Graphics Driver: Mesa RADV (latest version available in Bazzite)
Kernel: Latest Bazzite kernel
Proton: Latest Proton Experimental
Game
Halo Campaign Evolved
Description
I noticed a significant performance difference between Windows and Linux while playing Halo Campaign Evolved.
The game relies heavily on hardware ray tracing for its lighting, so disabling RT is not a viable workaround because it fundamentally changes the intended visuals.
Using the same hardware and comparable graphics settings, I consistently observe approximately:
Windows: ~105 FPS
Linux (Bazzite + RADV + VKD3D-Proton): ~82–85 FPS
This represents roughly a 20% performance loss on Linux.
At first I thought this might be specific to my system, but I later found an independent benchmark showing essentially the same behavior on the same scene (the beginning of Silent Cartographer), closely matching my own results:
Because both my own testing and the independent benchmark show a very similar performance gap, I thought it would be worth reporting.
I don't know whether the performance difference originates in VKD3D-Proton, RADV, shader compilation, or another part of the Linux graphics stack, but I hope this report can help identify where the remaining overhead comes from.
I also believe Halo Campaign Evolved is a particularly valuable title for evaluating Linux graphics performance because its ray tracing implementation is fundamental to the game's rendering rather than being an optional visual feature. Since the lighting pipeline is built around hardware ray tracing, performance differences cannot simply be avoided by disabling RT, making this an excellent real-world workload for identifying remaining overhead in the DirectX 12 → Vulkan translation layer or the Vulkan driver itself. If there are still optimization opportunities in VKD3D-Proton, RADV, or elsewhere in the Linux graphics stack, this title seems like an ideal candidate to help expose and measure them.
If there is any additional information, logs, or testing that would help identify the source of the performance difference, I'd be happy to provide them.
System information
Game
Halo Campaign Evolved
Description
I noticed a significant performance difference between Windows and Linux while playing Halo Campaign Evolved.
The game relies heavily on hardware ray tracing for its lighting, so disabling RT is not a viable workaround because it fundamentally changes the intended visuals.
Using the same hardware and comparable graphics settings, I consistently observe approximately:
This represents roughly a 20% performance loss on Linux.
At first I thought this might be specific to my system, but I later found an independent benchmark showing essentially the same behavior on the same scene (the beginning of Silent Cartographer), closely matching my own results:
https://youtu.be/sTG8f5TsfgA
Because both my own testing and the independent benchmark show a very similar performance gap, I thought it would be worth reporting.
I don't know whether the performance difference originates in VKD3D-Proton, RADV, shader compilation, or another part of the Linux graphics stack, but I hope this report can help identify where the remaining overhead comes from.
I also believe Halo Campaign Evolved is a particularly valuable title for evaluating Linux graphics performance because its ray tracing implementation is fundamental to the game's rendering rather than being an optional visual feature. Since the lighting pipeline is built around hardware ray tracing, performance differences cannot simply be avoided by disabling RT, making this an excellent real-world workload for identifying remaining overhead in the DirectX 12 → Vulkan translation layer or the Vulkan driver itself. If there are still optimization opportunities in VKD3D-Proton, RADV, or elsewhere in the Linux graphics stack, this title seems like an ideal candidate to help expose and measure them.
If there is any additional information, logs, or testing that would help identify the source of the performance difference, I'd be happy to provide them.