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 ...

 

[de5_Roy]

what about a test between i3 with 1 dissabled core (if possible) and a8 with 2 disabled cores, so as to made them a 2t vs 2t match.
This will simiulate that how will they perform in more threaded games (which may appear in future).?
With the charts of cpu and gpu usage.

I think a8 2c@2.6 will be equal to i3 1c@3.1.

So what you guys think!

i'd suggest looking up mp3 encoding benches from toms cpu review articles. iirc those were single threaded (dual? idk) e.g. itunes and lame encoding.
from gallardo's link:
http://www.xbitlabs.com/images/cpu/amd-a8-3870k/itunes.png

 

[palladin9479]

and what about non gaming use of i3-2100 and a8-3870k.
I3 can't stand against a8 in multitasking when using all 4 threads effectively
a8@3 seems to perform ~20% better than i3@3.1.
And overclocking cpu of a8 can widen the gap to a total of 50%.

Honestly the A8-38xx really doesn't have a market segment outside of HTPC. It's too expensive for low budget box's, for the same price you can pickup a better i3/i5 and a low end dGPU. Maybe trinity will solve this but I doubt it, GPU is simply bottle necked too much having to use the main memory and won't ever compete with a dGPU. The IGA takes up die space which drives the cost of the APU up and prevents it from competing with Intel on a same or similar price level. APU's will work really well if they can keep their cost to $125 USD or less, preferably under $100 USD.

 

[r0ck3tm@n]

Honestly the A8-38xx really doesn't have a market segment outside of HTPC. It's too expensive for low budget box's, for the same price you can pickup a better i3/i5 and a low end dGPU. Maybe trinity will solve this but I doubt it, GPU is simply bottle necked too much having to use the main memory and won't ever compete with a dGPU. The IGA takes up die space which drives the cost of the APU up and prevents it from competing with Intel on a same or similar price level. APU's will work really well if they can keep their cost to $125 USD or less, preferably under $100 USD.

What he said.
The review article on the A8-3870K did not recommend the A8-3870K.

What I do like about AMD APU's is they provide graphics that will handle anything less than what intensive gaming requires. I realize this is not a statement that will surprise anyone but for me it's a new thing. It's the alternative to a discrete card that I like. The reader might conclude that perhaps Intel does a better job at this and I would agree. Everyone seems concerned with the SLI or crossfire abilities of a heterogeneous cpu/gpu. The most impressive thing I have seen done with the cpu/gpu design is the energy savings that you can get with the motherboards that switch to the onboard graphics and back to the discrete graphics as required. Does AMD offer anything like this? If not, they should.

In my thinking the main computational abilities of the cpu should not be compromised for graphics. Of course, we want both a powerful cpu and a powerful gpu. What I would like to see is for AMD to get their cpu designs up to an impressive level again and continue on with whatever designs they have planned. Perhaps something radical is in order, like using technology from some other cpu design. Yes, I would like to see AMD shake things up WITH THEIR CPU DESIGN. It can be done. Someone needs to wake up over there. And I'm referring to engineering not downsizing. Engineering needs to overcome cost. Why does it need to be that placing everything on one chip is better? I think connectivity is at least as important for future design as integration.

 

[jimmysmitty]

Not strange at all. Stop looking at timed loop demo's and start actually playing games in real time. There is significantly more processing that needs to be done in a multiplayer environment then in a single player one running on auto-pilot. That processing lends itself to multiple threads working simultaneously. In single player you have one actor (the player) and everything in the environment reacting to that player, in multiplayer you have many independent actors causing the environment to react.

I only look at benchmarks as a way to see information and how it performs. Normally MP should perform the same, even with more people it shouldn't need extra cores to deal with it as all it normally is is information in the packet.

But BF3 must be different.

1: BF3 MP requires four cores. I did a blind test where is disabled two cores on my 2600k, and trust me, there was a difference on a single 570 GTX. In SP though, its GPU limited, thanks to those shiny DX11 effects.

