AMD Mantle API [updating]

PC DirectX 11 (or OpenGL compute) do not support multiple asynchronous compute queues.

Hopefully OpenGL 5.0 and DirectX 12 will introduce the support, allowing us finally fully utilize the modern GPUs on PC. Currently the only way to take advantage of multiple compute queues of the modern NVIDIA and AMD GPUs is to use CUDA or OpenCL (or Mantle?, I don't know?)

Perhaps @Dave Baumann would be kind enough to advise us on that one?
 
But the drivers do.
Automatic splitting of work from a single "thread" (=single queue) to multiple "threads" (=multiple queues) is not realistically possible in the generic case. You could do some coarse scale resource dependency analysis (DirectX resource granularity), but this is again not possible in the generic case (as it's impossible to know which cache lines each compute shader will read / write before actually executing that shader). Modern engines already have moved to sharing bigger buffers between multiple items of data (to reduce state change cost). Eventually we have everything in a big huge buffer (fully managed by the engine code) on PC as well.

I suppose the driver teams could do case-by-case optimizations to titles that have simple enough processing pipelines, or even some automation (as the amount of different resources in DirectX applications is quite small after all). If you have any inside knowledge about this, it would be nice to hear about it :)
 
I suppose the driver teams could do case-by-case optimizations to titles that have simple enough processing pipelines, or even some automation (as the amount of different resources in DirectX applications is quite small after all). If you have any inside knowledge about this, it would be nice to hear about it :)
I don't have an inside info - merely an hypothesis that the driver team can perform an analysis to determine parallelism in compute hotspots.

This is under the presumption that games have reached the stage where there is more than one compute task per frame and/or per tick.

It might simply boil down to striping single compute tasks across ACEs - e.g. to make compute and graphics co-execute in a more fine-grained, amicable fashion.

All guesswork - but I'm hoping that a comparison of low-count and high-count chips in games might indicate something. And whether there's an effect upon CPU bottlenecks.
 
Dragon Age: Inquisition offers support for Mantle, but according to one review still needs work.

For those of you that noticed it, Dragon Age Inquisition supports AMD Mantle. AMD Mantle up-to now has proved to reduce CPU overhead creating a little more headroom for slow processors versus gaming FPS. Typically this should help with GPU either very slow systems, or very fast GPU systems with Crossfire. As we have experienced often now, Mantle was showing negative scaling with a Radeon R9 295x2. If there is any GPU that should have benefited from Mantle, it was this solution.

Currently Mantle does work, but at performance decreases of up-to 10% in Full HD it just does not make any sense using it. We also noticed that when mantle was enabled, the scene load times took MUCH longer. For now we advise you to leave it disabled as it needs work.

http://www.guru3d.com/articles-page...-graphics-performance-benchmark-review,9.html
 
Dragon Age: Inquisition offers support for Mantle, but according to one review still needs work.



http://www.guru3d.com/articles-page...-graphics-performance-benchmark-review,9.html

I have about 33 hours into this game so far, have been using Mantle the whole time. Just tried with DX, and the load times seem to be about the same, I have it loaded on a hdd if that makes a difference. One other thing I noticed, once I have been to a area it never loads that area again from the hard drive, though it still takes a little time to load.
 
Dragon Age: Inquisition offers support for Mantle, but according to one review still needs work.



http://www.guru3d.com/articles-page...-graphics-performance-benchmark-review,9.html

Dont forget that it use a 5960X 8cores / 16thread CPU ).. and maybe it is question of a patch because i remember the first review of Dragon Age + Mantle, it had 10fps more in Mantle than on Dx11, at 2560x1440p...( a resolution at ultra setting who is far to be CPU bottleneck ) .

The problem there, at least for me, is that i have too a good CPU ( 4930K highly overclocked ), with a 1440P, monitor, so i should too be only worried by gpu bottleneck.. ( 1080P>1400p, be read to take an hit in performance )
 
I've found that the cutscenes in the game stutter relentlessly with Tessellation on Ultra. Lowering it to High fixes that, leaving all other settings at max. This is with a i7-3770K, 16GB RAM, GTX970, game installed on SSD.

The game looks to perform fairly well on my other rig with HD7950. I'll have to give Mantle a go and see if the problems are restricted to Crossfire.
 
My brother had to turn Mantle off on his 7950 because it made the load times extremely long.
 
Does Mantle only help newer GCN cards? Do the newer cards like Hawaii have the same load time problem?
 
Does Mantle only help newer GCN cards? Do the newer cards like Hawaii have the same load time problem?

Yes, the load times increase on all GPUs.

The loading screen seems to have some weird interaction with the frame rate, so I'm not sure it's a Mantle problem.
 
Back
Top