I am trying to document how much of an impact VFIO has on gaming performance. More benchmarks will come in the future, but for now, let’s look at Forza Horizon 5.
I picked Forza Horizon 5 to do these tests as it is somewhat demanding and has a benchmark mode which displays somewhat interesting results. The graphics are all maxed out and both vsync and g-sync are off, in game and in any applicable place.
The VFIO host’s hardware:
Intel Core i7 9700k (reduced to 6 cores for this benchmark) 64 GB DDR4 RAM 3000 MHz Intel 660p 1 TB NVME SSD GTX 1080 Ti 11 GB
The VFIO host’s disk configuration has a LUKS container with LVM volumes used for VMs.
6 cores and 32 GB of RAM are given to the gaming VM and nothing else is running on the host for these benchmarks.
Further configuration details can be seen (and similar setup can be deployed by setting up a FAI server from given playbooks) here:
Now onto the benchmark results, first results on the physical hardware with 6 cores and 64 GB of RAM:
Results on the virtual machine with 6 cores and 32 GB of RAM:
And finally, a comparison of frametimes in capframex:
As can be seen, the performance hit is not too bad, the biggest issue being increased stuttering, but it barely has any effect on the gameplay and happens on the physical machine too.
For the sake of completeness, here is the benchmark run on physical machine with all CPU cores enabled:
Any suggestions and criticisms are welcome.
More benchmarks to come.