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 ...

 

[triny]

By 2013 their apu with steamroller will be on 28nm leaving space for bigger IGP maybe 7950Zx paired with a hd 7970 discrete
HSA in full swing.
The extremists who want more are a minute minority .
Piledriver is a stepping stone .

Eventually only the extremists will want traditional Desktops housed in chunky clunky steel boxes clad in plastic
the vast majority will move on to sleeker much smaller but potent solutions.
less energy,less heat,less money but still potent.

 

[viridiancrystal]

My take on those slides is that there is nothing new in them. They say AMD is going to continue to go down the road of "good enough," power-efficient CPUs and "heterogeneous computing." That is exactly what their APUs are- okay but not stellar CPU performance, inexpensive to make, fairly low-power (especially Zacate), and use of OpenCL and GPGPU with the on-die GPU component. I take the part about not having "an emphasis on cores" meaning that they are going to do away with the emphasis on per-core IPC and the resultant "fat" cores as in Stars/K10 and move towards simpler cores such as in Zacate/Bobcat and Bulldozer. Either that or AMD is going to emphasize the power of the GPGPU portion rather than the CPU cores in their APUs. I have a hunch it's more the latter as "disruptive technology" from AMD pretty much means "GPGPU" from how I see it.

Bulldozer seems to fit in very well with what the slides appear to say. Bulldozer is a very modular architecture and I would absolutely imagine it is designed to be even more integrated with the GPU than the Stars-based Llano ever was. The best example I can think of is that the FPU is no longer very tightly associated with the cores. I would predict that before too long, AMD will use the IGP as the FPU and end up with absolutely massive FPU power and overcome some of the current restrictions and downsides of using a discrete GPU to do calculations. That would certainly be a disruptive technology as current GPGPU implementations are very fast but also very limited in what they can do well.

I don't think AMD is giving up the desktop computer market. The desktop computer market continues to shrink, and much of what remains is corporate desktops. Most of those are small-form-factor units with slower, low-powered CPUs and IGPs. That's exactly the kind of machine an APU will do well in, and AMD will certainly continue to sell chips to that market. AMD also realizes that their high-end APUs such as the A8 and their repackaged server CPUs (FX series) are also "good enough" for enthusiast desktop usage. They get playable framerates in games when paired with an appropriate GPU and are at least in the same ballpark as Intel's CPUs with doing most real high-end desktop work like video encoding and the like. Sure, they won't win many benchmarks against Intel's top-end chips, but they will get the job done in most cases. AMD has really only ceded two markets to Intel. The first is the >4-socket server market, which is a small fraction of a percent of the server market and continues to decline. AMD gave that up when they decided to arrange the HT link topology in G34 Opterons to a maximum of quad-socket operation as compared to the previous Opteron 800/8000-series' 8-socket operation. The advantage of that choice is that it made for a better-connected and better-scaling 4P setup. The other is the competitive benchmark market. To paraphrase one of AMD's marketing directors, "yes, we could be in those markets, but we would definitely spend billions in R&D to earn maybe millions in sales. You do the math."

I think exactly the opposite on that statement by AMD. I believe they are going to move to Better IPC with the cores they have. AMD people are not all idiots. They knew they messed up with bulldozer, and they know they need to fix it.

http://www.tomshardware.com/reviews/amd-financial-analyst-day-trinity-kaveri-kabini,3126.html
AMD says much about moving from more core to IPC, and they cancelled their 10 and 20 core server processors because they felt the IPC improvements to PD will make up for the cores.

If This is all this reins true, then AMD may be on the right track.

 

[Chad Boga]

By 2013 their apu with steamroller will be on 28nm leaving space for bigger IGP maybe 7950Zx paired with a hd 7970 discrete
HSA in full swing.
The extremists who want more are a minute minority .
Piledriver is a stepping stone .

Eventually only the extremists will want traditional Desktops housed in chunky clunky steel boxes clad in plastic
the vast majority will move on to sleeker much smaller but potent solutions.
less energy,less heat,less money but still potent.

:lol:

 

[g4114rd0]

- ...It will be fun to focus on the tablet boom segment to make people h@ppy.”

 It's all about executing FX 2 Gen PD project, turnaround AMD makes a lot of sense.

AMD +0.15
2012-02-03 16:00

 

[g4114rd0]

you a stocks guy 'g4'.?

@ long time ago.

 

[earl45]

On a side note: Hey, Chad is alive........now where's Baron?

Baron's trying to figure out what happen to Bulldozer, once he figure's that out he will let the forum know how to fix them.