2: Remember, that DX10/11 shifted a LOT of resource usage onto the GPU. Hence why DX11 titles tend to be GPU limited, and DX9 titles tend to require more CPU power [and scale better as a result]. The more that gets shifted to the GPU, the less important the CPU becomes in gaming.

So is the MP only DX9 then? It would make sense then. But then again why? Why not use that extra power in the GPU thats going to waste?

Honestly the A8-38xx really doesn't have a market segment outside of HTPC. It's too expensive for low budget box's, for the same price you can pickup a better i3/i5 and a low end dGPU. Maybe trinity will solve this but I doubt it, GPU is simply bottle necked too much having to use the main memory and won't ever compete with a dGPU. The IGA takes up die space which drives the cost of the APU up and prevents it from competing with Intel on a same or similar price level. APU's will work really well if they can keep their cost to $125 USD or less, preferably under $100 USD.

I prefer a dGPU in my HTPC. Mainly because its tried and true. Even if I bought a i3 2120 for a HTPC, I would probably keep my HD5450 or put a HD6450/7450 in for the best quality. Plus they use very little power and create almost no heat.

 

[mayankleoboy1]

http://www.obr-hardware.com/2012/03/apu-trinity-tested-good-gpu-still.html

tritnity gets pre-tested.

As i expected, Piledriver is the same crap as Bulldozer is ... ive told it you last year ... GPU working pretty well, but CPU is still ***. AMD is not able to make a good CPU.

 

[truegenius]

nope!
But why😀

 

[palladin9479]

I only look at benchmarks as a way to see information and how it performs. Normally MP should perform the same, even with more people it shouldn't need extra cores to deal with it as all it normally is is information in the packet.

Multiplayer has more independently moving objects. A game does more then render a screen over and over again, there are tons of background calculations that need doing. In a single player timed demo the only thing that needs calculating is what's directly in-front of the player and some mild AI. Timed demo requires even less as AI is all pre-programmed if it even exists, same with physics. The only thing your testing is the graphics rendering capacity. In actual play with a human making random movements, AI's responding to those movements, object collision and movements. If the environment is destructible (BF3 is) then you need to dynamically make terrain calculations and so forth. In MP there is no AI but there is significantly more objects being blown apart and events taking place, all requiring their own set of calculations.

I honestly can't believe how you thought multiplayer and timed loop demo's had remotely the same processing profile.

And it's incredibly hard to put a dGPU inside a small mini-itx case, which is where HTPC's are moving to. A few pages back I listed some very fashionable case's that would match my living room decor.

 

[bawchicawawa]

http://www.obr-hardware.com/2012/03/apu-trinity-tested-good-gpu-still.html

tritnity gets pre-tested.

As i expected, Piledriver is the same crap as Bulldozer is ... ive told it you last year ... GPU working pretty well, but CPU is still ***. AMD is not able to make a good CPU.

http://3.bp.blogspot.com/-zKx5TXQ_Z6I/T2eDssRvRxI/AAAAAAAACEg/qJcvH0kDZ9E/s1600/2012031911411890.jpg



http://1.bp.blogspot.com/-PARt_HNhX.../Xj8prN8We3A/s1600/16100462m64m3ohl6mzw4e.jpg

Super pi and 3dmark isn't much of a coverage of the cpu's real world performance.

Anyways, trinity = piledriver with no l3 cache. This is like saying the athlon II 651(the a8 3850 without the igp) is terrible because it doesn't match phenom II's performance, which isn't the point at all.

The point of trinity is to be an upgrade from llano(which it will be...), not the previous cpu generation...

 

[palladin9479]

Here are the case's I was looking at

http://www.streacom.com/products/fc5-od-fanless-chassis/

You have one expansion slot, and that requires a bracket. An APU can do everything a HTPC possibly needs, it even has the Radeon GPU to go with it. Thus you can use that slot for a digital tuner or wireless (if you float that way) card, or whatever else you want. Provided they keep the APU cost at $100 or less.

 

[jimmysmitty]

http://www.obr-hardware.com/2012/03/apu-trinity-tested-good-gpu-still.html

tritnity gets pre-tested.

