AMD "Piledriver" refresh of Zambezi - info, speculations, test, fans

[JumpingJack]

I find it interesting that in the Graphics/MM bullet that AMD is including "UVD 3 with Secure Asset Management & Video Compression Engine". Sounds a lot like Intel's Insider (for streaming copy protected video) and Quicksync for full HW transcoding. The former makes a lot of sense, but the latter is odd -- it short circuits on of the killer apps for GPGPU computing.

[Oliverda]

The guy with Trinity A0 says that it just matches Llano at 2.9Ghz in 3dmark and c11.5 ... He didn't say what model but I assume top 3.8Ghz. If so then there is no improvement versus bdver1...

If so there is.

A8-3850 (2.9 GHz): 3.45 pts
FX-4100 (3,6 GHz): 2,94 pts

Trinity (3.8 GHz?): 3,45 pts?

Anyway CB is one of the applications which doesn't suit for BD.


By the way let's see what were the ES BD's base clocks:





3.2 GHz max.


So we may expect 4 GHz+ base clocks for the start which should have already been there for the BDs.

[informal]

There is no way that it can be an improvement...
AMD is now 80% behind SandyBridge in Cinebench CPC, uses ~20% more power than the old architecture...giving a generous 10% leeway in a 30% gate shrink, 30% behind in WinRAR, which happens to be the only place so far that FX really "shines".
I hope Piledriver comes with some "I'm on acid" 50% IPC increase...

They need 20% more clock vs Llano so 700 Mhz over Llano, 3.8 Ghz should beat Llano by a few percent

Cinebench is irrelevant metric,but with same thread count AMD is just 10% or so behind SB. This is very good result in one of the benchmarks that don't fit well with Bulldozer's shared FPU. It does consume more power but this is entirely GloFo's fault. Design is probably not working as well as AMD had hoped but to call it a failure is wrong. Clock for clock is irrelevant for most buyers,general performance and performance/$ is what matters. I have already posted links in my previous post where BD destroys Thuban in almost all real world applications that are relevant for desktop users. In games it's on par or faster than Thuban/Deneb. It does cost more but with slight price correction Thuban is going the dodo way soon with 8170 and Piledriver coming. Just forget K10,it was a good core let it go

[Olivon]

I have already posted links in my previous post where BD destroys Thuban in almost all real world applications that are relevant for desktop users.

In real world applications, BD doesn't destroy Thuban or maybe I missed something

[informal]

In real world applications, BD doesn't destroy Thuban or maybe I missed something

Real world apps:
http://www.techspot.com/review/452-a...pus/page8.html
http://www.techspot.com/review/452-a...pus/page9.html
Games:
http://www.techspot.com/review/452-a...us/page10.html
Games show the smallest difference which is normal as they mostly are not CPU bound. But still ,BD is not slower at least and sometimes decently pulls ahead.

[Olivon]

I know you like Hardware.fr reviews, and applications test was well choosen by the author in order to use BD at maximum with all kinds of applications :



http://www.hardware.fr/articles/842-23/moyennes.html

Thuban is behind BD but by only ~7%, that's not a destroy and that's what I read in many reviews.

If I look to the conclusion of your link, they don't talk about an X6 slaughter :

We won't deny it, we really were hoping for a lot more from Bulldozer and AMD's eight-core processors. It's disappointing to find these newly launched processors do little to improve AMD’s situation. The FX processors come short of competing hand to hand with the now 9-months old Sandy Bridge processors, and in certain instances surpass their own Phenom II range. Still, this is just the start for Bulldozer, and there's much more to be seen from the FX range, or so AMD says.

[informal]

Well I did say real world applications. Most desktop buyers use just these: excel,winrar,adobe ps,handbrake,tmpgenc. In these tasks BD is much faster than Thuban/Deneb. Hardware.fr uses "synthetic" workloads and games that skew the average one way or the other. That's why I prefer a real world application tests.

[Oliverda]

In real world applications, BD doesn't destroy Thuban or maybe I missed something

The word "destory" is definitely too strong but according to our suite the FX-8150 beats 1100T in real world apps overall.



source


Just to add to my above post:

[muzz]

Well I did say real world applications. Most desktop buyers use just these: excel,winrar,adobe ps,handbrake,tmpgenc. In these tasks BD is much faster than Thuban/Deneb. Hardware.fr uses "synthetic" workloads and games that skew the average one way or the other. That's why I prefer a real world application tests.

