Why is the emulated GPU so weak?

[GalacticGamer]

The emulated GPU gets 509 test time 30m in basemark web 3.0 and the host gets 3248 test time 1m. Please improve it.

[kasper]

The video on my guest (virtual) machines is almost on par with the host (physical) machine.   You may need to check your configuration to fix your issue.   It looks like your benchmark program is 'basemark web 3.0' which I am not familiar with.   Try using other benchmark programs and see if they compare to the one your just used.

[kasper]

Interesting.  Basemark (now gpuscore) appears to be CPU dependent.   So the results of a GPU benchmark may vary greatly depending on the CPU.  For me, if I take a guest and give it 2 virtual CPU's, run the bench, and then change the guest to 4 CPU's, the bench results are greater.   What is interesting is running the benchmark on the physical host vs running the bench on a guest that has almost as many virtual CPU's.   The results on the guest are roughly %50 less which one would not expect.  Trying other benchmark apps in host vs guest shows weak guest results but not as great as the web Basemark bench.   Maybe someone could explain why this is so and whether I am missing some configuration option that would result in better numbers.

[richard612]

Strange little test. I got 1226 in a VM and 1476 on the [Linux] host.

In the VM, with Chrome's "override software rendering list" (chrome://flags/#ignore-gpu-blocklist) set to disabled the score drops to 430.  This flag might be what OP needs.

[RaSystemlord]

Out of common interest, I ran out a couple of tests.

Host: (Asus ROG Strix Scar, Ultra 9 275HX, RTX 5080 laptop, 24 cores, doesn't have hyperthreading, Win 11 Pro): 2047

VM: Kubuntu 24.04.02, a (almost) new clean system, Firefox, defaults everywhere
- with 8 processors, no VM Tools OR with Open VM Tools: 522
- with 16 processors: 502

[kasper]

Benchmarks are dependent on many factors.   The more general the benchmark (web browser based vs raw numbers drystone/whetstone/opengl) the more factors are involved in configuring the system and tweaking the results.  For VMWare Workstation, moving from 17.x to 25.x to 26.x, gave me an ever so small decline in efficiency but that may be me.   For me, the high road is leaning towards a Linux/Wavyland physical host.