As i expected, Piledriver is the same crap as Bulldozer is ... ive told it you last year ... GPU working pretty well, but CPU is still ***. AMD is not able to make a good CPU.

http://3.bp.blogspot.com/-zKx5TXQ_Z6I/T2eDssRvRxI/AAAAAAAACEg/qJcvH0kDZ9E/s1600/2012031911411890.jpg



http://1.bp.blogspot.com/-PARt_HNhX.../Xj8prN8We3A/s1600/16100462m64m3ohl6mzw4e.jpg

3.8GHz vs 2.9GHz. I would like to see that llano CPU at 3.8GHz just for kicks. Its a 31% higher clock speed with gains of 12% in 3DMark and 9% in SuperPi. Not quite what we are expecting.

That is if this is 100% legit. It looks like it could be considering the GPU score looks to be about what I would expect from it but wtill se have some time till it comes out.

I only look at benchmarks as a way to see information and how it performs. Normally MP should perform the same, even with more people it shouldn't need extra cores to deal with it as all it normally is is information in the packet.

Multiplayer has more independently moving objects. A game does more then render a screen over and over again, there are tons of background calculations that need doing. In a single player timed demo the only thing that needs calculating is what's directly in-front of the player and some mild AI. Timed demo requires even less as AI is all pre-programmed if it even exists, same with physics. The only thing your testing is the graphics rendering capacity. In actual play with a human making random movements, AI's responding to those movements, object collision and movements. If the environment is destructible (BF3 is) then you need to dynamically make terrain calculations and so forth. In MP there is no AI but there is significantly more objects being blown apart and events taking place, all requiring their own set of calculations.

I honestly can't believe how you thought multiplayer and timed loop demo's had remotely the same processing profile.

And it's incredibly hard to put a dGPU inside a small mini-itx case, which is where HTPC's are moving to. A few pages back I listed some very fashionable case's that would match my living room decor.

Most games run the same in single and multiplayer. It just suprises me that they wouldn't take that extra power of the CPU in SP and use it as well to add more to it.

As for mini-ITX, we use a very small one at work for our Intel one, a Antec case, and it still has room for a low profile dGPU. A HTPC doesn't need more than a HD5450.

Super pi and 3dmark isn't much of a coverage of the cpu's real world performance.

Anyways, trinity = piledriver with no l3 cache. This is like saying the athlon II 651 is terrible because it doesn't match phenom II's performance, which isn't the point at all.

The point of trinity is to be an upgrade from llano(which it will be...), not the previous cpu generation...

While its true that the no L3 is going to mean not 100% true performance, L3 only helps in some areas:

http://www.anandtech.com/bench/Product/188?vs=80

Overall Athlon II and Phenom II perform very much alike, in some areas the L3 helps. In Sandy bridge I would think it makes more of a difference because the L3 has a major role, it stores all L1 and L2 instructions. But thats not the same in AMDs CPUs, as of yet.

If its true it just gives a preliminary idea of performance until we see PD hit the market.

What I will find interesting is the actual clock speed of Trinity vs Llano, which is where I think all of the performance gains for Trinity will come from over Llano not because IPC has actually gone up.

But thats just my theory for now, and if those benchmarks are true as are the numbers, the 31% higher clock speed for that little of a gain performance wise is not very impressive.

 

[palladin9479]

I prefer a dGPU in my HTPC. Mainly because its tried and true. Even if I bought a i3 2120 for a HTPC, I would probably keep my HD5450 or put a HD6450/7450 in for the best quality. Plus they use very little power and create almost no heat.

Just making absolutely sure your saying a HD6450 because ...

http://www.tomshardware.de/Streacom-FC5-Test-Review,testberichte-240959-14.html

The older Llano's beat the HD6450.

Your getting a CPU and a GPU better then a HD6450 in a single package.

And wow ... talk about double speak.

Most games run the same in single and multiplayer. It just suprises me that they wouldn't take that extra power of the CPU in SP and use it as well to add more to it.

This could be said exactly opposite. That other multiplayer games are not taking advantage of the other two cores.
But no... that would mean the BD is actually useful for something ... and we can't possibly have that. So it's gotta be BF3's fault right, they just didn't program it right.

