AMD - Piledriver chitchat

[scaryjim]

If not I think AMD should drop the Hybrid Crossfire feature if they cannot get it to work properly like in the past.

I remember seeing at least one review where someone managed to Hybrid a 7750 with Richland, so actually you can *already* pair an APU with GCN. To be fair to AMD, I think their hybrid crossfire setups are actually one of the best things they've done: it seems to work almost flawlessly across multiple graphics architectures and segments...

This could literally be a new part as it's GCN1.1 and the only other 1.1 part is the 7790 AFAIK.

Naah, there's also Oland, which is the basis for the R7 250 and R7 240, and that's a 384 shader die. So a 384 shader IGP would make sense. Although I guess one of the harvested A8/A6/A4 parts might end up with 384 shaders, and get paired with Oland.

For a 512 shader harvested card you'd have to lose a *lot* of Bonaire, which is - what, 896 shaders? So you'd have to fuse off 384 - a whole Oland's worth - to get a 512 shader part? I mean, even a 640 shader part harvested from Bonaire would lose a lot of silicon. It all just seems a bit random, I suppose. I can't see them continuing to produce Cape Verde dies just to have a 512 shader part, but you'd think they'd want to get more performance out of a cut down Bonaire die. I suppose a 640 shader card to supplement 512 shaders in Kaveri wouldn't be ridiculous...

[Biscuit]

Kaveri against a Core i7 4770K with a GT630:

https://www.youtube.com/watch?v=HjAM2zYNqko

Not sure how this is useful as its not a comparable or purposeful system. The CPU is a top dog CPU and the GPU is a dog sh*t GPU. no-one in their right mind would combine the two for the purposes of BF4.

I would like to see it against price and/or power usage comparable low end intel CPU+Nvidia GPU combo to be honest.

[CAT-THE-FIFTH]

Not sure how this is useful as its not a comparable or purposeful system. The CPU is a top dog CPU and the GPU is a dog sh*t GPU. no-one in their right mind would combine the two for the purposes of BF4.

I would like to see it against price and/or power usage comparable low end intel CPU+Nvidia GPU combo to be honest.

Looking at those framerates it destroys a GT630,ie, over twice as fast. Look at one of the earlier links where it says the GT630 system was 12FPS to 14FPS and the AMD Kaveri system was 28FPS to 40FPS.

A GT630 costs around £50. So a £50 to £60 Intel CPU would be an IB based G3420 dual core.

Even if you got an HD7750 GDDR3 it would nearly be £60.

If you look on Amazon 1866MHZ RAM is literally the same price as 1600MHZ stuff,and the higher speed stuff is falling in price quickly.

Add the fact that BF4 is very multi-threaded,a low end Intel CPU would be crap in comparison,as you would only be getting a dual core. Add another card for Crossfire or a better single card,and the Intel CPU would be more of a bottleneck. Kaveri can be overclocked too.

Once,the Kaveri Athlon II X4 CPUs are out,I am not sure what the point of the dual core Celerons will be for,except for games based on very old engines. Even then the AMD CPUs will be good enough,and for newer games the Pentium dual cores would be meh.

Then consider this - the 65W versions of the Kaveri A10 will be hard to beat in size and power consumption in SFF PC systems. Intel has nothing in the desktop space under £150 which could compare.

The cheapest GT3E enabled Intel part for desktops is this:

http://ark.intel.com/products/76640/...up-to-3_20-GHz

That is £181 for the tray price excluding VAT.

It also has less L3 cache than normal Core i5 CPUs.

Edit!!

BF4 does thread well,so it actually is an advantage for the Intel CPU.

AMD should have used a Haswell Core i3 which will cost around the same TBH. However,I expect it would be utterly destroyed in gaming performance using the IGP.

Second Edit!!

Mantle is not being used in the video either.

[kompukare]

Actually, i5-4570R is mobile BGA-1364 not desktop LGA1150. Some 2013 iMac's uses it but then Apple do that kind of thing (using mobile parts for desktop).