 

[earl45]

ssshhhhh.
we don't want to 'feed the beast'.
stay on topic.

Sorry! lol

 

[jimmysmitty]

http://vr-zone.com/articles/amd-a8-3870k-makes-it-past-6ghz-with-memory-at-3505-mhz-/14719.html

AMD A8-3870K makes it past 6GHz with memory at 3505 MHz!

Not related to FX, but related to APUs,

Ahh another LN2 cooled overclock that will mean nothing to the majority. Love how its the 3870K. Does that mean it will be better than Intels 3770K IB CPU?

This will just fuel the rumors and when it hits and does not deliver per the fanboy rumors, will just be another let down.

At least the GPU will be decent.

By 2013 their apu with steamroller will be on 28nm leaving space for bigger IGP maybe 7950Zx paired with a hd 7970 discrete
HSA in full swing.
The extremists who want more are a minute minority .
Piledriver is a stepping stone .

Eventually only the extremists will want traditional Desktops housed in chunky clunky steel boxes clad in plastic
the vast majority will move on to sleeker much smaller but potent solutions.
less energy,less heat,less money but still potent.

Yea, I am sorry but my Corsair 500R is not clunky, its beautiful. In fact its, as Apple would say, "Elegant". Moreso than those bland white towers they produce.

On a side note: Hey, Chad is alive........now where's Baron?

Another self imposed exile perhaps? He was better than before but didn't like people not praising BD. It will remain that way until AMD produces something better than Intel. Then he will gloat as if he was right all along.

 

[earl45]

Ahh another LN2 cooled overclock that will mean nothing to the majority. Love how its the 3870K. Does that mean it will be better than Intels 3770K IB CPU?

This will just fuel the rumors and when it hits and does not deliver per the fanboy rumors, will just be another let down.

At least the GPU will be decent.



Yea, I am sorry but my Corsair 500R is not clunky, its beautiful. In fact its, as Apple would say, "Elegant". Moreso than those bland white towers they produce.



Another self imposed exile perhaps? He was better than before but didn't like people not praising BD. It will remain that way until AMD produces something better than Intel. Then he will gloat as if he was right all along.

+1

 

[jdwii]

+1

Come on guys give the guy a break, I know he liked to fight with people but....????.... Ok carry on

 

[triny]

I don't know should I laugh or cry..........

http://3.bp.blogspot.com/-X_vlXWSMKxI/TcksYZWcKcI/AAAAAAAAAGc/0S-Y7XsGHoA/s1600/facedesk..jpg


The HD7950 has a die size of 350mm, which is larger than even the FX chips......................which means, it's gonna be a really big frying pan, instead of a practical APU

If they can shrink a 5000 series down to fit on 32nm I don't see them not shrinking a 256 bit v chip down to fit on 28nm
and even the biggest v chips on 14nm
I may well be wrong but I doubt they are going through all the trouble to end up with a hd 6550D on the die when their dream project is complete.

 

[triny]

Ahh another LN2 cooled overclock that will mean nothing to the majority. Love how its the 3870K. Does that mean it will be better than Intels 3770K IB CPU?



Yea, I am sorry but my Corsair 500R is not clunky, its beautiful. In fact its, as Apple would say, "Elegant". Moreso than those bland white towers they produce.

My white 600t looks good too
but I don't need all the drive space do you?
size will shrink by a lot
sales of smaller form factors will drive the market the sales data shows this is already starting to happen.

 

[jimmysmitty]

Ahh another LN2 cooled overclock that will mean nothing to the majority. Love how its the 3870K. Does that mean it will be better than Intels 3770K IB CPU?



Yea, I am sorry but my Corsair 500R is not clunky, its beautiful. In fact its, as Apple would say, "Elegant". Moreso than those bland white towers they produce.

My white 600t looks good too
but I don't need all the drive space do you?
size will shrink by a lot
sales of smaller form factors will drive the market the sales data shows this is already starting to happen.

Actually I do. Right now I am only using 3 drive bays (one for my SSD and two HDDs in a RAID0) but I like the room and the air flow of it is amazing. My idle temps dropped by about 4c (went from 32c to 28c) from my old case, has room for fans on the drive cages and room for the larger GPUs. And for gaming, the GPUs are not getting any smaller.

My next setup I plan, if I ever get enough extra money, to do two SSDs in a RAID0, and probably 4 HDDs in a performance RAID as well.

