Trying Out Ballistic Overkill On Linux With Vulkan

With the rendering appearing the same when switching between the OpenGL and Vulkan renderers, I was curious about the performance impact. Sadly though there's no built-in benchmarking support for this Unity game and not even an in-game FPS counter (and no viable solution for independent FPS monitoring overlay with Vulkan on Linux). Just from running the game, it wasn't possible to tell any real performance difference on this i7-7700K + GTX 1050 Ti setup since the game isn't very graphically demanding to begin with.

When comparing the impact of OpenGL vs. Vulkan when launching the game, the CPU usage appeared overall similar between the different renderers, but there were less CPU spikes when using Vulkan.

So next I enabled the Phoronix Test Suite to carry out GPU/CPU utilization monitoring while firing up a Ballistic Overkill server (Sunnsquare Mall, King of the Hill) and using that empty server to basically do a fly-by/run-around for five minutes to see how the renderer options compared. Both times the same configuration and roughly same path/in-game action was done and obviously all of the same in-game settings....

During this rough comparison of OpenGL vs. Vulkan for Ballistic Overkill on Linux, we see the CPU utilization did end up being noticeably different... On this Core i7 7700K system, the average CPU usage in-game was 19.3% with a peak of 35%. When using OpenGL, the average CPU usage was 30.7% with a peak of 43%.

With both OpenGL and Vulkan, the GeForce GTX 1050 Ti was being fully utilized. But when using OpenGL, there were more dips/stalls in the GPU utilization, likely when waiting on the CPU to feed it, while the Vulkan renderer was able to keep the GPU consistently busy.

It's good to see the CPU usage lower with the Vulkan renderer and less fluctuation in the NVIDIA GPU utilization. It's unfortunate that the RADV Vulkan driver wasn't working with Ballistic Overkill, but more than likely that will probably see shortly any necessary fixes in Mesa Git. When it does, I'll try running some more rough tests (RadeonSI vs. RADV) but it's a pity that this game doesn't have better benchmarking capabilities.