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 ...

 

[jdwii]

The problem with "fixed function logic" circuits is that when you want something a little more custom, it falls flat on its face. Inte's QS is no exception to that rule (nor is DXVA and nVP).

APUs do offer better OGL compliance, so you can use more specific GPU capabilities that Intel CPUs can't muscle out.

Try using denoise on a Celeron in a 1080p BD playback for instance. The HD2k nor HD2.5K will get even close to an A6 or A10 in that regard. Sorry to be such a videophile, but that's the truth.

And actually, not even the E350 ("Brazos 1.0") could handle 1080p + filters, but it was able to do so in 720p+filters and pure 1080p playback.

That's just for video playback and not thinking about games.

You would really need a side to side comparison of the playback of either system to believe me, but take a look here: http://www.anandtech.com/show/5906/amd-llano-htpc-builders-guide

Cheers!

Yeah i had the E-350 which my friend owns now and it can play 1080P on VLC but not youtube but its only a 1366/768 screen and it can plat some small games such as minecraft and sims 3 but it does skip here or their.

 

[-Fran-]

Zero. Video playback uses media processing power, not graphics processing. APU fans like to shout from the rooftops how great Fusion is for HTPCs. But you don't really need to use the Graphics engines to playback video.

Wow, I didn't notice the "EDIT" part...

You seem to have drank the "anti AMD" coolaid or something, because this statement is very ignorant in every way. And it's really offensive to say the least.

If you only watch videos in Youtube or Vimeo or Netflix, you *could* be correct, but that's not even the case since they handle the decoding pipeline different from regular video players. I don't want to argue this in more than 1 line, but you really need to read this:

http://www.tomshardware.com/reviews/video-transcoding-amd-app-nvidia-cuda-intel-quicksync,2839-2.html

Cheers!

 

[jdwii]

Wow, I didn't notice the "EDIT" part...

You seem to have drank the "anti AMD" coolaid or something, because this statement is very ignorant in every way. And it's really offensive to say the least.

If you only watch videos in Youtube or Vimeo or Netflix, you *could* be correct, but that's not even the case since they handle the decoding pipeline different from regular video players. I don't want to argue this in more than 1 line, but you really need to read this:

http://www.tomshardware.com/reviews/video-transcoding-amd-app-nvidia-cuda-intel-quicksync,2839-2.html

Cheers!

Lol agreed, I think even windows 8 uses OpenCL now for everything.

 

[anxiousinfusion]

There is some interesting information I came across. I had no idea that Jim Keller would take such immediate action with the planned architectures. An unnamed source (presumably an AMD engineer) claims potential gains of up to 45%! I certainly hope this is legit.

http://vr-zone.com/articles/amd-pushes-steamroller-and-excavator-forward-bullish-about-performance-increases/17088.html

 

[viridiancrystal]

imo if someone has a quad module bd, pd is okay to skip. i made my assumption based on the rumors/info that
amd will incorporate 3rd party ip - rcm (according to my ...rather limited knowledge, should work up to (may be multiple resonant freq defined by the power management for stock use) or at a certain frequency),
clockspeed has minor bump,
both bd and pd are 32 nm chips,
the high vcore amd cpus (both bd and the pd e.s. from obr screenies) use at high clockrate even at stock, etc.
use of rcm can be seen from different angles. one is that amd made a wise choice using rcm for it's effectiveness. another is that amd Failed to reign in bd's powerhogging tendencies by themselves and Had to use rcm. another one is that amd is too short on money to re-work all of the cpu so they're incrementally improving their product such as introducing modular arch, then improving perf/watt, after that may be improving imc, pcie controller, power effeciency and so on. all these seem to contribute to improving stock perf. i don't see how overclocking fits into this. i am not saying amd won't make overclockable cpus, they certainly will have unlocked cpus. i am guessing that they're leaving the oc related things up to people who will oc the cpu. that means everything that comes with overclocking is the user's responsibility including the power use, heat dissipation, cooling management. they'll probably do another pd 8 core + amd branded lcs bundle like the one with closed loop liquid cooler + fx8150.
then there's the mantra - "real gamers/enthusiasts do not care about power efficiency". 😀
amd will definitely claim credit in case anyone manages to break any ghz records.
early reviews will make it worse. i noticed (after noob poitned it out iirc) that early cpu reviews almost always use high vcore for overclocking, to see how far the cpu goes in terms of clockrate and make power efficiency verdict on that info. toms do better, with their oc efficiency analysis.

Well, think way back to Tom's Trinity power numbers. The A-10 5800k averaged out at ~105w. Clocked at 3.8Ghz, with the graphics core on there. The Fx-4170 is clocked at 4.2Ghz, and is has a TDP of 125w (couldn't find any power numbers). I'd say that the graphics in the A-10 are somewhere between the 6570 and the 6670 which both draw around 60w of power, probably lower due to the 32nm, but still rather impressive either way. I've got some confidence in PD's power draw.

