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 ...

 

[lilcinw]

Yeah, I believe it will be termed an HD 7xxx but is based on VLIW4.

 

[viridiancrystal]

I would assume just the arch, no way are they putting a full cayman on there. Too bad, if they put some 28nm gpu's it would help offset the power difference from using Piledriver over stars.

 

[mlscrow]

I would assume just the arch, no way are they putting a full cayman on there. Too bad, if they put some 28nm gpu's it would help offset the power difference from using Piledriver over stars.

Exactly right. I did say 6900 "tech", in my defense. From other sources, people believe, according to what they can physically see on the picture of the Trinity that was leaked that it will use "six VLIW4 clusters for a total of 384 Radeon cores".

 

[esrever]

they have 400 radeon cores on llano now, why would they lower them to 384 given the fact that by using bulldozer they are freeing up die area?

 

[mlscrow]

I have to admit that I pulled the trigger on a i5-2500K and Z68 mobo as an upgrade for my i5-760 P55 set-up.

I'm keeping my 965BE and 990XA and patiently awaiting any improvements from stepping revision, possible FX-4170 and/or Piledriver.
since the Intel is my gamer unit and the AMD is my daily unit I'm keeping it regardless
as it would be nice to see AMD come thru and significantly improve what they have now.
windows 8 and it's core scheduling isn't going to be enough..

It appears that I have been in a very similar position to you with regard to current and desired hardware. I currently have a 955 C3@4GHz right now and have been eyeing the 2500K, just haven't pulled the trigger yet. The fact that it puts all current AMD chips to shame, is a fantastic overclocker (what do you have yours too btw? Surely over 4GHz, but how much farther if you don't mind me asking?) as well as the fact that Piledriver is so far out (Q3?) is pushing me toward the 2500K, but I keep telling myself to wait to see just how well Trinity performs before I make that call.

This is not to say that I think Trinity's performance is what I should be basing my expectations of the non-APU version of Piledriver on, especially seeing as how trinity will be made with 32nm process and supposedly the Piledriver will be 28nm, but it will at least give me enough information to make some assumptions as to whether or not I should wait for Q3 or not.

As far as the W8 Scheduling not being enough, couldn't agree more. It's nice that we can expect 5-10% improvement with that alone, but it has to dish out at least 20-25% more on top of that if it's going to sell, at least in my opinion. I'm holding on to the hope as well. Although, as much as I like AMD for the fact that they provide competition and keep prices down for us consumers, I'm honestly not that worried about the market if they do happen to go under as I feel that Nvidia and ARM are eventually going to be a figure to contend with. And just hypothetically, imagine if NVidia were to one day purchase AMD? That's a fun idea to me.

 

[mlscrow]

they have 400 radeon cores on llano now, why would they lower them to 384 given the fact that by using bulldozer they are freeing up die area?

This wasn't a for sure, 100% fact, it's just speculation, however, to potentially answer your question,

"The graphics in Trinity will be based on AMD's VLIW4 architecture that is used in the HD 6900 series and the Cayman circuit. According to the picture above it looks like it will get six VLIW4 cluster for a total of 384 Radeon cores, which is less than Llano's 400 cores that uses five VLIW5 clusters. The VLIW4 architecture is more efficiency and together with support for higher memory frequencies and overall higher frequencies we should expect better graphics performance from Trinity than Llano. Trinity also supports the dedicated H.264 decoder VCE, just like the graphics cards of the HD 7000 series and the new GCN architecture."

The article over at SemiAccurate is the source (not sure if I'm allowed to post links to other sites...I didn't read the "rules and guidelines" for these forums yet.
I know, I know.

 

[mlscrow]

I feel your pain bro.
easy clock and not trying to compete with some of these overclocking clowns, I'm running the i5-2500K @ 4.1GHz stock voltages.
unless AMD changes the plans and releases and AM4 socket and not convert everything to FM2, AM3+ is the last.

Stock voltage! Now that is nice. I'm not particularly a fan of overvolting. I'll do it a tad (1.4125v for my 4.0GHz 955, which, due to Vdroop, only goes up to exactly 1.4v under load in CPU-Z, not bad for a C3), as long as it doesn't create an unreasonable amount of increase heat, but yeah, I'm not out to break any records either. I like having a cool and stable build.

