16-Way NVIDIA GeForce vs. AMD Radeon Comparison - August 2017

[bridgman]

I have been waiting for AMD's new in-house OSS drivers to support my GCN 1.0 HD 7970 card.

But it now looks as if RadeonSI might be a better bet? Who would've thought...

What exactly have you been waiting for ?

As agd5f said, if you are running the OSS drivers on your hardware today you are using radeonsi for 3D already, albeit with the radeon kernel driver instead of the amdgpu kernel driver we use with newer HW. The only thing I can think of that you would get from amdgpu kernel driver vs radeon driver is radv support, but AFAIK radv is also a work in progress for SI.

[marek]

radeonsi vs amdgpu vs radeon confusion: https://people.freedesktop.org/~mareko/dot_mesa.png

radeonsi is only a piece of the puzzle. amdgpu and radeon are other pieces. You can choose some pieces (amdgpu vs radeon), but you'll always use radeonsi in both cases.

[edgar444]

Quick Q, is that fps-lock set on 30 for civ6.

Don't have the game, but if that fps on *1080p* is real unlimited framerate, then I'm really surprised... o_o

[ermo]

What exactly have you been waiting for ?

As agd5f said, if you are running the OSS drivers on your hardware today you are using radeonsi for 3D already, albeit with the radeon kernel driver instead of the amdgpu kernel driver we use with newer HW. The only thing I can think of that you would get from amdgpu kernel driver vs radeon driver is radv support, but AFAIK radv is also a work in progress for SI.

Good Vulkan + OpenGL performance for my old GPU under Linux using FLOSS drivers.

I was under the impression that I had to wait for (non-experimental) amdgpu GCN 1.0 support and the open-sourcing of AMD's currently proprietary Vulkan driver for this to become a reality?

In my original post however, not only did I confuse 'radeon' (the old <GCN 1.2 kernel drm driver) with RadeonSI, but I also assumed that RADV would work with radeon (I now understand that RADV is an interim alternative to AMD's currently proprietary Vulkan userpace driver and that RADV needs amdgpu).

I guess I missed the 4 PM clue train *sigh*.

[Apopas]

Since now the fps are up, I'm waiting for the prices to drop a bit and then I'll grasp a RX 570.

[FishPls]

Quick Q, is that fps-lock set on 30 for civ6.

Don't have the game, but if that fps on *1080p* is real unlimited framerate, then I'm really surprised... o_o

It should be unlimited framerate, the port is just _really_ CPU-intensive and basically bottlenecks fast no matter what you're running it on. I asked many months ago from Aspyr whether we could see any CPU usage improvements, their answer was basically "It's possible, but don't expect it".

[valici]

Good results. Looks like Radeon does better at lower settings and resolution. Why would that be? Lower overhead than Nvidia?

[pinguinpc]

Good tests but when stay 1080p? (Most common resolution used actually) tests

Without forget more test case race driver grid, dirt rally, dirt showdown, tomb raider, shadow of mordor, shadow warrior, spec ops the line, anima gate of memories, witcher 2, saint row games and many others (various require manual test)

Last titles are very interesting because amd have serious problems with various vp ports, shadow of mordor and others

4K actually dont have much relevance mainly display costs

[justmy2cents]

kudos to AMD developers. 580 so consistently beating shit out of 1060 is really nice

if only bitcoin miners wouldn't be hogging all stocks. damn thing is near impossible to get. i wanted to change both my gpus (one for 1060 and one for 580 since i need both vendors), but i simply have no luck

[bridgman]

Good Vulkan + OpenGL performance for my old GPU under Linux using FLOSS drivers.

Got it... so you should be in good shape for OpenGL performance today (assuming you run comparably new code to what Michael used) but not sure how solid SI support is in upstream amdgpu kernel driver (which you need for radv).

That said, I don't remember seeing amdgpu/SI bug reports recently but since it's not default I don't have a good feeling for how much it is being used. I know Dave/Bas are fixing SI bugs in radv so it must be working well enough for their testing.

Is anyone else on the thread running amdgpu+radv on SI today ?