Gregory's ServerI think it's neat. We're getting closer to an almost fully FOSS stack on the #NVIDIA side, while getting some nice additions to orchestration.
We already have #OCI runtimes and images specifically catered to #CUDA workloads, cutting away fat from orchestration and slimming down images.
You can now technically stick a GPU from every vendor into a machine and even share workloads between them. Compute with NVIDIA, doing OpenGL with AMD and transcoding with Intel.
Consider the possibility of setting up something like VirtGL to have all 3 vendor GPUs working together to render DOOM Dark Ages at 8K.
Even using either for various things like complex VJing setups, or signage computers with lots of fancy processing, demos, but also setting up 16 Xfire AI bots to fight for my amusement...
Or y'know, mint contracts or something.