Radeon Gaming/Graphics Performance: Windows 11 vs. Linux GPU Benchmarks

[phoronix]

Phoronix: Radeon Gaming/Graphics Performance: Windows 11 vs. Linux GPU Benchmarks

With the end of the year upon us it's a great time to see how the Windows vs. Linux gaming performance is looking as we enter 2023. In particular, it's interesting on the AMD Radeon side with the open-source Linux graphics driver stack having made great gains this year thanks to the continued investment by AMD and heavy contributions by Valve to the Mesa RADV Vulkan driver that is used by the Steam Deck and commonly in general by Linux gamers. Here is a look at the Windows vs. Linux GPU performance both for the mature RDNA2 support as well as the recently-released RDNA3 graphics.

https://www.phoronix.com/review/radeon2022-windows-linux

[andre30correia]

I will like to see with feral gamemode working if something really changes

[dimko]

More or less equal stuff where it matters.(i really dont care about high figures on 1080p. as I dont use it for many years now, on higher resolutions things start to be more equal)

[dimko]

I will like to see with feral gamemode working if something really changes

Also, would be curious to compare energy management. I suspect that on lower resolutions, power consumption may be less on Linux side of things(hence, money saving!)
Most people don't care past 144hz anyway.

High resolutions are almost equal anyway. And Unigine benchmarks are kind of legacy by now. OpenGL is practically not used by linux gamers, since everyone and their grandma are using Vulkan and proton(which uses Vulkan).

[skeevy420]

I will like to see with feral gamemode working if something really changes

That usually only matters if you're being throttled due to your CPU governor. If you're already in a performance mode from somewhere else it won't matter very much. Instead of gamemode, I'd be curious about is if changing from intel_pstate performance to acpi performance would have an impact.

Michael For some reason I just realized that there aren't divider lines between file systems, kernel versions, etc between Windows 11 and Ubuntu in the chart.

Is the reported disk size discrepancy between Linux and Windows due to 1000 and 1024...kb and kib?

Also, the 6800 and 7900 are using different drivers. I assumed they were using the same driver until I double checked the table. Are you using the drivers from AMD.com or whatever windows provides?

[skeevy420]

Also, would be curious to compare energy management. I suspect that on lower resolutions, power consumption may be less on Linux side of things(hence, money saving!)
Most people don't care past 144hz anyway.

High resolutions are almost equal anyway. And Unigine benchmarks are kind of legacy by now. OpenGL is practically not used by linux gamers, since everyone and their grandma are using Vulkan and proton(which uses Vulkan).

The one result I care about isn't in these tests: the lows. Only because high frame rates only tell part of the story and don't necessarily paint a complete picture of what is going on. Once a game starts running way over 60fps (guess my refresh rate) I'm more interested in what the average low frame rates are, frame timings, and things of that nature. It's cool that a game can hit 300fps and averages 230fps. Not cool if it dips down to 37fps a lot and has horrible frame times what that occurs so half the time you're in a choppy lag fest.

[Michael]

That usually only matters if you're being throttled due to your CPU governor. If you're already in a performance mode from somewhere else it won't matter very much. Instead of gamemode, I'd be curious about is if changing from intel_pstate performance to acpi performance would have an impact.

Michael For some reason I just realized that there aren't divider lines between file systems, kernel versions, etc between Windows 11 and Ubuntu in the chart.

Is the reported disk size discrepancy between Linux and Windows due to 1000 and 1024...kb and kib?

Also, the 6800 and 7900 are using different drivers. I assumed they were using the same driver until I double checked the table. Are you using the drivers from AMD.com or whatever windows provides?

They are from AMD.com... If you go there you'll see depending upon selecting RX 6000 or RX 7000 are different driver versions (as mentioned in the article) and trying to use the RX 6800 XT on the RX 7900 XT driver had not detected any GPU.

[piotrj3]

More or less equal stuff where it matters.(i really dont care about high figures on 1080p. as I dont use it for many years now, on higher resolutions things start to be more equal)

Wouldn't say so. Hitman 3, Quake 2 RTX, X-plane are 3 titles that have any significance (As they are more representantive for new titles) and for them diffrence significant exist. And there seems to be (% wise) greater perfomance diffrence for 4k. So I would say something is wrong here, and knowing it is issue in 4k, it is probably not CPU bound issue like on side of game.

Also keep in mind we talk about Vulkan mostly here, and Vulkan is supposed to be more exact and more 1-1 with operations directly done on GPU. The utilization in general for Vulkan is supposed to be good from start, it is not OpenGL or old directx when you can make a lot of magic optimalizations.

Seeing in hitman 3 4k ultra diffrence of 20% is concerning. Quake 2 RTX shows issues with VK_KHR_ray_tracing_pipeline on linux, and Xplane on 7900XTX shows issues in performance. Same goes for wierd things like Strange Brigade having FPS cap on linux etc.

[RejectModernity]

Michael should add more games, this lineup looks weak. Something like Doom Eternal, RDR2, Forza Horizon 5, Marvel's Spider-Man, Halo Infinite, Cyberpunk 2077 (without RT, since there were bugs). These games are kinda demanding and play on Linux pretty well.

[Michael]

Michael should add more games, this lineup looks weak. Something like Doom Eternal, RDR2, Forza Horizon 5, Marvel's Spider-Man, Halo Infinite, Cyberpunk 2077 (without RT, since there were bugs). These games are kinda demanding and play on Linux pretty well.

As mentioned in the article, Cyberpunk 2077 has rendering issues with RADV on RDNA3 at the moment. For the other games last check they could not be automated in a nice manner for the benchmarks.