 

[dogman_1234]

I thought they cancelled Trinity?

 

[bawchicawawa]

Anyone know some information on the win 7 update? I thought it was going to come out yesterday... =\

 

[bawchicawawa]

Clever, anyways... about the update?

 

[mlscrow]

I thought they cancelled Trinity?

No, they cancelled Wichita and Krishna, not Trinity.

 

[mlscrow]

Clever, anyways... about the update?

No idea on the Windows 7 patch. This is, however, and as you know, a thread about Piledriver. You may find better results in a Bulldozer thread as those who have purchased them will be the most inclined to find that information.

 

[dogman_1234]

No, they cancelled Wichita and Krishna, not Trinity.

Gottcha!

 

[-Fran-]

If what Semiaccurate said is correct, using VLIW4 384 SP's make me have a bad feeling...

They lowered the TDP on the Video side of things in favor of what exactly? I really hope its for the overall TDP of the APU, but my guts are telling me it's for the PD core and it's Turbo Core jumps that aren't that far away from the BD cores. I really really hope I'm wrong here >_<

Any other interpretation of that?

Cheers!

 

[mlscrow]

Clever, anyways... about the update?

I believe this is what you were looking for, no?

Bulldozer Windows 7 Patch that just came out today.

http://www.overclock.net/t/1198453/official-fx-windows-7-patch-s-released

Here are some results from someone who benchmarked:

Passmark 7 8T = Old Patch = 1.9% increase, New Patch 2.1% increase. (it doesn't have any of the performances drops on some tests like the old patch)
Passmark 7 4T = Old Patch = 5.3% increase, New Patch 11.5% increase.
Passmark 7 2T = Old Patch = 5.8% increase, New Patch 12.1% increase.
Passmark 7 1T = Old Patch = 0.3% increase, New Patch 0.7% increase.

Cinebench 8T = Old Patch = 0.01-0.02 points better, New Patch = 0.04-0.05 points better.
Cinebench 4T = Old Patch = 0.05-0.06 points better, New Patch = 0.09-0.10 points better.
Cinebench 2T = Old Patch = 0.05-0.06 points better, New Patch = 0.09-0.10 points better.
Cinebench 1T = Old Patch = 0.01-0.02 points better, New Patch = 0.01-0.02 points better.

3D Mark 11 Physics Score = No Patch 4.7ghz 7943, Old Patch 7981, New Patch 8025

Primebench mark, 4.5ghz No Patch 9.199, Old Patch 9.144, New Patch 9.128

 

[mlscrow]

If what Semiaccurate said is correct, using VLIW4 384 SP's make me have a bad feeling...

They lowered the TDP on the Video side of things in favor of what exactly? I really hope its for the overall TDP of the APU, but my guts are telling me it's for the PD core and it's Turbo Core jumps that aren't that far away from the BD cores. I really really hope I'm wrong here >_<

Any other interpretation of that?

Cheers!

Obviously the actual specs have yet to be released, but I'm personally not that worried, as even with a lower amount of SP's, they're more efficient and provide better performance overall. There were some potentially legit (equally potentially illegitimate) benchmarks of Trinity's performance over Llano and CPU wise, Trinity was ~13% faster than Llano and GPGPU wise, it was 54% faster. If those benchmarks are legitimate, then yeah, nothing to worry about.

However, we haven't heard the fat lady sing yet.

 

[viridiancrystal]

I really hope trinity does well. For someone that, if the chance came, has to build on a tight budget, Apu's seem like the best way to go, and really deliver.

 

[gamerk316]

I believe this is what you were looking for, no?

Bulldozer Windows 7 Patch that just came out today.

http://www.overclock.net/t/1198453/official-fx-windows-7-patch-s-released

Here are some results from someone who benchmarked:

Passmark 7 8T = Old Patch = 1.9% increase, New Patch 2.1% increase. (it doesn't have any of the performances drops on some tests like the old patch)
Passmark 7 4T = Old Patch = 5.3% increase, New Patch 11.5% increase.
Passmark 7 2T = Old Patch = 5.8% increase, New Patch 12.1% increase.
Passmark 7 1T = Old Patch = 0.3% increase, New Patch 0.7% increase.

