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 ...

 

[sarinaide]

I didn't understand that move. They could have made more money offering AMD certification or something. Let the memory vendors deal with the overstock.

The performance edition sticks are just too slow right now. They need 2133/1866 at least.

http://www.newegg.com/Product/Product.aspx?Item=N82E16820103006

They max out at 1600.

The AMD branded RAM has three makes the RADEON BE RAM yet to come out is native 1866 but every single one of these modules has a free overclock to a SPD higher on a mere voltage tweak and no adjustment needed to latencies.

Well socket unification is one thing AMD needs with FM1, FM2, AM3+ all selling in the mainstream market at the same time.

Well FM2 is confirmed like AM3+ will likely be to run until the end of the current roadmap. After excavator the word out is unification of socket and first pure fusion chip, that is high end x86 processor with highest end Radeon graphics on a socket to be announced later, after excavator AMD will be fully unified and fully HSA.

Its funny that whenever something is said on AMD's side its marketing, yet when Intel says 2x better iGP and 15% IPC its believable. Not only have intel never been able to improve IGP by that margin in a single evolution, on the IPC front I have been hearing IB issues manifesting again. I do however think both will be improved on the intel front but I think the margins will be rather modest yields.

more players going HSA.....this will become bigger news over the next 6 months to a year.

 

[esrever]

I wonder how much Intel is going to improve IPC with their new design i hear 15-20% if so that will show Amd up even more which actually makes me sad.

expect performance of haswell over ivybridge is only 10%. The entire haswell CPU is built for mobiles and lower power consumption so there really isn't going to be a 15-20% ipc gain. Intel is pretty much going to push haswell with crystalwell as the premium ultrabook solution with better battery life and twice the gpu performance as ivybridge. There has not been much talk about the CPU at all.

 

[sarinaide]

expect performance of haswell over ivybridge is only 10%. The entire haswell CPU is built for mobiles and lower power consumption so there really isn't going to be a 15-20% ipc gain. Intel is pretty much going to push haswell with crystalwell as the premium ultrabook solution with better battery life and twice the gpu performance as ivybridge. There has not been much talk about the CPU at all.

I sense as much Intel mobility graphics is pretty decent and they are strong on that front, on the DT I don't see a IGP anywhere close to Skyrim maxed full at 1080 though. Consider HD4000 is around 4 FPS in 3DM 11, the 7660D around 11FPS and a GTX 560ti around 20FPS, considering a GTX 560ti barely plays skyrim fully maxed at 1080 res I really don't see Intel coming close, if anything AMD's Kaviri based APU's are looking very promising breaking into legitimate mainstream gaming market. Evolved X86 cores, and a full GCN mainstream GPU, that will be a massive jump forward considering how much more performance a GCN part gives over VLIW

 

[de5_Roy]

i've been following haswell news for a while and have not yet seen any mention of 15-20% ipc improvement from the rumor/information sources i consider credible. afaik, at said that the cpu performance improvement will be in 'lower double digits'. if there is a talk of 15-20% improvements, it is very likely to be a maximum, in a single case or improvement of core i7 (ht) vs i5 (non ht) on some applications.
power consumption might go down, power efficiency might go up... again. intel tried to drill that into people's heads in the latest idf.

@blackkstar: you described amd's scorpius platform i.e. combination of amd fx cpus with 970/900x/990fx chipset motherboard etc.

@22nm production talk: do lga1155 ivy bridge xeons count? those seems to have been available since q2. intel launched sb server (the not-lga1155 cpus) lineup a year after sb desktop. this is might happen with 22nm cpus as well. it's not like intel is under serious pressure in the server sector. 😗

 

[amdfangirl]

I wonder what happened to AMD's sideport RAM. If they replaced it with GDDR5, can you imagine how well it would perform?

 

[sarinaide]

What does a FX 8150, 1100T and A10 5800K all have in common? When paired with a HD 7970 GHZ edition GPU it hits 83FPS fully maxed BF3 at 1080P.....and they said it had no game in it. 😛

I am planing on selling every power rig I have left, investing in a higher end FM2 platform and focusing purely on my moddifications, I have a big box of goodies from Million Dollar PC, Frozen CPU and PPC which I gotta get to, make the hottest looking APU ever.

 

[mayankleoboy1]

I agree with this as well. Intel knows it too, it's why Haswell talk is mostly about their iGPU. If AMD was serious, they could definitely catch up to Intel. Intel isn't playing the performance game anymore either.

