = first link found)

 

[zebrax2]

^ The first thing that came to my mind when i read the article.

Does steamroller have AVX2?

 

[eternalone]

Actually I think steam roller will be released early. Hopefully vishera is cancelled all together and they release steam roller/frontloader in January. That would be nice.

 

[Schmide]

Actually I think steam roller will be released early. Hopefully vishera is cancelled all together and they release steam roller/frontloader in January. That would be nice.

Is there a Front Loader?

I thought it went

Bulldozer
Steam Roller
Pile Driver
Excavator

 

[Olikan]

^ The first thing that came to my mind when i read the article.

Does steamroller have AVX2?

when i read the article, it says that FPU would lose peak frequency...
yet the cpu is made to reach high performance

AVX-2 ? probably not...way too soon.....maybe in excavator

 

[Jovec]

Is there a Front Loader?

I thought it went

Bulldozer
Steam Roller
Pile Driver
Excavator

BD
PD
SR
EX

The frontloader comment refers to the frontloader tractor pictured on the steamroller slide. Alas, if CPU design doesn't work out for them it appears they don't have much of a future in construction either...

 

[eternalone]

BD
PD
SR
EX

The frontloader comment refers to the frontloader tractor pictured on the steamroller slide. Alas, if CPU design doesn't work out for them it appears they don't have much of a future in construction either...

lol

 

[inf64]

when i read the article, it says that FPU would lose peak frequency...
yet the cpu is made to reach high performance

AVX-2 ? probably not...way too soon.....maybe in excavator

No the article doesn't state that. AT has a good coverage. FP unit is "streamlined" so that it's smaller and yet performs the same as old one. SO basically from execution POV they didn't expand it. Whether the rest of the core changes(which are significant) may help the FP execution ,remains to be seen. Clock frequency may suffer due to other factors (teh core is more complex now).
AVX2 is a possibility. They will give Jaguar full ISA support next year . I say full since Jaguar will probably launch around Haswell and AVX will still be the top level ISA one can support. AMD already supports FMA3 with PD so they need to add support for integer 256bit AVX instructions. Since the FP unit remains unchanged , the AVX2 execution will be done analog to how Bulldozer does FP AVX : across one module. So not a big deal I guess. Performance on the other hand may be a big deal since Haswell will probably eat it alive in AVX2 code 🙂.

 

[tweakboy]

I sure they fixed all their mistakes and what not in time. But Intel will always be the leader as far as CPU and only CPU , not the graphic card port...

 

[tweakboy]

What are you expecting out of it. They flopped once. I think they do good this time around, but Intel will always have better tech and what not in desktop field..