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 ...

 

[gamerk316]

For the first time, some execs do acknowledge that Bulldozer approach may not have been the best one at the time, and things need to change.

*whistles*

Seriously, can I say "I TOLD YOU SO" yet? Because I did, many times.

And again, I note the total lack of news regarding PD, which is telling me that there aren't going to be significant changes. If PD was significantly better, AMD would be telling us, ESPECIALLY after the BD debacle.

 

[jimmysmitty]

Huh ... how .. what ... brain .... hurting...

I am off my rocker or did you just compare an Atom to an APU? I m ean ... Via Nano's crush Atom's, and this is Via we're talking about. AMD's E series crush's Nano's, A4 APU's beat E's. So ... your comparing, with a straight face, the bottom most processor capable of running x86 with the highest mobile CPU AMD makes. That's as bad as that Anandtech review discussed earlier.

Technically this Atom is also a APU, lest you forget the integrated PowerVR IGP.

Still I was just noting it. AMD wants to get into that market, they have a long road ahead. Intel is already there. The normal use is 9+ hours, same as most tablets right now, and less standyby but still not bad.

I would assume AMD will push brazos there, at least the next gen or two.

Trinity will be fighting IB soon and we wont know the full specs yet but it may or may not be decent competition. I at least hope its better than AMDs Thunderbolt alternative, Lightning Bolt, which a USB 3.0 device wont even reach full 5Gbps speeds while TB is 10Gbps bidirectional with plans to 50Gbps bidirectional by 2014.

 

[jimmysmitty]

For the first time, some execs do acknowledge that Bulldozer approach may not have been the best one at the time, and things need to change.

*whistles*

Seriously, can I say "I TOLD YOU SO" yet? Because I did, many times.

And again, I note the total lack of news regarding PD, which is telling me that there aren't going to be significant changes. If PD was significantly better, AMD would be telling us, ESPECIALLY after the BD debacle.

Maybe not. They did just fire a bunch of people, possibly marketing. Don't want it to get over hyped again and let it come out to a uninspiring release. Better to under promise and over deliver.

 

[Reynod]

The twelve hour resting battery life is great but I wonder how Intel will do there. Their Atom for tablets is said to have 30 hours battery life in the same situation. It may not be a full CPU like a Core i3 but still its impressive.

I guess we shall see. AMD needs to pull something amazing out soon or they may just well become #3 in the CPU game.

Seriously jimmy ... have you ever tried to use one of those Atom powered netbooks?

I played with 3 or 4 of them at work and gave them back ... seriously ... you can't do stuff all past e-mail and browsing a BBS type of site ...

The graphics is so slow, and it has the processing power of my Nan doing her lotto numbers after her meds have kicked in ...

The Llano and Brazos (I have an E450 ASUS) is so good its amazing ... I think AMD should have called it "Brawndo" ... http://www.brawndo.com/

Its got electrolytes !!

 

[lilcinw]

Seriously jimmy ... have you ever tried to use one of those Atom powered netbooks?

I played with 3 or 4 of them at work and gave them back ... seriously ... you can't do stuff all past e-mail and browsing a BBS type of site ...

The graphics is so slow, and it has the processing power of my Nan doing her lotto numbers after her meds have kicked in ...

The Llano and Brazos (I have an E450 ASUS) is so good its amazing ... I think AMD should have called it "Brawndo" ... http://www.brawndo.com/

Its got electrolytes !!

:lol: I may have to steal that one.

 

[sarinaide]

Anyone have a ETA to Trinity, need something to occupy me during the longer wait to Piledriver.

 

[g4114rd0]

Trinity enabled integer division unit: Also on FX chips reactivated?..."

http://translate.google.de/translate?u=http%3A//www.planet3dnow.de/cgi-bin/newspub/viewnews.cgi%3Fid%3D1334532731&sl=de&tl=en&hl=de&ie=UTF-8

Two new keynotes announced for AFDS..."

http://semiaccurate.com/2012/04/16/two-new-keynotes-announced-for-afds/

AMD Expects Significant Performance Improvements with Steamroller Microprocessors..."

http://www.xbitlabs.com/news/cpu/display/20120416165428_AMD_Expects_Significant_Performance_Improvements_with_Steamroller_Microprocessors.html

 

[palladin9479]

For the first time, some execs do acknowledge that Bulldozer approach may not have been the best one at the time, and things need to change.

*whistles*

Seriously, can I say "I TOLD YOU SO" yet? Because I did, many times.

And again, I note the total lack of news regarding PD, which is telling me that there aren't going to be significant changes. If PD was significantly better, AMD would be telling us, ESPECIALLY after the BD debacle.

You realize their not talking about the modular fusion concept but their design process. The modular concept is required (splitting the processing components of the CPU apart) for them to execute fusion and heterogeneous computing. They just screwed up the implementation, especially with sharing something as critical and the L2 cache. All they needed to do was split the FPU off from the main CPU, not try to stuff two CPU's into the same real-estate as one.