Everyone talks about how the desktop is a dying breed, but it will never die. People need it to render, edit photos, do 3d modeling, create music, compile software, etc. Even if some of these tasks move to the cloud, not everyone is going to want to move to the cloud. There will always be a desktop market.

.

Exactly.
And this is where AMD comes with their "moar cores for less" theme. A 15% IPC improvement versus 2 moar cores. Guess which is better for the tasks you described ?

Also, this is exactly where AMD's APU's and HSA come. Lots of smaller stupider cores to do menial jobs. Works really well. Specially for rendering/3d modelling, which is FP heavy.

The more i think, the more i believe that SR will be epic win.If only they can move to atleast a 22/20nm node...

 

[mayankleoboy1]

The major softwares that are still single threaded are Web browsers. IMHO, they are the major roadblocks in the path of multithreaded software.

Even most modern games use 2-4 cores.

 

[mayankleoboy1]

The major softwares that are still multithreaded are web browsers. IMHO, they are the major roadblocks to a multithreaded software future.

 

[sarinaide]

Exactly.
And this is where AMD comes with their "moar cores for less" theme. A 15% IPC improvement versus 2 moar cores. Guess which is better for the tasks you described ?

Also, this is exactly where AMD's APU's and HSA come. Lots of smaller stupider cores to do menial jobs. Works really well. Specially for rendering/3d modelling, which is FP heavy.

The more i think, the more i believe that SR will be epic win.If only they can move to atleast a 22/20nm node...

more cores doesn't equate to IPC, that relates to metal level, while PD is a overhaul SR is something really different and it seems like x86 performance is going to be a massive improvement. But as before AMD is going the CPU/GPU paralellism route, with many gaming and software writes going that route as well we have seen the potential a low spec APU can deliver when the GPU does the heavy FP calculations. As I said before AMD is going to unify the A-Series and FX after Excavator on a yet to be announced platform, but if SR A-series is encorporating mainstream GCN architecture, Excavator is rumored to run a mid-high enthusiast level Radeon core then reason suggests that after that AMD will have a full high end heterogeneous chip well catered for the general movement in computing, HSA is becoming real and this is where AMD needed to lock it out, when they decided to follow the Fusion is future mission they basically knew back in 06 they were going to abandon the pure x86 battle but its much bigger than that.

The theory is sound but its about execution from now on in.

 

[noob2222]

Last I heard it was April 2013 for the desktop parts, which is early Q2. And considering the ton of sku's for Ivy parts, in plentiful supply, and the 4th fab nowhere near capacity, all that indicates their 22nm is better than expected, perhaps better than their 32nm node which was the best in the industry..

Haswell will be released to market before they start to ship their 22nm server cpus based on Ivy bridge? doesn't sound too exciting.

If 22nm was above expectations, they would show up this year.

No, server parts have to be extensively tested and validated which is why they show up later, at least for Intel.

sorry thought I put the link in there.

http://www.xbitlabs.com/news/cpu/display/20121001215924_Intel_to_Boost_Number_of_Cores_Inside_Next_Gen_Xeon_Core_Extreme_Microprocessors.html

 

[sarinaide]

Lol 30mb L3, thats going to ramp up the TPD ratings, on DT I can see Intel abondoning the dieshrink philosphy, they are going to hit the single threaded performance brick wall and then what?

 

[de5_Roy]

AMD Launches Z-60 APU (Hondo) for Windows 8 Tablets
http://www.bjorn3d.com/2012/10/amd-launches-z-60-apu-windows-8-tablets/
no it's not related to piledriver or trinity.
edit: however, tablet arena is one where amd can make money, if they put out the right chip. one of the biggest achievement of hsa is going to be building a powerful and power efficient tablet/smartphone chip.

 

[sarinaide]

Fighting really hard on a budget well less than competitors, they have done well on that front the last year or so.