 

[cgner]

Right, because Intel can automatically outperform any other company at anything they choose to do, huh? By the way, in spite of their focusing on CPUs, we still have this other mildly successful company around, which is called AMD.

As far as GPU production goes, they are not mildly successful to produce about 40% of all GPU's sold on the market. In fact, they are very successful to hold that kind of stake when only 2 companies compete.

 

[amuffin]

Well, look at the % of people that actually care about actual system specs, to the % of people that don't. Also, look at the % of people that play games, and the % of people that don't.

 

[cgner]

Well, look at the % of people that actually care about actual system specs, to the % of people that don't. Also, look at the % of people that play games, and the % of people that don't.

According to toms, 100% of people care about the specs and 95% play games. 😀

 

[skitz9417]

hey do u guys think the pd cpu will better than the intel haswell cpus?

 

[amuffin]

hey do u guys think the pd cpu will better than the intel haswell cpus?

No.

 

[nsouter853]

There is some interesting information I came across. I had no idea that Jim Keller would take such immediate action with the planned architectures. An unnamed source (presumably an AMD engineer) claims potential gains of up to 45%! I certainly hope this is legit.

http://vr-zone.com/articles/amd-pushes-steamroller-and-excavator-forward-bullish-about-performance-increases/17088.html

May I please have my grain of salt?

 

[viridiancrystal]

May I please have my grain of salt?

I feel like that is something one would write while intoxicated.

 

[davemaster84]

hey do u guys think the pd cpu will better than the intel haswell cpus?

No, and if I aint wrong Haswell will be released the same time as steamroller or even excavator . But I hope PD will be a match for SB and therefore a price - competitive option against Intel.

 

[de5_Roy]

Well, think way back to Tom's Trinity power numbers. The A-10 5800k averaged out at ~105w. Clocked at 3.8Ghz, with the graphics core on there. The Fx-4170 is clocked at 4.2Ghz, and is has a TDP of 125w (couldn't find any power numbers). I'd say that the graphics in the A-10 are somewhere between the 6570 and the 6670 which both draw around 60w of power, probably lower due to the 32nm, but still rather impressive either way. I've got some confidence in PD's power draw.

you are right about trinity's power consumption. desktop 6670 uses around 66w, ddr3 version. i did like trinity's stock power figures. i was slightly disappointed that toms didn't test overclocking of the k series apus. i also didn't find any info that refuted claims of design issues with trinity motherboards or trinity having a new stepping. this is all speculation on my part.
toms could have tested if the igpu was inhibiting cpu performance. speculation: if the igpu indeed shares some of the cpu performance (because both are under the same 100w tdp), piledriver arch. hasn't delivered it's full potential yet, considering improvements.
even after saying these... nothing about overclocking-related stuff.
edit: iirc amd does not measure tdp using worst case scenario like intel does. so at full load and stock settings, 4170 might use around 130w or slightly higher or lower.

 

[jaguarskx]

No, and if I aint wrong Haswell will be released the same time as steamroller or even excavator . But I hope PD will be a match for SB and therefore a price - competitive option against Intel.

Sorry, but that's not going to happen. Phenom II / FX is basically equal to Intel's Core 2 Duo / Quad CPUs. Sandy Bridge CPUs are about 23% more powerful than C2D / C2Q so PileDriver will have to more or less squash AMD's promise of 5% - 15% performance increase with every new CPU. Even 15% would be very difficult. I believe a 10% performance improvement is a more realistic estimate of what PD's performance will be.

AMD can't afford to hit the low end of their performance scale. A 5% increase is marginally better than a "side-grade".

 

[cgner]

hey do u guys think the pd cpu will better than the intel haswell cpus?

nay.

 

[sarinaide]

Our dear friend OBR at it again, this time just echoing the sentiments many feel about his "FX 8350" benches. He responded with a equally dimwitted attempt at salvation calling the FX 8350 a BD core debug and not a PD core in the sense that Trinity and Abu Dhabi is. He claims that AMD accused him of making things up, I can assure him that AMD does not bother making contact with him and second AMD about 3 months ago released a official memorandum of all metal level changes to the BD architecture to be employed in all PD cores obviously he would have known that if he didn't make stuff up.

He is punting results so he can point back later if its close enough and proclaim himself a messiah to the people. Since he is not objective his points are moot and have no credibility at all. Just pure hate mongering.

While most will follow PD intensively notably the Trinity and Abu Dhabi parts most AMD users are looking towards the SR release.