Small form factor may drive the normal market but you would be suprised. We sell a mix of PCs at work, even some ITX setups, and our midsize to full size sell a lot more. And the Corsair lineup we got in ahs sold very well, all except the 800D but thats also very expensive. But the 500R sells out almost as fast as we get them in, I have done plenty of builds with the 600T and 650D.

Trust me though, desktops will never go away. Tablets will never replace them. As great as they will get, they will still lack power. Even laptops, which were supposed to kill the DT market and have evolved greatly with better power still are weak compared to a DT.

Trinity will replace AMDs low end, and sadly I feel it will not be a proper replacement for the current Athlon IIs but thats my opinion. Just as if someone asks for a Phenom II we have to give them a FX and I tell them that if they do they need at least the FX 6 core or better to get equivalent performance. No need to lie and say that the FX 4100 is the same when in reality its not.

But still, as I said the DT market will always exist. There is plenty or room for it.

 

[jdwii]

Ahh another LN2 cooled overclock that will mean nothing to the majority. Love how its the 3870K. Does that mean it will be better than Intels 3770K IB CPU?



Yea, I am sorry but my Corsair 500R is not clunky, its beautiful. In fact its, as Apple would say, "Elegant". Moreso than those bland white towers they produce.

My white 600t looks good too
but I don't need all the drive space do you?
size will shrink by a lot
sales of smaller form factors will drive the market the sales data shows this is already starting to happen.

That case is not the reason why this thread was made, But if you keep persisting about it i'm going to say gross.

 

[r0ck3tm@n]

Actually I do. Right now I am only using 3 drive bays (one for my SSD and two HDDs in a RAID0) but I like the room and the air flow of it is amazing. My idle temps dropped by about 4c (went from 32c to 28c) from my old case, has room for fans on the drive cages and room for the larger GPUs. And for gaming, the GPUs are not getting any smaller.

My next setup I plan, if I ever get enough extra money, to do two SSDs in a RAID0, and probably 4 HDDs in a performance RAID as well.

Small form factor may drive the normal market but you would be suprised. We sell a mix of PCs at work, even some ITX setups, and our midsize to full size sell a lot more. And the Corsair lineup we got in ahs sold very well, all except the 800D but thats also very expensive. But the 500R sells out almost as fast as we get them in, I have done plenty of builds with the 600T and 650D.

Trust me though, desktops will never go away. Tablets will never replace them. As great as they will get, they will still lack power. Even laptops, which were supposed to kill the DT market and have evolved greatly with better power still are weak compared to a DT.

Trinity will replace AMDs low end, and sadly I feel it will not be a proper replacement for the current Athlon IIs but thats my opinion. Just as if someone asks for a Phenom II we have to give them a FX and I tell them that if they do they need at least the FX 6 core or better to get equivalent performance. No need to lie and say that the FX 4100 is the same when in reality its not.

But still, as I said the DT market will always exist. There is plenty or room for it.

I am thinking now that someone making decisions at AMD is planning for their APU's to replace the desktop high end CPU's. I don't agree with it right now but it's what I'm seeing. If Piledriver and what comes after is sufficiently powerful, that can happen and might even be really cool and I'm thinking more of the high performance computing aspect of it rather than the graphics aspect which I believe is such a demanding subsystem it should have it's own chip. That is until we develop substantially different processes for computer chips or whatever will take the place of computer chips.

My biggest problem with all this is that if AMD quits doing the research for a high performance desktop or server CPU, how will they produce a good mid range and low end CPU? I have no problem with merging the server CPU architecture with the desktop CPU architecture. Modularity in design should allow this.

I'm running one SSD for OS and four disk drives in RAID 10. I don't think it's too much, I think it's ideal. I just need one more SSD lol. And my FT02 is beautiful. If I want a computer in a tiny box I'll build one.

 

[jimmysmitty]

If they can shrink a 5000 series down to fit on 32nm I don't see them not shrinking a 256 bit v chip down to fit on 28nm
and even the biggest v chips on 14nm
I may well be wrong but I doubt they are going through all the trouble to end up with a hd 6550D on the die when their dream project is complete.

The HD5K they shrunk down was akin to a HD55X0, not a full fledged 1600SPU HD5870. And 14nm. Funny you mention that as Intel is already building their 14nm FAB. By the time AMD makes it to 22nm, Intels FAB will be up, running and probably popping out test chips.

And from what I have heard, the on board GPU for Trinity is still VILW which means, even though its being called a HD7K its still just a HD6K just like the Llano IGP was a HD5K with the HD6K branding.

