[md_rumpf]

Are there really use cases for faster chips? I can run all models I want on an H100 pod. No models exist that I can't run with at least 64 H100s. NVIDIA should just stop.