 

[truegenius]

Hindi language words

then there's the mantra

magic spell

proclaim himself a messiah to the people

superhero

edit: iirc amd does not measure tdp using worst case scenario like intel does. so at
full load and stock settings, 4170 might use around 130w or slightly higher or
lower.

paladin vs jimmy 😛

Lol agreed, I think even windows 8 uses OpenCL now for everything.

damn ! my current sapphire HD6770 does not support openCL

 

[gamerk316]

And in other news:

http://semiaccurate.com/2012/09/04/microsoft-xbox-next-delay-rumors-abound/

I'm calling a major fiasco on the Xbox Next right now. No real issues with the AMD CPU [makes sense for MS if they want to unify their OS], but I'd imagine the 7000 series GPU is going to KILL yields at launch.

I also note that this more or less means the console WILL be running Win8, and will basically be a dedicated, unupgradable mini-PC. On the plus side, porting should be ridiculously easy.

 

[de5_Roy]

rage3d rages about amd performance cpu. apparently he didn't get the memo...
http://www.rage3d.com/articles/kombuting/2012_sept1/
amd's own widi
http://semiaccurate.com/2012/09/05/amd-is-catching-up-in-the-wireless-display-game/

 

[sarinaide]

Another idiot that thinks improvements are linear. That if AMD improve 50% by steamroller Intel conversely do the same. A small change in architecture could bring a much bigger gain so as to make the minimum targets seem obselete, the honest truth is nobody will know, but AMD have that target in mind bear in mind its a target and not what may actually be. Similarly by Broadwell Intel may only be 15% better than IB so it is difficult to actually say until it happens.

What I will say with 100% certainty is its another first prized chop of a fanboi to join the list of fanbronies.

 

[fazers_on_stun]

A surprising amount of enthusiasts have a vendetta against Intel graphics that originates from some time before Sandy Bridge was released. Even now, the standard answer the Intel's increasingly powerful iGPU is that the Intel's graphics drivers are terrible. This is no longer the case. There was a time when Intel graphics (hardware and driver) were almost unuseable, but since then Intel has produced a great iGPU solution.

Most who complain about Intel graphics have not used them on Ivy Bridge, and quote lackluster high resolution gaming benchmarks as proof that HD graphics are not suitable for even average users. However, the budget Intel CPUs provide a very capable CPU/GPU combination for the average user with enough 3D graphics power for casual gaming, and unmatched media performance for HTPCs.

Intel graphics have already more than doubled in processing power since Nehalem with much more driver stability. There's nothing that indicates that this trend will not continue with Haswell and beyond.

Either vendetta or wishful thinking IMO 😛.. I get the feeling a lot of AMD fans hope Intel always remains behind in graphics, leaving just NV as the main competition.

I forget whether it was you or Cazalan who had the discussion on this thread about TSV interposers and Haswell's use of stacked memory for the iGPU, but S/A has a somewhat decent article explaining the benefits and problems with chip stacking: http://semiaccurate.com/2012/09/05/hot-chips-talks-all-about-chip-stacking-good-and-bad/

 

[fazers_on_stun]

Just hard to see how a company with no GPU core experience can produce a chip that will all of a sudden be at a "monster level" which to me sounds like something that will be seen in SteamRoller using a state of the art GPU core design.....

IIRC Intel has hired away some good graphics people over the last few years, so it's not like they are total newbies groping in the dark 😛.

That 7x brought it closer to Llano but there were still behind by a decent degree. If Intel get it right then I will happily say well done but I think to assume "Monster" without context is rather ambitious. Monster for all we know could mean llano in Intel terms.

IIRC that "7X" was Haswell vs. Sandy Bridge, which would mean maybe 3.5X times Ivy Bridge. So a quick & dirty comparison would be to take some Trinity vs. IB benchmarks and then multiply IB's by 3.5 and recompare.

Of course, nothing is guaranteed and obviously we need to wait for the actual released product reviews to come out, in 6 months or so. IMO however Intel has come closer to their predictions than AMD has, in the last 4-5 years.

BTW, didn't IB's iGPU compare more favorably when compared to Llano's or Trinity's at the 17W power level? I think this is where the biggest growth area will be - 17W or lower - in the next few years..

 

[cgner]

Another idiot that thinks improvements are linear. That if AMD improve 50% by steamroller Intel conversely do the same. A small change in architecture could bring a much bigger gain so as to make the minimum targets seem obselete, the honest truth is nobody will know, but AMD have that target in mind bear in mind its a target and not what may actually be. Similarly by Broadwell Intel may only be 15% better than IB so it is difficult to actually say until it happens.

uh huh. 100% gain is as easy as a walk int he park. That's how AMD achieved 500% improvement over phenom II in just 3 years of developments.

 

[fazers_on_stun]

http://www.rawrecipebox.com/wp-content/uploads/2011/03/large-salt-pile-792875.jpg
I feel like that is something one would write while intoxicated.

Heh, that's Theo Valich for ya 😛..

Wasn't he the guy who proclaimed AMD engineers were "dancing in the aisles" some 3 months before Barcelona came out in Sept. 2007? Or am I mixing him up with Charlie D?