I usually use stk on a computer with dual gpus. (Intel 4th Gen Integrated Graphics and Radeon HD 8870M)
When I run stk (with all graphics options at max) on my default (less powerful) gpu, the bottleneck appears to be the gpu. FPS are about 20, stk uses ~9% of one cpu core.
When I run stk (with all graphics options at max) on my additional (much more powerful) gpu, the bottleneck appears to be the cpu instead. FPS are only about 30, but stk uses 100% of one cpu core.
While the Vulkan driver is much faster (60 FPS at 30% of one cpu core), I don't think it is a fair comparison because many graphics features are automatically disabled due the the limitations and age of the gpu.
Is the bottleneck on the powerful gpu the limit at which one cpu core can transfer data to/from the gpu?
Is this related to Why SuperTuxKart struggles with 10k scene nodes in a scene, where they discuss the cpu/ram bandwidth saturation?
I usually run stk in 2 or 4 multiplayer and turn down the graphics. Otherwise is lags to 10-20 FPS on gl driver and powerful gpu with visible stuttering of the screen. I think the powerful gpu should be able to handle it somewhat faster if it were not for the cpu bottleneck.