Cinebench 8T = Old Patch = 0.01-0.02 points better, New Patch = 0.04-0.05 points better.
Cinebench 4T = Old Patch = 0.05-0.06 points better, New Patch = 0.09-0.10 points better.
Cinebench 2T = Old Patch = 0.05-0.06 points better, New Patch = 0.09-0.10 points better.
Cinebench 1T = Old Patch = 0.01-0.02 points better, New Patch = 0.01-0.02 points better.

3D Mark 11 Physics Score = No Patch 4.7ghz 7943, Old Patch 7981, New Patch 8025

Primebench mark, 4.5ghz No Patch 9.199, Old Patch 9.144, New Patch 9.128

About as predicted then; I'm not shocked the 2T/4T got the bulk of the optimization, considering how the changes were implemented. Still inferior to SB though.

 

[piesquared]

I really hope trinity does well. For someone that, if the chance came, has to build on a tight budget, Apu's seem like the best way to go, and really deliver.

That's a good point. They don't offer outrageous performnce in the traditional CPU sense, but they offer something better. Gives customers a noticeably better user experience by allowing them to do more things in a more productive way. There's no tangible benefits to a customers user experience by just running the same thing faster than they did before. And as you say, it is affordable to even those on a tight budget.

Just to go off on a tangent, GPGPU is a game changer it opens up a whole new set of possibilities. Some dismiss it nonchalantly, it seems because they stand something to lose, but there is a lot of programming going on in the backround for GPGPU computing. That's proven by the big increase in GPGPU developer activity on developer forums around the internet. There may not be 500,000 apps currently, it just got rolling, but there is enough to produce a number of killer apps and that will cause an immediate paradigm shift due to it's huge performance potential. Of course there's a final piece of the puzzle, and that is ARM and it's massive volume. If interested refer to AFDS11.

Hope it's ok to post about Trinity in this thread, I couldn't find it's sticky. Noticed the Haswell/IVB thread, but no mention of the products direct competitors anywhere...

 

[Chad Boga]

That's a good point. They don't offer outrageous performnce in the traditional CPU sense, but they offer something better. Gives customers a noticeably better user experience by allowing them to do more things in a more productive way. There's no tangible benefits to a customers user experience by just running the same thing faster than they did before. And as you say, it is affordable to even those on a tight budget.

Just to go off on a tangent, GPGPU is a game changer it opens up a whole new set of possibilities. Some dismiss it nonchalantly, it seems because they stand something to lose, but there is a lot of programming going on in the backround for GPGPU computing. That's proven by the big increase in GPGPU developer activity on developer forums around the internet. There may not be 500,000 apps currently, it just got rolling, but there is enough to produce a number of killer apps and that will cause an immediate paradigm shift due to it's huge performance potential. Of course there's a final piece of the puzzle, and that is ARM and it's massive volume. If interested refer to AFDS11.

Hope it's ok to post about Trinity in this thread, I couldn't find it's sticky. Noticed the Haswell/IVB thread, but no mention of the products direct competitors anywhere...

You should post more often, your posts are a joy to read. 😀

 

[palladin9479]

That's a good point. They don't offer outrageous performnce in the traditional CPU sense, but they offer something better. Gives customers a noticeably better user experience by allowing them to do more things in a more productive way. There's no tangible benefits to a customers user experience by just running the same thing faster than they did before. And as you say, it is affordable to even those on a tight budget.

Just to go off on a tangent, GPGPU is a game changer it opens up a whole new set of possibilities. Some dismiss it nonchalantly, it seems because they stand something to lose, but there is a lot of programming going on in the backround for GPGPU computing. That's proven by the big increase in GPGPU developer activity on developer forums around the internet. There may not be 500,000 apps currently, it just got rolling, but there is enough to produce a number of killer apps and that will cause an immediate paradigm shift due to it's huge performance potential. Of course there's a final piece of the puzzle, and that is ARM and it's massive volume. If interested refer to AFDS11.

Hope it's ok to post about Trinity in this thread, I couldn't find it's sticky. Noticed the Haswell/IVB thread, but no mention of the products direct competitors anywhere...