Did you ever thing that it's the additional computations from the destructible environment and independently moving objects (other players) that require the additional CPU power? As I said, stop taking single player timed demo's as any indication of performance in a multiplayer setup. Or would you prefer if they hamstrung their engine so that it only use's two cores in multiplayer, like the previous games did.

 

[palladin9479]

Review of that case + APU in a fanless setup.

http://www.techpowerup.com/reviews/Streacom/FC5_OD/1.html

There is a market for these, especially amongst system integrators and OEMs.

 

[jimmysmitty]

I prefer a dGPU in my HTPC. Mainly because its tried and true. Even if I bought a i3 2120 for a HTPC, I would probably keep my HD5450 or put a HD6450/7450 in for the best quality. Plus they use very little power and create almost no heat.

Just making absolutely sure your saying a HD6450 because ...

http://www.tomshardware.de/Streacom-FC5-Test-Review,testberichte-240959-14.html

The older Llano's beat the HD6450.

Your getting a CPU and a GPU better then a HD6450 in a single package.

And wow ... talk about double speak.

Most games run the same in single and multiplayer. It just suprises me that they wouldn't take that extra power of the CPU in SP and use it as well to add more to it.

This could be said exactly opposite. That other multiplayer games are not taking advantage of the other two cores.
But no... that would mean the BD is actually useful for something ... and we can't possibly have that. So it's gotta be BF3's fault right, they just didn't program it right.

Did you ever thing that it's the additional computations from the destructible environment and independently moving objects (other players) that require the additional CPU power? As I said, stop taking single player timed demo's as any indication of performance in a multiplayer setup. Or would you prefer if they hamstrung their engine so that it only use's two cores in multiplayer, like the previous games did.

Better performance does not always mean better quality. But then again as I said, thats my preference. Just like its my preference to go with ATI GPUs over nVidia.

As for the game, its not about BD. Its about the fact that they are each different. I do wonder then as well why a lot of the major reviews of BF3 didn't include MP as well since it is a major part of the game.

Still intersting to know.

 

[bawchicawawa]

3.8GHz vs 2.9GHz. I would like to see that llano CPU at 3.8GHz just for kicks. Its a 31% higher clock speed with gains of 12% in 3DMark and 9% in SuperPi. Not quite what we are expecting.

That is if this is 100% legit. It looks like it could be considering the GPU score looks to be about what I would expect from it but wtill se have some time till it comes out.



Most games run the same in single and multiplayer. It just suprises me that they wouldn't take that extra power of the CPU in SP and use it as well to add more to it.

As for mini-ITX, we use a very small one at work for our Intel one, a Antec case, and it still has room for a low profile dGPU. A HTPC doesn't need more than a HD5450.



While its true that the no L3 is going to mean not 100% true performance, L3 only helps in some areas:

http://www.anandtech.com/bench/Product/188?vs=80

Overall Athlon II and Phenom II perform very much alike, in some areas the L3 helps. In Sandy bridge I would think it makes more of a difference because the L3 has a major role, it stores all L1 and L2 instructions. But thats not the same in AMDs CPUs, as of yet.

If its true it just gives a preliminary idea of performance until we see PD hit the market.

What I will find interesting is the actual clock speed of Trinity vs Llano, which is where I think all of the performance gains for Trinity will come from over Llano not because IPC has actually gone up.

But thats just my theory for now, and if those benchmarks are true as are the numbers, the 31% higher clock speed for that little of a gain performance wise is not very impressive.

Obviously clock speed doesn't matter when they are way different when it comes to the arch, that's like comparing a 955 with a 2500k at the same clock for a comparison. So you saying "the 31% higher clock speed for that little of a gain performance wise is not very impressive." is a moot point. You are acting like it's the same arch...

Also, l3 cache is very important for real world performance where there will be more misses than a scripted benchmark.

 

[palladin9479]

Better performance does not always mean better quality. But then again as I said, thats my preference. Just like its my preference to go with ATI GPUs over nVidia.