 

[Cazalan]

AMD Expects Significant Performance Improvements with Steamroller Microprocessors..."

http://www.xbitlabs.com/news/cpu/display/20120416165428_AMD_Expects_Significant_Performance_Improvements_with_Steamroller_Microprocessors.html

I wonder if that means they're ditching the module design and going back to enhancing K10. 😉

 

[jimmysmitty]

Seriously jimmy ... have you ever tried to use one of those Atom powered netbooks?

I played with 3 or 4 of them at work and gave them back ... seriously ... you can't do stuff all past e-mail and browsing a BBS type of site ...

The graphics is so slow, and it has the processing power of my Nan doing her lotto numbers after her meds have kicked in ...

The Llano and Brazos (I have an E450 ASUS) is so good its amazing ... I think AMD should have called it "Brawndo" ... http://www.brawndo.com/

Its got electrolytes !!

I have. With 2GB of RAM they run fine. Nothing I would ever use but they don't run as slow as some people make them out to be.

The best thing is that they are all up from there. Improvements to performance through new tech or old tech added on, such as OoO, or better process tech.

If AMD gets into the market soon, they might have a chance. But if they are too far away they might just enter the race to be left way behind as this market does not sit still. New CPUs and phones come out almost twice a year. Intel has a good chance because they have the funds to push it, AMD on the other hand does not. So they better hope for either something just insanley amazing or they get very lucky there.

I wonder if that means they're ditching the module design and going back to enhancing K10. 😉

Doubt it. I heard this was to be their design for the next few CPUs. Steamroller would be another enhancement, possibly some tweaks to PD/BD with a generally good outcome in performance or so they hope. They are pushing for 10-15% a year, but that may not be enough as Intel currently does 5-15% per die shrink and more than 20% normally in new arch so thats about 25-35% per tick-tock while AMDs would be 20-30% per tick-tock. It would be enough to keep them right where they are, unfortunatley not enough to push them ahead.

That is unless Intel makes a major screwup and pulls a Willamette (first Pentium 4) which was underperforming Coppremine (last gen Pentium III also the basis for Core) per clock. I would assume they are shooting to not do that again as that was one of the major screw ups that allowed AMD to take the performance crown.

That said, the 20-30% a tick-tock for AMD should at least allow them to compete in the mainstream (i.e. 1155/1150) market to stay alive for hope of a monster CPU arch like K8 again.

 

[skaughtz]

Trinity enabled integer division unit: Also on FX chips reactivated?..."

http://translate.google.de/translate?u=http%3A//www.planet3dnow.de/cgi-bin/newspub/viewnews.cgi%3Fid%3D1334532731&sl=de&tl=en&hl=de&ie=UTF-8

Would someone care to summarize what is being discussed in this article? The translation errors didn't help my understanding...

 

[Cazalan]

Doubt it. I heard this was to be their design for the next few CPUs. Steamroller would be another enhancement, possibly some tweaks to PD/BD with a generally good outcome in performance or so they hope. They are pushing for 10-15% a year, but that may not be enough as Intel currently does 5-15% per die shrink and more than 20% normally in new arch so thats about 25-35% per tick-tock while AMDs would be 20-30% per tick-tock. It would be enough to keep them right where they are, unfortunatley not enough to push them ahead.

I read the article again and the cited quote wasn't nearly as aggressive as the article heading. I guess that's the norm. No real news just a reminder that Steamroller is still in the queue. Considering all the advanced info coming out on Haswell the Steamroller news is non-existent.

 

[noob2222]

I wonder if that means they're ditching the module design and going back to enhancing K10. 😉

better integration between CPU and GPU chips and other improvements

Steamroller is their original plan for a true fusion chip. The GPU will be taking over FPU calculations when it can. Other than the sheer design aspect, here is the key factor why the didn't or couldn't jump straight to the design.

Right now there around 200 apps that can be accelerated by stream processing units of GPUs, but several years from now that number will increase and the compute performance of GPU cores will be critical for success on the mass market. Still, x86 performance clearly remains an important factor, but the breakthrough in that direction may occur only in two years down the road.

http://www.microsoft.com/hpc/en/us/solutions/gpgpu-acceleration.aspx
The thing is when you attach it directly to the cpu and have the APU accelerate the computations, it gets even faster than routing it all the way through the cpu over to the gpu and back. Execution will be critical in whether this is a success, and there will be a huge change in the scheduler for that to happen.

 

[Cazalan]

Right now there around 200 apps that can be accelerated by stream processing units of GPUs, but several years from now that number will increase and the compute performance of GPU cores will be critical for success on the mass market. Still, x86 performance clearly remains an important factor, but the breakthrough in that direction may occur only in two years down the road.

http://www.microsoft.com/hpc/en/us/solutions/gpgpu-acceleration.aspx

The thing is when you attach it directly to the cpu and have the APU accelerate the computations, it gets even faster than routing it all the way through the cpu over to the gpu and back. Execution will be critical in whether this is a success, and there will be a huge change in the scheduler for that to happen.