Iris Pro is very much a brute force approach: it uses lots of die-space to get its performance. Despite Intel's near limitless R&D budget, they still don't seem to be able to do a decent GPU. Die size (and more importantly transistor count since Intel are at 22nn) vs performance it isn't very impressive at all. And Intel's drivers are still very poor especially in terms of game compatibility.

[DanceswithUnix]

What kind of part number is 7850K? If that is a mobile chip, then very nice indeed, but...

The top end went from 3870K to 5800K to 6800K. If it is a true part number, it sounds like they are leaving space for a 7870K.

Also, 3.7GHz just isn't fast enough. That cedes 10% clock speed to the 6800K, which with a good IPC increase would only break even and people are expecting this to go faster.

Mind you it could be one of these odd translation things, like when Alfa re-numbered the 164 for China because in Chinese it sounded like "Chariot of Death"

Pinch of salt applied.

[scaryjim]

Actually, i5-4570R is mobile BGA-1364 not desktop LGA1150.

Probably truer to say it's an embedded product, rather than mobile. Intel list it as desktop and it's got a 65W TDP - not very mobile friendly

What kind of part number is 7850K? ... 3.7GHz just isn't fast enough. ... Pinch of salt applied.

It's definitely a desktop part. Can't see them getting 4 cores running at 3.7GHz in a mobile TDP envelope any time soon (even with the switch to 28nm). My A10-4600 (2.3GHz) can actually hit 2.7GHz across all three cores in high performance mode, so we could potentially see 3GHz stock clocked mobile parts.

How good 3.7GHz is depends on how much improvement they've got from the twin decoders and other architectural enhancements, and how much turbo headroom they've got. Like I said, my mobile chip has a 2.3GHz nominal stock clock, but it only takes setting the power option to "high performance" to have it happily sitting at 2.7GHz across all 4 cores. So if 3.7GHz is the nominal stock speed but it has a +4 bin all core boost setting that can be fired up when there is TDP and thermal headroom for it, you could be looking at a chip that runs at 4.1GHz in most situations, with single thread boost clocks going higher than that. I find it curious that AMD's graphics arm is leaning more towards advertising peak speeds while their CPU arm still stick with base speeds, but it'd fit with previous APU announcements.

Only now I'm wondering whether 720MHz for the GCN shaders is a CPU department figure or a GPU department figure

[CAT-THE-FIFTH]

Interesting:

http://translate.google.com/translat...-audio-dsps%2F
http://www.tensilica.com/news/444/33...udio-Voice-DSP

[scaryjim]

Can't see VIA trying to clone AMDs products, but they've already got a very solid foundation in embedded systems - more so than AMD - and moving to heterogenous architectures and licensing high-quality audio DSPs would make sense in a lot of those markets. Also I believe their nano architecture is actually pretty good on performance/watt already - at least from a CPU side. Perhaps involvement in HSA will help them to get preferential licensing on AMD graphics IP?

[DanceswithUnix]

Probably truer to say it's an embedded product, rather than mobile. Intel list it as desktop and it's got a 65W TDP - not very mobile friendly



It's definitely a desktop part. Can't see them getting 4 cores running at 3.7GHz in a mobile TDP envelope any time soon (even with the switch to 28nm). My A10-4600 (2.3GHz) can actually hit 2.7GHz across all three cores in high performance mode, so we could potentially see 3GHz stock clocked mobile parts.

How good 3.7GHz is depends on how much improvement they've got from the twin decoders and other architectural enhancements, and how much turbo headroom they've got. Like I said, my mobile chip has a 2.3GHz nominal stock clock, but it only takes setting the power option to "high performance" to have it happily sitting at 2.7GHz across all 4 cores. So if 3.7GHz is the nominal stock speed but it has a +4 bin all core boost setting that can be fired up when there is TDP and thermal headroom for it, you could be looking at a chip that runs at 4.1GHz in most situations, with single thread boost clocks going higher than that. I find it curious that AMD's graphics arm is leaning more towards advertising peak speeds while their CPU arm still stick with base speeds, but it'd fit with previous APU announcements.