Considering the graphics core in the APU and the GPU are made by the same people, your kinda showing your bias here. You like ATI GPU's but you don't like AMD produced ATI GPU's? The GPU inside Llano is a RV-800, similar to the HD55xx GPUs. Only difference is they fused their MMU with the CPU. Even use the same drivers. I should know, I happen to have one.

http://en.wikipedia.org/wiki/Evergreen_%28GPU_family%29

Redwood Pro 400:20:8 128-bit DDR2/GDDR3/GDDR5 memory interface. Core clock 650, memory clock 400 / 900. The sabine version of the Llano use's a 400Mhz clocked GPU component.

All AMD did was die-shrink the graphics core and implant it alongside a die-shrunk Phenom II x4. The trinity will be an "enhanced BD" (whatever that means) coupled with a HD-6xxx graphics core.

As for the game, its not about BD. Its about the fact that they are each different. I do wonder then as well why a lot of the major reviews of BF3 didn't include MP as well since it is a major part of the game.

Still intersting to know.

Nobody includes realistic scenarios because you've all been using timed demo's for so long that it's stuck in your collective heads as the only measurement for performance. Timed demo's worked great during the time of UT / Doom 3, not so much anymore. People are under the impression that all games do is render a screen over and over again, when that's false. MP is really hard to control for, you need to simulate it in a lab with an actual server, most reviewers don't have time for that and just run the same canned recorder run over and over again on different graphics settings and call it a day.

Just a warning to be careful about swearing by benchmarks, they only show what their designed to show, nothing more and nothing less.

 

[palladin9479]

Obviously clock speed doesn't matter when they are way different when it comes to the arch, that's like comparing a 955 with a 2500k at the same clock for a comparison. So you saying "the 31% higher clock speed for that little of a gain performance wise is not very impressive." is a moot point. You are acting like it's the same arch...

Also, l3 cache is very important for real world performance where there will be more misses than a scripted benchmark.

L3 is mostly ~bleh~, then again BD has a history of poor prediction rates so it might need it more then Intel.

It's the bias showing, clock rates don't matter when it favors their opinion. Clock rates do matter when it does favor their opinion.

Only two things matter when considering performance, power consumption and cost. That is why you compare products at various price points for their given market segments, everything else is just how the manufacturer presented their product.

When trinity comes out they will bench the crap out of it. Compare it to an i3 / i5 at the same price point. It'll be behind in the CPU department, it'll have a nicer GPU then it does now. If games are already playing on a Sabine then they'll be even more playable on a Trinity. And if AMD's Llano is anything to go by, then Trinity will be a Turks (HD65xx/66xx) GPU core bolted onto the APU. Turks is 480:24:8 with 128-bit memory bus and 650 ~ 800MHZ core clock. It's using VLIW5 as is Trinity, which is what makes me believe they'll just reuse this design. Die shrink it and tweak it's clock out in accordance with the TDP of the chip.

 

[g4114rd0]

AMD Launches New Platform for Dedicated Web Hosting Providers..."

http://www.marketwatch.com/story/amd-launches-new-platform-for-dedicated-web-hosting-providers-2012-03-20
http://www.techpowerup.com/162666/AMD-Launches-New-Platform-for-Dedicated-Web-Hosting-Providers.html

 

[palladin9479]

AMD Launches New Platform for Dedicated Web Hosting Providers..."

http://www.marketwatch.com/story/amd-launches-new-platform-for-dedicated-web-hosting-providers-2012-03-20
http://www.techpowerup.com/162666/AMD-Launches-New-Platform-for-Dedicated-Web-Hosting-Providers.html

Blasphemy, it's Blasphemy I say!!!

Why .. their using TDP numbers, and we all know AMD lies and only use's ACP.

It's all Blasphemy!!!


/sarcasm

 

[palladin9479]

Kinda why I find it hard to really recommend a desktop system over an A6, IMO, the A4s kinda the limit between APU and CPU+Discrete Graphics sometimes.




