Since I can’t edit this thread I will do some clarifications on this reply. What NVIDIA Linux drivers are missing is GTT (graphics translation table)/GART (graphics address remapping table) which something like the Shared GPU memory that exists on windows, but this is how is called technically on the Linux kernel.
GTT has been on the Linux kernel for 12 years or more already and it has been implemented on the AMD drivers (amdgpu) and intel (i915) since a long time ago and it’s baffling how nvidia has been unable to implement GTT on their drivers. This is not a brand new feature in constant development like Wayland.
Edit: The GTT feature from the Linux Kernel is not the same as nvidia-smi -gtt (GPU Target Temperature)
References:
- AMDGPU Glossary — The Linux Kernel documentation
- drm/i915 Intel GFX Driver — The Linux Kernel documentation
- GART
- Graphics address remapping table - Wikipedia
- GitHub - T-X/linux-amdgpu-radeon-vram-swapping-test: Linux amdgpu Radeon VRAM Swapping Test
Note for experienced people: If you have a AMD GPU, you can execute radeontop -d - -l 1 to get something like this:
Dumping to -, line limit 1.
1720222198.786508: bus 01, gpu 0.00%, ee 0.00%, vgt 0.00%, ta 0.00%, sx 0.00%, sh 0.00%, spi 0.00%, sc 0.00%, pa 0.00%, db 0.00%, cb 0.00%, vram 14.08% 574.87mb, gtt 7.33% 289.45mb, mclk 20.00% 0.300ghz, sclk 18.21% 0.214ghz
In the end, you can see this: vram 14.08% 574.87mb, gtt 7.33% 289.45mb. vram is obviusly the used VRAM of the GPU and gtt is the System RAM being used by the GPU.
If you have an Integrated Intel GPU, you open htop (to monitor your RAM usage) and execute Release v0.5.0 - Tune behavior with large PCIe BARs · GpuZelenograd/memtest_vulkan · GitHub under the Intel GPU. You will see how your RAM usage increases in less than 1 second. This would also apply for AMD users but I would recommend to use the script from this repository instead: GitHub - T-X/linux-amdgpu-radeon-vram-swapping-test: Linux amdgpu Radeon VRAM Swapping Test