Most desktop users I know use their DT to play games, and they use their laptops to do Excel spreadsheet work stuff, so games ARE more important to most folks I know. In an office environment that uses Desktops to do Adobe and the other programs that BD does better at then yeah, BD is the winner there in most applications.
I certainly don't play with WinRar very often on MY desktop, I use it to play games, or else I'd have to use my LT with it's slow as molasses integrated graphics....I shudder at the thought.

[informal]

If you look at gaming results ,FX is on par with Thuban or better,sometimes by a big margin. So gaming is a non-issue IMO.

[Olivon]

Hardware.fr uses "synthetic" workloads and games that skew the average one way or the other. That's why I prefer a real world application tests.

Hardware.fr made a separate average for games with real scenes (not benchmarks) with special cpu usage and no classical GPU bound scheme. BD loose vs. Thuban and Deneb by a small margin :

[FlanK3r]

Hardware.fr made a separate average for games with real scenes (not benchmarks) with special cpu usage and no classical GPU bound scheme. BD loose vs. Thuban and Deneb by a small margin :

Guys,the chinese man told about trinity,right?There is second good info.Trinity for Fm3 has not L3 cache and example in Cinebench make it 10%!So,maybe Piledriver will be here good.Fx 4170 hit 3 points in 11.5

[informal]

Olivon, the gaming numbers range wildly with different reviews. Look here:
http://www.pcgameshardware.de/aid,84...U/Test/?page=3
With GTX480 FX8150 is generally faster than 1100T,especially when minimum fps is in question.

edit:
one more review with HD6970 done on CHV motherboard:
http://www.hardwaresecrets.com/artic...Review/1402/12
FX8150 is usually on par/ahead core i5/7 here. Strange isn't it?

[Olivon]

Guys,the chinese man told about trinity,right?There is second good info.Trinity for Fm3 has not L3 cache and example in Cinebench make it 10%!So,maybe Piledriver will be here good.Fx 4170 hit 3 points in 11.5

You're right FlanK3r, back to Piledriver

If you look at frequencies, OPN 3820 got around 30% more frequency (40% with Turbo mode) than A8 3850 at base frequency. Slide taks about up to 30% performance better than Llano so IPC might be on par with it.
But what about Fm3 ? You mean FM2 ?

@Informal : I used hardware.fr reference because I knew that you used it for your speculation chart on your blog, I thought we have a similar reference. But anyway, no problem, go back on topic as FlanK3r said

[muzz]

If you look at gaming results ,FX is on par with Thuban or better,sometimes by a big margin. So gaming is a non-issue IMO.

Some folks with Bulldozer have serious issues with Steam Games, and Steam Games DOMINATE the market by a LARGE margin.
These problems have been around since launch, and still are unresolved AFAIK.
Maybe AMD should tell THOSE folks that upgraded to their "Newest and bestest" that the fix will be coming in the next 6 months or so, in the meantime you should put your older gen AMD chip back in(which work fine BTW).

I haven't seen the "big margin" that you speak of, and considering the average gamer is still using an AM2+ or AM3 board, it's not like it's an EZ "pop the new chip in after spending $280", Asus bragged about compatibility with their AM3 boards B4 launch, and they just started to give folks the CH1V support.....I don't know about the rest of the boards they stated they would support(there was decent length list IIRC), but it seems to me that the folks complaining about CH1V got first dibs on support. I know of the MSI board too.
I had every intention of UG to an AM3+ and BD( I've been waiting for it), but I am going to skip this gen(like I did PH1) and hope for better down the road.

[undone]

Seems nobody notice the GPU part in Trinity gain turbo boost technology, it always means the higher tdp level right?

http://pics.computerbase.de/3/8/0/2/4/1.jpg

[informal]

@Informal : I used hardware.fr reference because I knew that you used it for your speculation chart on your blog, I thought we have a similar reference. But anyway, no problem, go back on topic as FlanK3r said

Yeah I know mate,but the thing is they changed their methodology after FX launched so I can't directly compare the "new" results with their older average numbers.If you look here , FX is generally around 9-11% or so faster than 1100T in apps and games . It is even more faster with apps and games in low res. ( 11% vs 9% for high resolution in computerbase.de tests).