Only now I'm wondering whether 720MHz for the GCN shaders is a CPU department figure or a GPU department figure

But the 6800K is 4.1GHz normal, 4.4GHz boost. Unless the bulk 28nm process is a bit of a clockspeed dog, they should be able to manage the same. After all the big change on the CPU is to duplicate some decoding that was already there, that shouldn't slow the clock speed down as nothing has become more complex.

[CAT-THE-FIFTH]

There was some noise the 28NM bulk process GF was developing is no so well optimised for high clockspeeds,and is more orientated towards lower power devices.

Anyway,if IPC has gone up a decent amount it should not be an issue. AFAIK,another consideration is whether AMD is looking at much lower power consumption too.

Also,a major win for GF:

http://www.theverge.com/2013/11/11/5...sors-for-apple

[scaryjim]

But the 6800K is 4.1GHz normal, 4.4GHz boost. Unless the bulk 28nm process is a bit of a clockspeed dog, they should be able to manage the same. After all the big change on the CPU is to duplicate some decoding that was already there, that shouldn't slow the clock speed down as nothing has become more complex.

Guess we won't know for sure until the actual parts come out. I think there was a rumour that they were reducing TDPs slightly, and I can't believe they'd lower the clocks without good reason. But until we get that reason, it's all a bit pointless speculation. Personally, I'm a little suspicious that there's no full shots of the slide that the footnotes comes from: that's quite unusual for an AMD presentation (the footnotes *should* be at the bottom of another slide, after all).

OTOH, a poster on the semi-accurate forums seem to concur with your opinion that the leak may be for the second-tier part, with a premium APU on top of that. 7850K/7870k would echo the naming scheme of Llano (while intimating that the second-tier part is also unlocked)...

Tell you what, if the 7850k is an unlocked 65W part with those specs (not dissimilar to the 6700 specs, after all) that'd be pretty impressive...

EDIT: significant speculation: the 6700 has an 844MHz graphics core with 384 shaders. What if running the 512 shaders makes the whole thing much more TDP-bound by the GPU? We know an 800MHz 512 shader GPU + associated memory and power delivery has a TDP of around 70W. Based on other calcs I've seen (including for the 9800GT Eco edition) I'd guess the actual GPU silicon pulls around 55W of that. Perhaps the CPU base clock is being hampered by the additional TDP demands of the GPU? That'd also fit with my specualtion that we might see more boost states available to the CPU (for instances where the GPU isn't chewing all the TDP...).

[CAT-THE-FIFTH]

Guess we won't know for sure until the actual parts come out. I think there was a rumour that they were reducing TDPs slightly, and I can't believe they'd lower the clocks without good reason. But until we get that reason, it's all a bit pointless speculation. Personally, I'm a little suspicious that there's no full shots of the slide that the footnotes comes from: that's quite unusual for an AMD presentation (the footnotes *should* be at the bottom of another slide, after all).

OTOH, a poster on the semi-accurate forums seem to concur with your opinion that the leak may be for the second-tier part, with a premium APU on top of that. 7850K/7870k would echo the naming scheme of Llano (while intimating that the second-tier part is also unlocked)...

Tell you what, if the 7850k is an unlocked 65W part with those specs (not dissimilar to the 6700 specs, after all) that'd be pretty impressive...

EDIT: significant speculation: the 6700 has an 844MHz graphics core with 384 shaders. What if running the 512 shaders makes the whole thing much more TDP-bound by the GPU? We know an 800MHz 512 shader GPU + associated memory and power delivery has a TDP of around 70W. Based on other calcs I've seen (including for the 9800GT Eco edition) I'd guess the actual GPU silicon pulls around 55W of that. Perhaps the CPU base clock is being hampered by the additional TDP demands of the GPU? That'd also fit with my specualtion that we might see more boost states available to the CPU (for instances where the GPU isn't chewing all the TDP...).

Kaveri is 47% GPU,so even taking into consideration the TrueAudio block,the CPU cores must be a decent size larger in terms of transistor number.

[scaryjim]

Kaveri is 47% GPU ...

