“30x throughput gain on the Llama 3.1 405B test versus the H200 NVL8 system. With @nvidia ‘s GB200 NVL72 system — connecting 72 NVIDIA Blackwell GPUs to act as a single, massive GPU ⚙️ Time to First Token and Time per Output Token Production settings care about time to first https://x.com/rohanpaul_ai/status/1907864056054296823