On topic of PD. If Piledriver launches (big IF) at around 4.1-4.2Ghz base clock and IF it gets ~3-5% IPC improvement versus bdver1(IPC being very moot point since it varies wildly with different workloads of course),then PD might land at 17-22% faster than 8150,a very good improvement which might put it at 2700K level in the "new" hardware.fr overall score testing chart. 2700K overall is very close to 980x.

[BeepBeep2]

Cinebench is irrelevant metric,but with same thread count AMD is just 10% or so behind SB. This is very good result in one of the benchmarks that don't fit well with Bulldozer's shared FPU. It does consume more power but this is entirely GloFo's fault. Design is probably not working as well as AMD had hoped but to call it a failure is wrong. Clock for clock is irrelevant for most buyers,general performance and performance/$ is what matters. I have already posted links in my previous post where BD destroys Thuban in almost all real world applications that are relevant for desktop users. In games it's on par or faster than Thuban/Deneb. It does cost more but with slight price correction Thuban is going the dodo way soon with 8170 and Piledriver coming. Just forget K10,it was a good core let it go

I'm not sure if you realize what you said but essentially that would mean that:
FX-4100=2500K
You say its GloFo's fault entirely that FX consumes this much power? So if they DID happen to shrink Thuban, it would also have consumed nearly as much as it did on 45nm process?

CPC is very irrelevant, however when your competitors product is 50 to 70% faster CPC and overclocks just as high as your product...You can't really ignore it.
We can all try to forget K10, but I'm sure at 32nm it would have performed better than Bulldozer does currently in both Desktop and Server markets.

I hope Piledriver does 5.5+ stable on air and brings at least a 15% IPC improvement, both of which I doubt will happen.

[informal]

I said same thread count. That implies 8 threads on FX8150 vs 8 threads on 2600K. How you got to FX4100 being equal to 2500K I really don't know.
Both chips OCed to approx. the same clock (FX8xxx/2600K) have comparable MT performance(win some lose some). Gaming perf. ,as showed in many reviews, is not worse than Thuban. The only bad thing is power draw which is GloFo's fault.

edit:
I assumed you understand that Bulldozer threads are different than SMT threads in i7. Each SMT thread is 60% of a full i7 core,but you have 8 of them in MT workload. Hence you can gain around 20% speedup in that case. The i7 core is still a "fat" core though. This means it can have higher average single thread IPC than slimmer Bulldozer core (in general case of course) at similar clock.

On the Bulldozer side you have each core representing one thread. The FPU side can be considered a core though,so in FP/SIMD workload "8 core" Bulldozer is actually behaving like 4 core with SMT on. In pure integer workloads it behaves like a "true" 8 core though. That's why we see a mixed bag results with Bulldozer. Sharing the FPU made it possible to pack in more integer cores which is always good for server workloads. But,on the other side,everything non-recompiled is going to bring less speedup or no speedup at all versus K10. It's even a success for BD to outperform X6 in legacy code since it has only 4 FPUs with SMT versus 6 FPUs in K10.

When the day comes for desktop apps to receive the FMA/XOP/AVX optimizations ,bulldozer may gain even more ground on K10. The situation right now is actually a worst case scenario when it comes to BD's desktop performance. Things can only improve from now on: via clock increase/TDP reduction (this is speed racer design after all),via IPC improvements(many things can be improved as AMD states in optimization manual) and via adding more CUs (for a smallish price in die area).

[RussC]

Seems nobody notice the GPU part in Trinity gain turbo boost technology, it always means the higher tdp level right?

http://pics.computerbase.de/3/8/0/2/4/1.jpg

Yup, higher clocks mean higher power.

Um, looking at everything, FX is an enthusiast CPU, and I see no reason for enthusiast to upgrade except the lure of super high clk speeds under LN2/HE. While all those test Informal points out, the Thuban is at stock speeds, clk for clk BD is under-whelming and dissapointing(as all the reviews point out clearly on almost every last one). My 955c2 beats a 4100 clk/clk easily in almost everything. Theres absolutely no reason to do anything. Since BD is a enthusiast CPU, we all over-clock, so theres no there/there. Plus, BD wont show up in Dell, HP, Gateway, eMachines systems either(at least I see no reason why they would), so there see through will be horrific, speaking of which, the power draw issure wont get fixed, not as long as their using SOI/HKMG in its current form. GF won't fix this anytime as there still to me just a poor mans TSMC. If you clock this chip proccess high, power draw will be off the charts, sorry. I no more believe GF can do anything they say as much as I believe AMD will make those improvements in those charts. They never do, but som gents here keep saying "if AMD gest x% blah blah blah improvement with PD", thell be fine. Nope wont happen.

