I'm about to upgrade from the 5800X3D to 9800X3D and thought I'd record benchmarks comparing the two systems.
However, I'm utterly confused by the results, benchmark settings and how fps is measured by the plugin.
Here's an example using the default settings. As you can see, my gpu is fully utilized, even at 1080p during the most cpu heavy part of the benchmark.
The frame rate is much higher than the physics rate, resulting in physics updates only happening every 2-3 frames, which means the benchmark's update cap of 20 is effectively the same as an update cap of 1.
This means that it's impossible to bench the cpu using the standardized benchmark settings. In order to shift the bottleneck to the cpu, I tried disabling any gpu intensive settings as well as cranking physics to the max.
This should give us a target physics rate that's higher than the achieved fps while also limiting physics updates per frame to 1, effectively limiting fps to what the cpu can handle.
As it turns out, the plugin doesn't cope with the fps dropping below the target physics rate.
What happens is that the benchmark slows down time progression and displays the fps as 288, which is incorrect.
Long story short, it appears to be impossible to measure cpu bound scenarios using this benchmark. The only metric that might give a glimpse at the actual cpu performance might be the default run's one percent lows, which
should be from frames which had physics updates.