I am thinking now that someone making decisions at AMD is planning for their APU's to replace the desktop high end CPU's. I don't agree with it right now but it's what I'm seeing. If Piledriver and what comes after is sufficiently powerful, that can happen and might even be really cool and I'm thinking more of the high performance computing aspect of it rather than the graphics aspect which I believe is such a demanding subsystem it should have it's own chip. That is until we develop substantially different processes for computer chips or whatever will take the place of computer chips.

My biggest problem with all this is that if AMD quits doing the research for a high performance desktop or server CPU, how will they produce a good mid range and low end CPU? I have no problem with merging the server CPU architecture with the desktop CPU architecture. Modularity in design should allow this.

I'm running one SSD for OS and four disk drives in RAID 10. I don't think it's too much, I think it's ideal. I just need one more SSD lol. And my FT02 is beautiful. If I want a computer in a tiny box I'll build one.

I doubt it. I don't think PD will bring enough to the table and I doubt AMD would go with only APUs. They need powerhouse CPUs for workstations, high end DT and servers and a on board GPU is wasted space for that. Thats why Intel still has a "extreme" high end with LGA1366 and LGA2011.

And as far as I can see, I don't think they will be able to merge DT and server. They are different. Its like trying to get a HD6970 to do real time 3D rendering that is done by the FirePros. The FirePros are made for the real time rendering while the HDs are geared towards gaming.

it does seem though as if AMD tries to always keep server and DT on the same arch. Maybe thats why they are failing. Maybe they need to do like Intel. Their DTs only are the same as entry level servers. But high end servers are different chips.

And thats how I have it right now. SSD OS, RAID0 data/games. The RAID0 allows for decent performance, not as fast as the SSD but still fast enough for games. And I personally hate tiny PCs. No room, horrible air flow and restricts design. Plus the high end cases from Corsair, Thermaltake and the such (not quite Antec right now, their current lineup is pretty bad) have great wire routing which allows for a clean looking PC, inside and out.

 

[g4114rd0]

http://www.xbitlabs.com/news/other/display/20120205041713_IBM_Quietly_Starts_to_Make_Chips_for_AMD.html

...so is not the end, unlocking the Winds of Change.

 

[g4114rd0]

that's interesting..

http://semiaccurate.com/2012/01/18/xbox-nextxbox-720-chips-in-production/

- ...and then I see a synthesized darkness, 2X APU with a blue logo in parallel into the MS console."

 

[triny]

The HD5K they shrunk down was akin to a HD55X0, not a full fledged 1600SPU HD5870. And 14nm. Funny you mention that as Intel is already building their 14nm FAB. By the time AMD makes it to 22nm, Intels FAB will be up, running and probably popping out test chips.

Regardless of which hdxxxx is on the die I believe they are working on allowing it to work in conjunction with any video card
by the time they mature their dream product in 2014.
Piledriver +25%cpu+50%igp looks good but steamroller will bring 20-30%{10%-15% per core}to cpu and GCN in 2013
by 2014 another 20%-30% another bump in igp
looks pretty darn good if it pans out
Intel has a lead with cpu raw power but if they should stumble on 14nm they may get run over, igp wise amd is light years ahead.
I believe the APU will eventually not only mature into a power desktop chip but will force Intel to heavily utilize HSA as well.
The idea of having cpu igp and 4 discrete all working in harmony under HSA by 2014-2015 tingles my fancy.
Though I have always bought Intel ,even I can see there is a better way than raw power. finesse always trumps raw power.
There always will be 8core+ server chips for applications that need more grunt.

 

[jimmysmitty]

http://www.xbitlabs.com/news/other/display/20120205041713_IBM_Quietly_Starts_to_Make_Chips_for_AMD.html

...so is not the end, unlocking the Winds of Change.

Not suprising. Since we knew that the next XBox was using a PowerPC CPU with a ATI GPU, it makes sense. That and AMD and IBM have teamed up all the time on process tech.

I doubt IBM will take over full time and product manufacturing. Doubt they have the FAB capacity anymore.

Regardless of which hdxxxx is on the die I believe they are working on allowing it to work in conjunction with any video card
by the time they mature their dream product in 2014.
Piledriver +25%cpu+50%igp looks good but steamroller will bring 20-30%{10%-15% per core}to cpu and GCN in 2013
by 2014 another 20%-30% another bump in igp
looks pretty darn good if it pans out
Intel has a lead with cpu raw power but if they should stumble on 14nm they may get run over, igp wise amd is light years ahead.
I believe the APU will eventually not only mature into a power desktop chip but will force Intel to heavily utilize HSA as well.
The idea of having cpu igp and 4 discrete all working in harmony under HSA by 2014-2015 tingles my fancy.
Though I have always bought Intel ,even I can see there is a better way than raw power. finesse always trumps raw power.
There always will be 8core+ server chips for applications that need more grunt.

