Mastodawn

⚠️ I had posted this not knowing that the Nvidia drivers for #Linux are still not as good as the ones for Windows OS: namely they don't offload #VRAM excess to #RAM..
Which means that one needs lots of VRAM, especially for #AI.
Since the whole #Nvidia AI business is on Linux servers/compute nodes, I suppose that they want companies to buy GPUs at a premium with precious VRAM. It's not great for the environment for sure, nevermind linux ecosystem.. #nvidiafication

https://forums.developer.nvidia.com/t/non-existent-shared-vram-on-nvidia-linux-drivers/260304/152

Non-existent shared VRAM on NVIDIA Linux drivers

new 595 beta mentions Improved support for falling back to system memory when available video memory is low, to help prevent Wayland desktop freezes. will test on Spider-Man remastered once the kernel modules release on github.

NVIDIA Developer Forums

Nicolas MØUART 6d ago

The other day I was looking at their repo, I mentioned it.
I saw that model, and was "wot!? NVFP4?? What's that?!" -> It's a model optimized for blackwell Nvidia GPUs, yes dear, how weird, it's just weird. #distro #redhat #linux #nvidiafication

https://huggingface.co/RedHatAI/gemma-4-26B-A4B-it-NVFP4

RedHatAI/gemma-4-26B-A4B-it-NVFP4 · Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.