How do we fix Nvidia's lack of async compute?

File: Nvidiots BTFO.jpg (1MB, 3451x1362px) Image search: [Google]

1MB, 3451x1362px

We dont need to.

>>56571610

exactly since they fundamentally do NOT support dx12, so there is nothing to fix.

>>56571589
>all kepler and maxwell chips
But what about Pascal?

>>56571837
Pascal is Maxwell die shrink with bit of driver improvements

>>56572095
Except not.

there's no gain to concurrently executing massively parallel code, especially when you don't have utilization issues like AMD does (because the front end is bottlenecked in hawaii and newer gcn chips).

>>56572248
Except it is.

>>56571589
>he doesn't understand what divergence is

>>56571589
We don't. Nvidia will, if they need to. Or they won't. Whatever works for them I guess.

>>56571589
They don't need to. Nvidiots will buy it regardless.

Nvidia is 10x the size of AMD and much more relevant in the GPU world. If they decide Async isnt important, it's not important.

It's like saying Apple should do something with the iPhone because some chinkphone nobody cares about did it.

>>56573100
AMD make every good choice in GPU world.
Async, Mantle (who became Vulkan).

Nvidia just take your money and fuck all of you

>>56573148
Why are those good?

>>56573186
Example with RX 470:
+26 fps with Vulkan.

Nvidia card gain nothing, they are shitty in async

( http://www.guru3d.com/articles_pages/msi_radeon_rx_470_gaming_x_8gb_review,14.html )

>>56573262
All that tells me is that AMD has shit support for openGL.

>>56573282
*facepalm*

you seems don't know the diff between sync and async

>>56573262
You are incredibly naive and misinformed if you think the only difference between OpenGL and Vulkan is async compute. It's not even the biggest deal. It wasn't that long ago when everyone was hyping up DX12 and Vulkan for an entirely different reason, namely parallelism and reduced CPU overhead.

AMD's driver is notoriously shit in DX9/11/OGL and this can be easily seen by examining the behavior of AMD's graphics cards. Scenes with a lot of draw calls perform poorly despite having low GPU load. This is the absolutely major bottleneck that DX12 and Vulkan remove for Radeon GPUs and why you sometimes see such huge performance increases, because Radeons are pretty damn nice hardware but the driver was always shit (even if it isn't unstable shit). Async is just the cherry on top.

NVIDIA however doesn't suffer from this problem as significantly as AMD and as such DX12/Vulkan don't really manage to squeeze any performance increase on systems with fast CPUs, because you're GPU bottlenecked both in DX11 and in DX12/Vulkan. Still, even with NVIDIA cards I've seen a few scenes in ROTTR where simply switching from DX11 to DX12 gets me a very solid boost from ~40FPS to 60+, but it's only in a few specific areas.

>>56573317
You are talking about things you don't understand and will probably never understand. This has very little to do with async compute and a lot to do with AMDs poor driver optimization. Unlike DX11, it's up to the vendor of graphics hardware to package it with their driver and AMD is doing and has always done a mockery of a job of it.

Doesn't help that AMD rendering pipeline is single threaded in both DX11 and OGL, which means shitty performance.

>>56571589
trash gsync and work together with amd for a common platform