Pretty much agree. But a point about GPGPU and trying to run generic logic on a GPU. GPU's are basically weak VLIW SIMD CPUs with enormous core counts. Due to the fundamental difficulties with VLIW, its very bad to run generic logic operations on. But, the fundamental strengths of VLIW make it amazing to run large quantities of vector math on, calculations that would take a hundred cycles on a generic CPU can instead be done in a half dozen cycles on a VLIW CPU with the width of a modern GPU. Basically it's really good at anything that can run large math operations in parallel, large arrays, fractals, primes and various logarithmic operations. Of course it's bad at comparisons / memory ops / moves and such. Makes for an interesting co-processor and definitely worth investing money / time to create compatible software engines for.

 

[gamerk316]

Pretty much agree. But a point about GPGPU and trying to run generic logic on a GPU. GPU's are basically weak VLIW SIMD CPUs with enormous core counts. Due to the fundamental difficulties with VLIW, its very bad to run generic logic operations on. But, the fundamental strengths of VLIW make it amazing to run large quantities of vector math on, calculations that would take a hundred cycles on a generic CPU can instead be done in a half dozen cycles on a VLIW CPU with the width of a modern GPU. Basically it's really good at anything that can run large math operations in parallel, large arrays, fractals, primes and various logarithmic operations. Of course it's bad at comparisons / memory ops / moves and such. Makes for an interesting co-processor and definitely worth investing money / time to create compatible software engines for.

Exactly. GPU's are basically really, really weak CPU's with very large core counts. Great at anything that can be parallelized [Rasterization and Physics computations are both great examples].

That being said, given how GPU's vary wildly in performance, and the fact they are typically overworked when gaming, and you will see some resistance to writing software that relies on a powerful GPU to do large amounts of work.

 

[-Fran-]

Exactly. GPU's are basically really, really weak CPU's with very large core counts. Great at anything that can be parallelized [Rasterization and Physics computations are both great examples].

That being said, given how GPU's vary wildly in performance, and the fact they are typically overworked when gaming, and you will see some resistance to writing software that relies on a powerful GPU to do large amounts of work.

hahaha, you made me remember Java in it's beginning. The JVM running boggus/sloppy, a not so friendly / basic API (all prior 1.5, off course) and performance on the floor compared to the finest C++ code you could have/program/compile/run. Not even top performance HW could make it shine, lol. Now it's more close thanks to a lot of optimization and work within the JVM, but still is slower to the C/C++ finest.

I think we're still seeing immature APIs for GPGPU and we'll need a few more years to really have programmers digging around it (just look at OGL; it's still a PITA to code for it compared to DX and look what's the defacto API for games nowadays). More libs and more importantly, matured libs. Oh well, that's a whole topic all by itself, but it's interesting to do a lil' OT to chat about it 😛

So... The GPUs will become the 2013's math coprocessor of the 90's xD

Cheers!

 

[viridiancrystal]

It all relates back to optimization. I say it all the time. Everyone knows how easy it is to run Valve games. Its not like they look bad, or have a N64-like polygon count (portal 2 looked amazing). It is just very well optimized.

Back to GPGPU, all computers have to have some kind of GPU, even if you optimized what you could, however low end, it would make the computer seem all that faster to the average consumer who has no clue why it was quicker, but they would be happy about it.

Also (don't hold me to this, i'm no programmer) there must be some way to program software to optimize to as many cores are present. Photoshop uses 8 or more cores if they are present, but it doesn't run like crap if you don't have all of them.

 

[gamerk316]

^^ Here's the problem: The thread is you lowest unit of execution. And most times, you only have one or two threads that do any real heavy lifting at any one time. That limits how well you can scale your software. I've always seen more cores as a way to do more APPLICATIONS at one time, but simply do not see software scaling much beyond four cores, except in some very special cases. Todays games may have 80+ threads running, but only a handful are doing any real work.

Hence my primary argument of why BD was destined to fail: for software like you see on a PC, its simply far too difficult to scale efficently. Once performance is "good enough", designers stop trying to optimize farther, because its a waste of time and money. At the end of the day, very few people (if any) buy a game because it spits out 120 FPS on a 460; they buy it because they want to play it.

That being said, I consider the next leap forward in gaming to be implementing real-time physics effects into games, which almost requires a dedicated GPGPU. [Multiple-object dynamic physics equations get very complicated very fast]. For example, instead of set values for bullet damage, I want the effects computed in real-time based on distance, velocity, armor, etc.