From my experience, the MP is usually more taxing than the Single player, so I doubt it's DX9 only....
not to mention all those HBAO effects that's still there

APU's make the most sense where your space / power limited, basically mobile computing and small mini-itx cases. I would of really liked to see AMD release FS1 CPU's for public purchase along with vender's making mini-itx boards that mount FS1 CPUs. A 45W CPU + GPU DDR3-1600 (1866 soon) on a small board would make a great set-top style PC / light gaming device. AMD's growth in the mobile market shows how popular they've become.

Desktop on the other hand ... too easy to buy a $70 USD dGPU that will smoke anything inside an APU due to memory bandwidth limitations.

 

[g4114rd0]

AMD Radeon HD 7990 Specifications..."

http://img.donanimhaber.com//images/haber/32277/amdhd7990habergor.jpg

http://vr-zone.com/articles/report-amd-radeon-hd-7990-specifications/15285.html

edit: Another Picture

 

[bawchicawawa]

APU's make the most sense where your space / power limited, basically mobile computing and small mini-itx cases. I would of really liked to see AMD release FS1 CPU's for public purchase along with vender's making mini-itx boards that mount FS1 CPUs. A 45W CPU + GPU DDR3-1600 (1866 soon) on a small board would make a great set-top style PC / light gaming device. AMD's growth in the mobile market shows how popular they've become.

Desktop on the other hand ... too easy to buy a $70 USD dGPU that will smoke anything inside an APU due to memory bandwidth limitations.

Well, the cpu portion of an a8 3870k costs about 92$ (although the athlon x II 651 comes out of a 3850 so it's not unlocked so bad overclocking, lol) while the whole 3870k costs 125$.

http://www.anandtech.com/show/4448/amd-llano-desktop-performance-preview/3

Do the math. =p
I'd suggest the 3850/3870k over anything when building/buying a cheap, general use computer.

 

[truegenius]

trinity have 30% more clock speed and that provides 30% more performance than a llano.
Do some math
and you will find that cpu of trinity (e-bd, pd) is having ipc equals to ph2. And at ultra low power usage that too without l3 which saved the bd's performance (bad branch prediction).
So l3 can give some more boost in pd.

Conclusion
pd will have 10% better ipc than ph2, 30% than bd and sb will be only 40% more better in ipc that too without l3 cache.
And with l3, sb will have only 30% better ipc.
Also pd may start from 4ghz and maybe upto 5ghz while sb toped out at 3.5ghz so another 40% performance from clock that will contributes to 10% better pd (5g) than sb@3.5g.

No prediction, only expectations (based on trinity).

 

[bawchicawawa]

http://wccftech.com/amd-trinity/
Hold up...

"The A10-5800K would be the flagship part of the Trinity Lineup featuring a Quad Core Design clocked at 3.8GHz Stock and 4.2GHz Turbo Core 2.0 Frequency. The APU also features a 4Mb L3 Cache, HD7660D graphics solution and a rated TDP of 100W."

 

[gamerk316]

Better performance does not always mean better quality. But then again as I said, thats my preference. Just like its my preference to go with ATI GPUs over nVidia.

As for the game, its not about BD. Its about the fact that they are each different. I do wonder then as well why a lot of the major reviews of BF3 didn't include MP as well since it is a major part of the game.

Still intersting to know.

Hard to bench reliably. You'd basically need your own server and have a group of people carry out a pre-planned routine in order to get something that was even somewhat reliable as a benchmark.

 

[lilcinw]

trinity have 30% more clock speed and that provides 30% more performance than a llano.
Do some math
and you will find that cpu of trinity (e-bd, pd) is having ipc equals to ph2. And at ultra low power usage that too without l3 which saved the bd's performance (bad branch prediction).
So l3 can give some more boost in pd.

Conclusion
pd will have 10% better ipc than ph2, 30% than bd and sb will be only 40% more better in ipc that too without l3 cache.
And with l3, sb will have only 30% better ipc.
Also pd may start from 4ghz and maybe upto 5ghz while sb toped out at 3.5ghz so another 40% performance from clock that will contributes to 10% better pd (5g) than sb@3.5g.

No prediction, only expectations (based on trinity).

That would be epic if it is true, but after BD you will excuse me if I don't hold my breath.

Still... hoping for something epic to fill my GA-990FXA-UD3 with.