 

[truegenius]

T = IPC * Clockspeed * Number of cores * scaling factor

instead of calling it ipc, we can call it general relative ipc ( 😛 though not related to general relativity theory i think ) or performance per clock (ppc) 😗

i remember that i too stated/applied a formula/equation (with relative ipc) like this or maybe i am not feeling well to do a thorough reread on previous pages or maybe i dreamed of that. though not sure about that 😀
😗 looks like i need to backoff dreaming or i should increase omega3, vitamin B (all) and iron intake 😗

 

[sarinaide]

http://www.kitguru.net/components/cpu/jules/can-intels-haswell-beat-amds-apu-in-2013/

It's more a case will Haswell (i7 ) beat Llano in 2013, maybe it will be on parity or slightly ahead, but around the same time as Haswell Kaviri will be released, Kaviri APU's will use mainstream level integrated graphics in the form of HD7700 GCN architecture which is faster all round for CPU/GPU accelerated performance added to that mainstream level gaming at 1080 reslolutions, by Excavator I don't think Intel will be making IGPU's anymore 😛

I always looked at APU's with a glance of this is really low end but having gotten the opportunity to use them, and see others test them, reading the results Trinity was a massive step forward in heterogenous computing, ignoring x86 where AMD have shifted focus from, in every other aspect the Trinity not only exceeded performance of similarly priced parts, but also made near $400 parts cry mercy a big feat for what is a low scale x86 processor with low end IGPU.

Sleeping Dogs, BF3, Skyrim, Shogun 2, F1 2012 were among the games tested in IGPU and Discrete modes, the IGPU performance was a obvious landslide with Intel parts barely offering double digets at 1080 res, the Trinity at that res is playable in practically all the titles albeit on nothing more than medium settings but this is determined game to game these are all graphically challenging titles. On to discrete and only in Skyrim did a i3 and intel chips develop a decent margin over the A10 using a 7950 and 6970 I don't sense much change using Nvidia cards so I never got to that.

Overall I think a trinity APU goes beyond the "mere budget level" stigma that llano befell, all around it is a faster chip and it shows AMD are not that far from delivering mainstream level performance in A-series trim.

 

[jdwii]

more cores doesn't equate to IPC, that relates to metal level, while PD is a overhaul SR is something really different and it seems like x86 performance is going to be a massive improvement. But as before AMD is going the CPU/GPU paralellism route, with many gaming and software writes going that route as well we have seen the potential a low spec APU can deliver when the GPU does the heavy FP calculations. As I said before AMD is going to unify the A-Series and FX after Excavator on a yet to be announced platform, but if SR A-series is encorporating mainstream GCN architecture, Excavator is rumored to run a mid-high enthusiast level Radeon core then reason suggests that after that AMD will have a full high end heterogeneous chip well catered for the general movement in computing, HSA is becoming real and this is where AMD needed to lock it out, when they decided to follow the Fusion is future mission they basically knew back in 06 they were going to abandon the pure x86 battle but its much bigger than that.

The theory is sound but its about execution from now on in.

How is Steamroller that different? It's still a module that shares resources just more efficiently.

2 decodes per module vs 1
dynamic L2 cache(which sounds like it saves power not any faster)
28nm vs 32nm
L1 cache is now bigger(should've never been so small on bulldozer vs Phenom)

http://hothardware.com/News/AMDs-Next-Gen-Steamroller-CPU-Could-Deliver-Where-Bulldozer-Fell-Short/

Honestly i'm expecting a bigger improvement than Piledriver but not as much as people are thinking lets not hype steamroller. Since Steamroller is going to be made on 28nm we can probably expect a small improvement in power consumption as well plus they're improving their FPU a little for power consumption and space, i'd say Steamroller will be a bigger improvement in both areas(power consumption+Performance) then piledriver was but not as much as people are thinking.

With the talk of HSA maybe steamroller and the AM3+ will be the last CPU only platform? I'm not expecting much from Amd on CPU performance until Excavator especially when Steamroller is going to use AM3+ not to mention this should give jim keller enough time to put some work into Excavator as well since steamroller is most likely done and headed at samples soon to get ready for a 2Q 2013 launch on APU's.

 

[sarinaide]

More instructions per cycle, less branch miss predictions, lower latencies. If we speculate numbers some mention of 20% over Vishera, though we already know that AMD are dedicating to 15% per evolution, some in AMD are targeting more.

Yes SR samples will be out January time, A collegue of mine is privy to such I guess it comes with having well over 100k subscribers, review for major websites and in the top 20 overclockers globally, add 30 years experience he is justified testing AMD products.

As to chipset, I have been led to believe that AM3+ and FM2 will run until the end of the current road map, thereafter it will be a unification of socket to be announced. But there will be chipset updates for SR and EX.

 

[jdwii]

Just with the mentioned improvements i fail to see improvements in Cache latency L1 cache is improved on this but its going to be bigger so latency should be the same at least i hear no improvements in L3 or L2 cache except dynamic L2 which is only going to improve Power consumption. Maybe their not giving us all the info.

