RadeonSI Picks Up Primitive Culling With Async Compute For Performance Wins

[AsuMagic]

This is what happens when games are "tweaked" to do unnecessary computations to benefit a certain "green" gpu maker

source?

edit: just beyond crysis 2 and decade old examples

[Bitter]

https://www.reddit.com/r/pcmasterrace/comments/36j2qh/nvidia_abuse_excessive_tessellation_for_years/

[lucrus]

Well, a RX480 that comes out ahead of a GTX 1080Ti in games would just be groundbreaking news. Hard to believe until I see it myself, but I have high hopes.

[Bitter]

Well, a RX480 that comes out ahead of a GTX 1080Ti in games would just be groundbreaking news. Hard to believe until I see it myself, but I have high hopes.

Probably not, but the practice of drawing hidden surfaces might still be on their agenda.

[artivision]

This is what happens when games are "tweaked" to do unnecessary computations to benefit a certain "green" gpu maker

Doesn't happening that way friend, there is no conspiracy. The thing is that Gameworks use low overhead extensions provided by Nvapi and the game developer cannot do anything about it. From here two things can be done: 1) Amd support some of them outside D3D and DXVK some of them upon Vulkan for all GPUs. 2) Benchmarkers stop using Gameworks effects or at least stop saying that they test D3D wile they actually testing D3D+Nvapi.

[schmidtbag]

As much as I love everything Marek does, those results seem waaaay too good to be true. A 30% improvement is a hefty achievement, but the Radeon 7 nearly got a 300% improvement. That's insane, when you consider that GPU was running approximately where it was expected to run before async compute.

[soulsource]

The cool thing here: Although the feature is disabled by default for consumer chips, we can still play around with it:

People using the Pro graphics cards can disable this by setting AMD_DEBUG=nopd, and people using consumer graphics cards can enable this by setting AMD_DEBUG=pd. So you always have a choice.

[Ground0]

Damn another day i would try git version ... was going back to semi stable opensuse Tumbleweed builds (Performance was enough to going back from git Versions) When comes the gaming test

[boxie]

As much as I love everything Marek does, those results seem waaaay too good to be true. A 30% improvement is a hefty achievement, but the Radeon 7 nearly got a 300% improvement. That's insane, when you consider that GPU was running approximately where it was expected to run before async compute.

This is an Async compute optimisation on a workload that is very unoptimised. I would suggest that the subset of games that will benefit is small and the margins even smaller.

though, I do wonder how much Async compute DXVK uses and if this optimisation technique could apply to RADV (if it has not already been done)?

[tuxd3v]

As much as I love everything Marek does, those results seem waaaay too good to be true. A 30% improvement is a hefty achievement, but the Radeon 7 nearly got a 300% improvement. That's insane, when you consider that GPU was running approximately where it was expected to run before async compute.

I agree, Fury X, Vega56, and VII were the ones profiting the most..
But if you look into there from a Hardware perspective...the hardware are already there, the drivers are the ones to mature..
But I think the games will not see a so insane big jump in performance,
However, its a big jump though!

Kudos for Marek for that!