The 25% bump is only vs Llano, not vs BD. BD is already about 15-20%, if not more, faster than llano as Llano is Athlon II based, not Phenom II. The 50% for the GPU needs to be taken lightly, as most of the time they talk about FLOPS, not actualy FPS. Increases in FLOPS are pointless in games.

Intel wont stumble. Especially in process tech. Who went HK/MG first? Intel. Whos just got to it? Everyone else. Whos going to have 22nm 3d Tri Gates? Intel. Who will follow? Everyone else. Intel has never had a bad process. They have had bad archs, but their process is almost always top of the line.

And I doubt Intel will allow for themself to mess up. They invest way too much money and probably test more than anyone else.

 

[triny]

Not suprising. Since we knew that the next XBox was using a PowerPC CPU with a ATI GPU, it makes sense. That and AMD and IBM have teamed up all the time on process tech.

I doubt IBM will take over full time and product manufacturing. Doubt they have the FAB capacity anymore.



The 25% bump is only vs Llano, not vs BD. BD is already about 15-20%, if not more, faster than llano as Llano is Athlon II based, not Phenom II. The 50% for the GPU needs to be taken lightly, as most of the time they talk about FLOPS, not actualy FPS. Increases in FLOPS are pointless in games.

Intel wont stumble. Especially in process tech. Who went HK/MG first? Intel. Whos just got to it? Everyone else. Whos going to have 22nm 3d Tri Gates? Intel. Who will follow? Everyone else. Intel has never had a bad process. They have had bad archs, but their process is almost always top of the line.

And I doubt Intel will allow for themself to mess up. They invest way too much money and probably test more than anyone else.

I doubt Intel will stumble either in process application but time to perfect at 14nm may be longer than expected.
none the less amd is expecting the fully matured apu to be their main chip for everything I just can't believe they are investing all the effort on apu for it not to mature into a chip for all applications it does not fit their profile .
As for 50% + gpu some will come from more cpu power some from an IGP bump though 50% is a lot that's why I was thinking that perhaps they were going to use a cut down hd7770 or 256 bit chip perhaps a cut down hd 6790
that would fit with 50%+ ?

 

[triny]

I am thinking now that someone making decisions at AMD is planning for their APU's to replace the desktop high end CPU's. I don't agree with it right now but it's what I'm seeing. If Piledriver and what comes after is sufficiently powerful, that can happen and might even be really cool and I'm thinking more of the high performance computing aspect of it rather than the graphics aspect which I believe is such a demanding subsystem it should have it's own chip. That is until we develop substantially different processes for computer chips or whatever will take the place of computer chips.

My biggest problem with all this is that if AMD quits doing the research for a high performance desktop or server CPU, how will they produce a good mid range and low end CPU? I have no problem with merging the server CPU architecture with the desktop CPU architecture. Modularity in design should allow this.

I'm running one SSD for OS and four disk drives in RAID 10. I don't think it's too much, I think it's ideal. I just need one more SSD lol. And my FT02 is beautiful. If I want a computer in a tiny box I'll build one.

at least your using the box mine has 1ssd 1hd and 1 bluray drive it looks empty

APU will be leveraged for all applications no doubt about it ,though we will see more improved server chips until
the APU comes of age .

 

[jimmysmitty]

I doubt Intel will stumble either in process application but time to perfect at 14nm may be longer than expected.
none the less amd is expecting the fully matured apu to be their main chip for everything I just can't believe they are investing all the effort on apu for it not to mature into a chip for all applications it does not fit their profile .
As for 50% + gpu some will come from more cpu power some from an IGP bump though 50% is a lot that's why I was thinking that perhaps they were going to use a cut down hd7770 or 256 bit chip perhaps a cut down hd 6790
that would fit with 50%+ ?

As I said before, its VLIW whcih means its HD6K based. I would say at max, 6670 IF its 480 SPUs. I doubt it will use a higher bus as it will use the CPUs memory bus, whatever that may be.

 

[esrever]

http://www.wired.com/wiredenterprise/2012/02/papermaster/

 

[jimmysmitty]

http://www.wired.com/wiredenterprise/2012/02/papermaster/

Not sure if its a wise move to try and enter a market thats already flooded with a lot of other already well hunkered down companies. Companies the size of or even larger than Intel. And considering how poorly AMD has been doing in X86 recently, why would they magically do better in ARM?