 

[sarinaide]

Just with the mentioned improvements i fail to see improvements in Cache latency L1 cache is improved on this but its going to be bigger so latency should be the same at least i hear no improvements in L3 or L2 cache except dynamic L2 which is only going to improve Power consumption. Maybe their not giving us all the info.

With what happened with BD do you blame them for keeping things under wraps.

 

[jdwii]

With what happened with BD do you blame them for keeping things under wraps.

No not really in fact i hope they underestimate their performance figures that way people will be happier when steamroller does arrive at least that is what i would do.

 

[-Fran-]

They can't underestimate either.

It's a delicate game, that if you don't know how to play it, you better stay away or just play it by the book.

If you say it won't, but at the end it does, then you already lost sales to those that trusted your words before and went with the competition. And if you hype and don't deliver, the backslash is terrible for your product.

There is some point in the middle that AMD is pointing to with, first, the 2 tier preview and review releases and maybe they'll do something else for next CPUs. They don't have the money to cover another BD marketing fail and they're being cautious about things, which is good IMO.

That's my marketing "life" major degree at its best, hahaha.

Cheers!

 

[mayankleoboy1]

Why dont they use TSMC's 28nm HKMG process for their CPU's too ?

AFAIK, 28nmHKMG >> 32nmSoI
and their 40nm Z series chips ......*facepalm*

 

[jdwii]

7% IPC increase 🙁
http://www.youtube.com/watch?v=o5hxOsrukPY&feature=player_embedded

Where have you been LOL we ALL been saying this.

 

[fazers_on_stun]

http://www.kitguru.net/components/cpu/jules/can-intels-haswell-beat-amds-apu-in-2013/

It's more a case will Haswell (i7 ) beat Llano in 2013, maybe it will be on parity or slightly ahead, but around the same time as Haswell Kaviri will be released, Kaviri APU's will use mainstream level integrated graphics in the form of HD7700 GCN architecture which is faster all round for CPU/GPU accelerated performance added to that mainstream level gaming at 1080 reslolutions, by Excavator I don't think Intel will be making IGPU's anymore 😛

I always looked at APU's with a glance of this is really low end but having gotten the opportunity to use them, and see others test them, reading the results Trinity was a massive step forward in heterogenous computing, ignoring x86 where AMD have shifted focus from, in every other aspect the Trinity not only exceeded performance of similarly priced parts, but also made near $400 parts cry mercy a big feat for what is a low scale x86 processor with low end IGPU.

Sleeping Dogs, BF3, Skyrim, Shogun 2, F1 2012 were among the games tested in IGPU and Discrete modes, the IGPU performance was a obvious landslide with Intel parts barely offering double digets at 1080 res, the Trinity at that res is playable in practically all the titles albeit on nothing more than medium settings but this is determined game to game these are all graphically challenging titles. On to discrete and only in Skyrim did a i3 and intel chips develop a decent margin over the A10 using a 7950 and 6970 I don't sense much change using Nvidia cards so I never got to that.

Overall I think a trinity APU goes beyond the "mere budget level" stigma that llano befell, all around it is a faster chip and it shows AMD are not that far from delivering mainstream level performance in A-series trim.

Since you suddenly started quoting synthetic benchies a few posts before this last one 😀

http://www.xbitlabs.com/articles/cpu/display/amd-a10-5800k_7.html#sect0

The picture here is dramatically different from what we saw in our previous test session, when we tested processors with their integrated graphics cores. With a discrete graphics card that doesn’t cap the power of the x86 cores Trinity doesn’t look so rosy any more. Of course, it does outperform Llano, just the way it should. However, it doesn’t bring it any closer to Intel Core i3, which run much faster under gaming load. You could object that we are testing Trinity in a non-typical usage scenario, and that they will primarily be used with activated integrated graphics, for which the x86 cores performance would be quite sufficient. But we disagree. AMD obviously anticipates that their new APU will also be used with discrete graphics. Otherwise, why would they roll out A85X chipset with CrossFireX support and add processor models with the GPU disabled on the hardware level to their product line-up?

All in all, as we can see, a Socket FM2 platform with a pretty fast discrete graphics accelerator is not such a good fit for a gaming system. And the main reason for that is the low performance of the Piledriver cores in games. However, I would also like to remind you that Trinity processors do not support PCI Express 3.0 bus.