Recent benchmarks indicate Linux surpasses Windows by 20% in World of Warcraft and 6% in Doom Eternal
Recent benchmarks indicate Linux surpasses Windows by 20% in World of Warcraft and 6% in Doom Eternal
The benchmark numbers cited below below were done by Flightless Mango, the author of MangoHUD: https://github.com/flightlessmango/MangoHud The linked pages have charts and graphs, plus links to YouTube videos that permit image quality comparisons. At present, there is no one better at doing video game performance comparisons between Linux and Windows than Flightless Mango as far as I know. He records samples of frame time data from the entire run, which is then shown in the graphs and used to calculate numbers. These days, Linux relies on DXVK to translate Direct3D 9, 10 and 11 into Vulkan, so the performance of the Linux Vulkan driver is extremely important in terms of how it does in comparison to Windows. A couple months ago, Nvidia released a Vulkan beta driver that increased performance in World of Warcraft by 25%: https://flightlessmango.com/benchmarks/-5CbWvvY7EM The result is that using DXVK to translate Direct3D 11 into Vulkan on Linux enables it to perform ~20% higher than windows: https://flightlessmango.com/benchmarks/V2_LlSaYNUo In additition, Doom Eternal is a native Vulkan game, so there is no API translation overhead. Nvidia seems to have added a workaround for the game to get some of the performance that ID Software seems to have left on the table. The result is a 6% performance increase over Windows: https://www.gamingonlinux.com/articles/n...lay.16343/ https://flightlessmango.com/benchmarks/h-XnlUMfkjM It should be noted that the comparisons here are not fair to Windows for a couple reasons. One is that you likely could install the Vulkan beta driver on Windows for the same workaround to be used in Doom Eternal. You likely could use DXVK and install the vulkan beta driver on Windows to play World of Warcraft with a similar boost in performance. If you do those things, I would expect that the two would come close to performance parity. However, this is definite progress compared to a couple years ago where you would never see a Windows game performing better on Linux than on Windows (or even on par) under any circumstance. It should also be noted that the portrayal of Linux is somewhat inaccurate. It is using the Nvidia vulkan beta driver, but that is not what most users have installed (although they could install it). It also is using "fsync" patches from Valve meant to reduce CPU overhead in Wine. These patches are not merged in the mainline kernel at this time. However, these two things are things that will likely be avaliable on Linux systems in the future, so these benchmark results give us a glimpse into future Linux performance. There are a couple more differences between how the Linux system was setup and how typical Linux systems are setup. The first is that wine staging was used. This doesn't really matter that much because Proton has basically all of the patches needed for performance. However, the second is that a different CPU scheduler is used, which matters a bit more. Here is an older benchmark (from before the newer Nvidia vulkan beta driver that improved things for Linux so much) that shows the difference between the default CFS and the PDS used in the benchmarks: https://flightlessmango.com/benchmarks/phA-M1biogE It shows that the impact of the CPU scheduler is between 3% to 5% in the games tested. The PDS scheduler is not in the mainline Linux kernel and likely never will be. That being said, when things are GPU bound, the CPU scheduler likely won't matter very much, so the added benefit from a different CPU scheduler will not matter very much. It should also be noted that the games results that I put into the subject were cherry-picked. Other benchmarks done by flightlessmango since the nvidia vulkan driver that improved performance show Linux getting 93% to 97% of the performance of Windows when using DXVK with Nvidia hardware: Resident Evil 2 at 93%: https://flightlessmango.com/benchmarks/TijlTJC7KYc Deus Ex: Mankind Divided at 96%: (although only 75% with the native port, which suggests it is not well done) https://flightlessmango.com/benchmarks/mQZd-pA21CY Control at 97%: https://flightlessmango.com/benchmarks/1D82_XEwvpA If you look at the VKD3D numbers where Direct3D 12 is used, the performance Linux obtains is even worse. However, VKD3D is not as mature as DXVK, so that is to be expected. Coincidentally, the Direct3D 12 code paths in all games are slower than the Direct3D 11 codepaths, so Linux is not gaining an advantage over Windows from doing a purely Direct3D 11 comparison in those three. The above mentioned World of Warcraft does see a slight improvement from Direct3D 12, but that just kept the Linux performance number from being more than 20% higher. It should be said that on AMD hardware, Linux tends to perform more favorably compared to Windows. You can see this in the benchmarks that were done on AMD graphics hardware: Rage 2 at 100%: https://flightlessmango.com/benchmarks/4BmZt7v8dpY Assassin's Creed: Odyssey at 84%: https://flightlessmango.com/benchmarks/3dyQUMebYGc A Hat in Time at 148%: https://flightlessmango.com/benchmarks/2YXm3Okfs0M Minecraft at 122%: https://flightlessmango.com/benchmarks/ytgDMgYL0eo Coincidentally, the benchmark results show that using DXVK with AMD hardware on Windows is not as good as using it on Linux. This is likely due to the Linux vulkan graphics driver being better for AMD hardware. With Nvidia hardware, the vulkan graphics driver is the same on both, so I would not expect DXVK on Linux to outperform DXVK on Windows in this game. I should add that while I do not have any current numbers, I know that A Hat In Time is a game where DXVK makes it outperform the native Windows Direct3D 9 implementation on Nvidia hardware regardless of whether the OS is Windows or Linux, but the margin is not quite as big as it is with AMD graphics hardware. Lastly, I have been told that the benchmarks themselves have the issue that they do not represent real game play. Doing repeatable benchmarks of real game play is hard, especially in World of Warcraft, given that you are playing with other people and not bots, so I am not sure if it is possible for flightless mango to address this concern. Regardless, his benchmark data suggests that future Linux systems will perform comparably to Windows, which is good news.
Blizzard hasn't released any statements about their anticheat feature. It appears they have confirmed their ability to identify when players are using Wine, with built-in detection for that software. It's possible they can spot attempts to exploit Wine for cheating. World of Warcraft has been largely stable in Wine on Linux over the past decade. Occasionally, newer updates caused compatibility issues, prompting developers to enhance Wine support, though these incidents have become less frequent. I think this discussion likely started around ten years ago, and Blizzard hasn't attempted to restrict Linux usage or Overwatch play on Linux despite some misleading reports.
In these tests, none focus on the key aspect of image quality comparison. If rendering isn't consistent—skipping elements or producing lower quality results—the performance gains might come from faster execution. Running at medium or low settings also helps. I don’t understand what you’re observing, but according to the graphs you shared, Windows generally shows better frame consistency and higher FPS stability compared to Linux.
Times and frame rates change together. Higher FPS means shorter frames. In the tests where Linux succeeded, it performed well overall, matching Windows in its winning benchmarks. For steadier performance, I’m uncertain since Linux consistently showed better results across all moments, except for one exception in Doom Eternal.
WoW benchmarking presents challenges because it varies based on location and activity, affecting different game aspects. A quick pass over Dazar'alor tends to strain the GPU more, while most sessions are CPU-intensive or limited by internal systems and frequent RNG checks in MMORPGs. Testing is necessary, even if minor drops like 1% or 0.1% FPS are desirable. DXVK should also be evaluated against WoW's DX12 mode on Windows to compare low-level APIs, especially since WoW has made significant progress with DX12 during BFAs and now experiments with ray-traced shadows.