AMD Mantle API [updating]

Am I reading the StarSwarm-slides correctly - Mantle is performing just like the game would perform without any API overhead?
I'm assuming it's GPU-limited with Mantle and CPU-limited without. It's not unrealistic to see those sorts of performance differences in a demo like that.
 
Unlikely, if it was Mantle and discrete GPUs used, surely it would say Mantle, and in the next slide why would they state only A8 if it was A8+discrete, when they clearly state in other slides if discrete GPU is used?

I prefer to believe that someone forgot to put the complete system info on a slide, relying on the end slides (which are referred on those slides) than to believe that they are "making numbers" without actually being able to run the Oxide demo (which is the source of the performance numbers, AFAICT) OR that they somehow made drivers for an Intel IGP themselves, porting Mantle for an Intel platform...
 
AMD-Catalyst-14.1-BETA.jpg


http://wccftech.com/amd-catalyst-14...-mantle-trueaudio-frame-pacing-fix-gpus-apus/
 
Just a question about Mantle and dual GPU systems.

I did just buy an r7 260x (cheap, for having nice audio with thief, since I dont really have time to play games, but want to play that one... ).

With Mantle, can I pair it with an r9 290 or a 7950 and still have good scaling or is it worth less?

...do I need PCI3 or can I still use my PCIE2 MB? I have an old OC nehalem, and still I dont feel the need for an haswell; clock-to-clock AMD's SR seems to still be inferior to that (am I right?).
 
It's possible Mantle would allow better asymetrical Crossfire performance, there were some (or one) AMD slide to that effect. But maybe not all Mantle games. The promise is more about running GPGPU tasks on one GPU (which can be the Kaveri's IGP) and graphics tasks on the bigger GPU efficiently, too ; though assuredly graphics performance scaling should increase somewhat. But it can turn out terrible in the end.

For what we know, it's foolish to do such a severe asymetric crossfire, it will be useless in all current games and the result in future games is unknown but probably next to useless unless proved otherwise. Ideal result is R7 260X + R9 290 or 7950 behaves as two R7 260X for graphics performance, with additional horsepower used for gaming GPGPU tasks, but it might not be the case.
A lone 290 would be just much easier.

PCIe 2.0 16x would not be that bad, as modern configurations tend to run a pair of graphics at PCIe 3.0x 8x.
Kaveri + big GPU (Pitcairn or Tahiti or Hawaii) is maybe nice, in a scenario where main graphics run only on the big GPU (and you have PCIe 3.0 16x)
 
Kaveri + big GPU (Pitcairn or Tahiti or Hawaii) is maybe nice, in a scenario where main graphics run only on the big GPU (and you have PCIe 3.0 16x)

Interesting for sure, but we'll kind of have to wait and see won't we? Right now the iGPU on AMD APUs isn't used if you have a discreet card, right?
 
For what we know, it's foolish to do such a severe asymetric crossfire

ah, sorry, I was not clear. cfx was excluded, I'd never bother to use it. Just mantle games for me. I have time to play 1-2 games/year, if I am lucky :/

It might be a good pairing then, 7950 for 3dfx and 260x for trueaudio&additional GPGPU. Should it work, then (no cfx cable, only PCIE2)?

Ok, now I only need to compare SR's IPC to nehalem and see what's worth. Thanks.
 
For what we know, it's foolish to do such a severe asymetric crossfire, it will be useless in all current games and the result in future games is unknown but probably next to useless unless proved otherwise. Ideal result is R7 260X + R9 290 or 7950 behaves as two R7 260X for graphics performance, with additional horsepower used for gaming GPGPU tasks, but it might not be the case.
A lone 290 would be just much easier.

The question was not if a R7 260X would pair up well with a HD7950 or a R9 290 in Crossfire in DirectX.
I think that part of the question was pretty obvious, even more considering that this is the Mantle thread.

From what I know, the answer to the question is: we still don't know what kind of multi-gpu combinations and performance boosts will be made available with Mantle.

I'd guess it'll all come down to either Mantle's Crossfire will still use AFR or not. Predicting per-frame performance between GPUs with different capabilities and memory sizes will always be very hard AFAIK, which is why nVidia doesn't allow SLI between anything other than two identical graphics cards.

