20-Way NVIDIA/AMD Vulkan Linux Gaming Performance Comparison

[humbug]

The kernel driver is the same - AMDGPU.
OpenGL: Catalyst (fglrx) is slower than radeonsi.
Vulkan: AMDVLK is faster than RADV. And AMDVLK is contributed to open source.

I was talking about Vulkan. Specifically the driver included in the Radeon software package. It includes proprietary shader compiler which is fast. Sometimes there is no difference but sometimes there is a significant gap.
https://www.phoronix.com/scan.php?pa...oft-1810&num=3

I am guessing that shader compiler is one of the reasons that in Windows AMD always beats Nvidia at Vulkan. Hopefully amdvlk can come up to that level in the next few months.

[boxie]

I'm having difficulty wrapping my head around GTX 1080 Ti pushing 151.36 FPS on DOTA @ 1080p.

And then pushing 147.83 FPS on DOTA @ 4k.

That's 400% as many pixels, what dirty tricks or meaning does this have.

A nice analogy would be a road. a Single lane would cope nicely with 50 cars / minute at 100kph, but trying to push 200 cars a minute @ 100kph does not work, a 6 lane freeway would have no problems.
The 1080ti would be the 6 lane freeway so coping with 50 cars / minute is no problem.

The limiting factor for this game is that it is CPU bound, it can only push so much game information a second. Turning that game information into pixels does not take lots more CPU time - the textures and objects are loaded into VRAM and then the scene is rendered.

[smitty3268]

Vulkan: AMDVLK is faster than RADV. And AMDVLK is contributed to open source.

That's false, at least as of the last tests Michael posted.

AMDVLK < RADV < AMDVLK PRO (or whatever you call the proprietary version)

[haagch]

The kernel driver is the same - AMDGPU.
OpenGL: Catalyst (fglrx) is slower than radeonsi.
Vulkan: AMDVLK is faster than RADV. And AMDVLK is contributed to open source.

The vulkan driver from amdgpu-pro is supposedly largely the same as amdvlk but it uses a proprietary shader compiler that IIRC has provided a decent performance advantage in some past benchmarks (but in some it also produced worse results).

[Tuxee]

I'm having difficulty wrapping my head around GTX 1080 Ti pushing 151.36 FPS on DOTA @ 1080p.

And then pushing 147.83 FPS on DOTA @ 4k.

That's 400% as many pixels, what dirty tricks or meaning does this have.

That the graphics card is no longer the limiting element.

[ElectricPrism]

I'm having difficulty wrapping my head around GTX 1080 Ti pushing 151.36 FPS on DOTA @ 1080p.

And then pushing 147.83 FPS on DOTA @ 4k.

That's 400% as many pixels, what dirty tricks or meaning does this have.

[microcode]

The Vega 64 minimum framerates seem way ahead of everything else (especially on Mad Max), I'd love to see the 0.1% low (a.k.a. 99.9th percentile frame time) numbers for a comparison like this.

[SvenK]

Proprietary Radeon Software driver for AMD would have performed better.

On Vega you could be right, but on the other AMD cards the proprietary driver would be worse.

[AlB80]

Radeon R7 260X: radeon.si_support=0 amdgpu.si_support=1

Should be "cik_support". Are that tests ok?

Proprietary Radeon Software driver for AMD would have performed better.

The kernel driver is the same - AMDGPU.
OpenGL: Catalyst (fglrx) is slower than radeonsi.
Vulkan: AMDVLK is faster than RADV. And AMDVLK is contributed to open source.

[humbug]

Proprietary Radeon Software driver for AMD would have performed better.