Mesa NVK Benchmark! RTX 2060 SUPER | Mistral Nemo @ llama.cpp

Published on ● Video Link: https://www.youtube.com/watch?v=bnW_QoEo4_M



Duration: 0:00
40 views
0


In case you are wondering. Linux 6.12 with Mesa 24.3 gets like 8 tokens a second with Mistral Nemo on the current master of 2025 March. Vulkan backend. It's NVIDIA for ML, but FOSS and no CUDA.

Well I got rid of that Pascal based Tesla P4 for a RTX 2060 SUPER, which is a hand me down. But as the SUPER is Turing, it supports NVK and the GSP blob for automatic power control.
In exact, to get rid of the proprietary drivers and see if Mesa/Nouveau is finally viable for NVIDIA gaming and ML!