The cherry on top would be to have Mantle natively (therefore flawlessly) support something akin to Lucid Hydra's solution.
That way, a R7 260X + R9 290 with normalized clocks could be similar to a single GPU with 54 GCN Compute Units, 216 TMUs, 80 ROPs and 6GB of memory and the 260X would give a healthy ~30% performance boost to the 290.
Load balancing on the CPUs and its induced latency would probably be the hardest thing to handle, of course.


AMD would benefit greatly from this, especially for having their APUs to significantly boost the performance of mobile discrete GPUs.
 
Judging from what we know of Mantle so far, Mantle games will intrinsically have *no* Crossfire support unless the developers explicitly put it there themselves. AMD could probably provide some sort of library to help with a simple AFR implementation, but if you want something asymmetric, that's fundamentally something the game developer is going to have to do.

At least with the way it's exposed in Mantle it's *possible* to do fancy stuff, but the question still remains how much effort a given developer is willing to put into a very small fraction of the market, even if it is the halo end. It's not trivial to write asymmetric multi-gpu/accelerator code, particularly if you want it to do something reasonable across a wide range of hardware.

Maybe some of the engine folks will see it as further differentiation for their engines, but while it would be fun to mess around with, if I was making a game with a deadline it would not be high on my list.

That way, a R7 260X + R9 290 with normalized clocks could be similar to a single GPU with 54 GCN Compute Units, 216 TMUs, 80 ROPs and 6GB of memory and the 260X would give a healthy ~30% performance boost to the 290.
Complete pipe dream man... even with something like NVIDIA's parallelized geometry pipe (that AMD has not duplicated) there's still giant crossbars on GPUs that obviously aren't applicable if your memory isn't on the same chip... or even the same PCI-E device. There's reasons why people manufacture huge single-GPUs and do AFR. Obviously if there was any scalable way to just combine multi-chip resources no one would make big GPUs.
 
From what I know, the answer to the question is: we still don't know what kind of multi-gpu combinations and performance boosts will be made available with Mantle.

I'd guess it'll all come down to either Mantle's Crossfire will still use AFR or not.
As Andrew Lauritzen already said, a Mantle game will use the multi GPU rendering method, that the developer puts in their game, so in doubt none. Of course this has the potential that some studio may come up with a super clever method tailored to their engine far better than AFR could ever hope to be, but considering the relatively small market of multi GPU systems, I wouldn't hold my breath.
 
Interesting for sure, but we'll kind of have to wait and see won't we? Right now the iGPU on AMD APUs isn't used if you have a discreet card, right?

Depends what level discrete card you have, the APUs support Dual Graphics with certain dPGUs
 
Maybe some of the engine folks will see it as further differentiation for their engines, but while it would be fun to mess around with, if I was making a game with a deadline it would not be high on my list.

Then I guess it's up to developers.
Are you sure that AMD won't be capable/interested in implementing load-balancing Multi-GPU in the OS itself, in the future?


Complete pipe dream man... even with something like NVIDIA's parallelized geometry pipe (that AMD has not duplicated) there's still giant crossbars on GPUs that obviously aren't applicable if your memory isn't on the same chip...

Lucid's results with mixed-vendor graphics cards weren't a pipe dream.
When the solution worked, of course.
 
I know its what we can call like an extreme case, but well seeing the difference in fps between the dx phase and mantle... ouch.
 
Did it mention the GPU used for that video? I know it mentioned the i7 920 CPU. I particularly liked the woman narrating who sounded like she didn't have the foggiest what she was saying. The jump from DX to Mantle was certainly noticeable, and a little more than at least I expected.
 
Did it mention the GPU used for that video? I know it mentioned the i7 920 CPU. I particularly liked the woman narrating who sounded like she didn't have the foggiest what she was saying. The jump from DX to Mantle was certainly noticeable, and a little more than at least I expected.

If you look at the video, is Radeon R9 200 series and i7 980 cpu.
I am not an expert, and english is not my first language, but I see "too much" emphasis in CPU. I would like to see this test on the new Kaberis.
 
If you look at the video, is Radeon R9 200 series and i7 980 cpu.
I am not an expert, and english is not my first language, but I see "too much" emphasis in CPU. I would like to see this test on the new Kaberis.

Thanks, I watched it last night but haven't got access to the video at the moment.

Yeah sorry I miss typed the CPU, they even made a point of mentioning the old 6 core CPU.
 
Back
Top