h200 with wide-ep

(blog.vllm.ai)

50 points | by robertnishihara 11 hours ago

2 comments

kingstnap 15 minutes ago
Impressive performance work. It's interesting that you still see these 40+% perf gains like this.
Makes you think that you will continue to see the costs for a fixed level of "intelligence" dropping.
danielhanchen 13 minutes ago
Love vLLM!