Maybe, but that doesn't tell you about distribution of TDP or power consumption in the silicon. Given the main difference between the A10-6700 and A10-6800k is 400MHz on the CPU, I'd suggest that Richland devotes a reasonable proportion of its TDP budget to the CPU, particularly in the 100W desktop parts. It's possible that in Kaveri the GCN cores are actually given higher priority on the TDP budget, which means the base clock for the CPU cores is kept in check to ensure the GPU has enough TDP when it needs it. I guess if they're ratifying the memory controller to a higher speed that might have a TDP impact that could limit other parts of the chip too (after all, the reason Hawaii got a 512bit memory bus is because it uses slower PHYs for the memory, reducing actual speed, but also reducing TDP and silicon space). A memory controller that's capable of running faster DDR3 might be more complex, take up more silicon and generate more heat: in which case something else will have to give up that TDP headroom.

[CAT-THE-FIFTH]

Maybe, but that doesn't tell you about distribution of TDP or power consumption in the silicon. Given the main difference between the A10-6700 and A10-6800k is 400MHz on the CPU, I'd suggest that Richland devotes a reasonable proportion of its TDP budget to the CPU, particularly in the 100W desktop parts. It's possible that in Kaveri the GCN cores are actually given higher priority on the TDP budget, which means the base clock for the CPU cores is kept in check to ensure the GPU has enough TDP when it needs it. I guess if they're ratifying the memory controller to a higher speed that might have a TDP impact that could limit other parts of the chip too (after all, the reason Hawaii got a 512bit memory bus is because it uses slower PHYs for the memory, reducing actual speed, but also reducing TDP and silicon space). A memory controller that's capable of running faster DDR3 might be more complex, take up more silicon and generate more heat: in which case something else will have to give up that TDP headroom.

The 47% is from AMD!

[Biscuit]

Looking at those framerates it destroys a GT630,ie, over twice as fast. Look at one of the earlier links where it says the GT630 system was 12FPS to 14FPS and the AMD Kaveri system was 28FPS to 40FPS.

A GT630 costs around £50. So a £50 to £60 Intel CPU would be an IB based G3420 dual core.

Even if you got an HD7750 GDDR3 it would nearly be £60.

If you look on Amazon 1866MHZ RAM is literally the same price as 1600MHZ stuff,and the higher speed stuff is falling in price quickly.

Add the fact that BF4 is very multi-threaded,a low end Intel CPU would be crap in comparison,as you would only be getting a dual core. Add another card for Crossfire or a better single card,and the Intel CPU would be more of a bottleneck. Kaveri can be overclocked too.

Once,the Kaveri Athlon II X4 CPUs are out,I am not sure what the point of the dual core Celerons will be for,except for games based on very old engines. Even then the AMD CPUs will be good enough,and for newer games the Pentium dual cores would be meh.

Then consider this - the 65W versions of the Kaveri A10 will be hard to beat in size and power consumption in SFF PC systems. Intel has nothing in the desktop space under £150 which could compare.

The cheapest GT3E enabled Intel part for desktops is this:

http://ark.intel.com/products/76640/...up-to-3_20-GHz

That is £181 for the tray price excluding VAT.

It also has less L3 cache than normal Core i5 CPUs.

Edit!!

BF4 does thread well,so it actually is an advantage for the Intel CPU.

AMD should have used a Haswell Core i3 which will cost around the same TBH. However,I expect it would be utterly destroyed in gaming performance using the IGP.

Second Edit!!

Mantle is not being used in the video either.

I dont dispute any argument that the platform makes sense at this price point, my point is that as an active demonstration, its a bit rubbish. Why they didnt use an example such as the ones you detailed, is slightly beyond me.

Edit... whoahhhh loads of discussion from when i started typing to when i finished. Stupid job.

[CAT-THE-FIFTH]

I dont dispute any argument that the platform makes sense at this price point, my point is that as an active demonstration, its a bit rubbish. Why they didnt use an example such as the ones you detailed, is slightly beyond me.

Edit... whoahhhh loads of discussion from when i started typing to when i finished. Stupid job.

I having a feeling its more the case,here is a £280 to £300 worth of competitors products,which are slower in the game we run.