Im essentially out, my 955 will be the last AMD product I build to, even though I got it for free. The power issue with AMD is un-acceptable now. That melted power conn in the other thread was very telling. The 955, It'll be good enough for now till IB hits, then any money I would have spent gets dumped into that. I cant see any money spent on AMD is worth it no matter how much stuff I end up with for nothing doing upgrades for friends/family.

To bring this back around, I dont believe PD then SR will do anything they state. Or better yet, take their #s and subtract 90%, then you'll be alot closer.

RussC

[muzz]

Yeah I know mate,but the thing is they changed their methodology after FX launched so I can't directly compare the "new" results with their older average numbers.If you look here , FX is generally around 9-11% or so faster than 1100T in apps and games . It is even more faster with apps and games in low res. ( 11% vs 9% for high resolution in computerbase.de tests).

On topic of PD. If Piledriver launches (big IF) at around 4.1-4.2Ghz base clock and IF it gets ~3-5% IPC improvement versus bdver1(IPC being very moot point since it varies wildly with different workloads of course),then PD might land at 17-22% faster than 8150,a very good improvement which might put it at 2700K level in the "new" hardware.fr overall score testing chart. 2700K overall is very close to 980x.

I'm not seeing anything great for BD in that Chart, Clock for Clock 1100 holds it's own for the most part, and is even ahead in some of the testing.
Am I missing something?

[informal]

I'm not seeing anything great for BD in that Chart, Clock for Clock 1100 holds it's own for the most part, and is even ahead in some of the testing.
Am I missing something?

The German's website chart? Look at the overall score and you can see FX8150 being 9-11% faster. Yeah,it's not great but this is desktop space and not many workloads are MT friendly. If you look what a 2nd generation core brought it wont be much in that respect either: 3.2Ghz 4C/8T i7 960 has 9% lower clock maximum(T) clock and is 17% slower than 2600K.
My point is that with Bulldozer they can crank clock and slightly improve the core. Both of these things will net almost linear increases in performance versus 8150. So if they launch 3.9Ghz 8170 it will be ~8% faster than 8150. If they launch 4.2Ghz PD with slightly improved cores it will be up to 20% faster than 8150. For a small tweak in design this is quite a big boost in desktop application performance and this is thanks to bulldozer design decisions.

[muzz]

I think their decisions are a failure for what the AVERAGE desktop person is looking for, with the caveat being TODAYS workloads.
Forget about tomorrows software, by the time the AVERAGE software is available that will make better use of it, there will be 3 newer generations of chips out there to handle that.
Now MAYBE it is a step in the right direction(pretty risky, and certainly not guaranteed) and their upcoming decisions will offer derivatives that offer this/ or that, but unless I'm missing something, todays software still seems to be largely dependent on IPC/single threaded, and it doesn't do well at all in that scenario, certainly not well enough to justify it over a SB(or their new lineup), or even a PH11 for folks that will need to upgrade their board.
I personally see it as a serious disappointment, and I'm a pretty big fan of AMD.

[informal]

Yes,for desktop. Desktop favors few(er) "faster" threads (clock+IPC). BD has clock,lacks on IPC. Clock can be further increased and IPC can be also increased. Tempo of doing these things will dictate the level of competitiveness of AMD.
No for server. (no being it's not a disappointment since Cray wouldn't be building a supercomputer based on Interlagos,would they?)

[muzz]

I don't care about Cray, I care about the health and competitiveness of AMD in the desktop market, or else the Evil sister will be able to do what and for whatever price they want, and NONE of us want that(even the Intel fans).
We all need an at least somewhat competitive AMD(at least for the $) to keep them somewhat honest, and I'm hopeful for that to come to fruition in the upcoming generations.
I've only built one PC using an Intel product, and that's only because that was what they gave us in school (part of the cost of school- we had no choice)..........that was in 1999
That chip was soon changed out for a faster AMD chip when I got the machine home.

Come on AMD!