Something tells me it's going to take a lot more than 2 years for GPGPU to take off. It will be great for new apps but when people buy faster CPUs they expect all their old software to run faster.

 

[viridiancrystal]

So yeah, interesting story about those windows hotfixes that Microsoft released after BD came out, if anyone wants to hear it.

 

[esrever]

Something tells me it's going to take a lot more than 2 years for GPGPU to take off. It will be great for new apps but when people buy faster CPUs they expect all their old software to run faster.

most old software run well enough.

 

[palladin9479]

Something tells me it's going to take a lot more than 2 years for GPGPU to take off. It will be great for new apps but when people buy faster CPUs they expect all their old software to run faster.

Doesn't need to. GCN GPU's can process SIMD (SSE) instructions along with Integer operations, they can do this right ~now~. All that needs to be done is the dispatcher to send the instructions to the iGPU instead of the SIMD unit (FPU). We won't see this until steamroller era.

People need to realize that there is no intrinsic difference between what a modern GPU does and what a modern "FPU" does. They both are vector CPU's that were grown from two different sources but grew to the same goal, executing large quantities of math operations simultaneously.

 

[jimmysmitty]

I read the article again and the cited quote wasn't nearly as aggressive as the article heading. I guess that's the norm. No real news just a reminder that Steamroller is still in the queue. Considering all the advanced info coming out on Haswell the Steamroller news is non-existent.

I said this a while back but we will have more info on Haswell before we know the 100% on Piledriver even. AMD has been keeping its lips shut pretty tight and I think its a bad idea since you want some hype, maybe not "Barcelona 40% better" to be misconstruded but at least something. Right now we are still selling more Intel CPUs than AMD even though AMD does give more cores per $.

But when theres nothing to hook people to the future or any real idea they tend to go with what they know, which thanks to marketing is Intel.

 

[king smp]

I will say it again
for the general computer buyer AMD has a great marketing strategy with BD
consumers dont know about benches and IPC
they know about GHZ and cores
they can grasp more speed in GHZ and they can grasp more cores are better
along with a cool name like Bulldozer if AMD gets out there with some funny commercials of a Bulldozer running over Intel computers with a funny spokesperson
plus push speed and cores they have a chance of really gettting "Joe Consumer" to buy their product
combined with a good economy desktop APU with decent enough graphics power to play WoW and Farmville which can made into sub-$500 OEM systems
they have a chance to really break the Intel branding domination

the ghz speed approach worked well with Intel with the Pentium 4
sure the Athlon XP was better but more computers got sold with Intel
remember most people buy towers premade
they dont buy retail CPUs and read benches and reviews

 

[gamerk316]

Doesn't need to. GCN GPU's can process SIMD (SSE) instructions along with Integer operations, they can do this right ~now~. All that needs to be done is the dispatcher to send the instructions to the iGPU instead of the SIMD unit (FPU). We won't see this until steamroller era.

People need to realize that there is no intrinsic difference between what a modern GPU does and what a modern "FPU" does. They both are vector CPU's that were grown from two different sources but grew to the same goal, executing large quantities of math operations simultaneously.

Exactly. But you would expect a larger latency using a GPGPU as opposed to a FPU built directly on the CPU die. But you have a much larger capability with the GPU, especially with large datasets. I don't see the traditional FPU going away for that reason, but I do expect the GPGPU in some form to become much more important in the future.

 

[palladin9479]

Exactly. But you would expect a larger latency using a GPGPU as opposed to a FPU built directly on the CPU die. But you have a much larger capability with the GPU, especially with large datasets. I don't see the traditional FPU going away for that reason, but I do expect the GPGPU in some form to become much more important in the future.

I'm referring to the GPU on the die not the GPU in the slot. The on die GPU is directly tied into the CPU itself, it even shares the same memory bus. We'll see true "Fusion" chips in the mobile version of Steamroller first, although AMD might end up putting a Radeon GPU into the Steamroller desktop and just remove the display components so that only the processing array is present.

 

[noob2222]

Could be for the initial desktop steamroller, the other link hinted at AM3 being around for another 2 years. Either way with a discrete gpu, the on die gpu should be 100% useable for gpgpu.

Edit: am3+, some people are too picky on exact syntax

 

[dogman_1234]

I thought AM3 was a dead socket?

 

[truegenius]

Could be for the initial desktop steamroller, the other link hinted at AM3 being around for another 2 years. Either way with a discrete gpu, the on die gpu should be 100% useable for gpgpu.

Edit: am3+, some people are too picky on exact syntax

i have am3+, now where is my steamroller

 

[palladin9479]

i have am3+, now where is my steamroller

Well AM3+ already has all the interconnects for PCIe and dual channel memory, only reason Llano needed a new socket was for the external display for the GPU. AMD could in theory mount steamroller onto AM3+ and it should be compatible with a BIOS update. The only issue I see is PCIe 3.0 vs PCIe 2.0. Does PCIe 3.0 need a new pin out or is it electrically compatible with PCIe 2.0. I think it's compatible but don't quote me on that.