NVIDIA

DeepSeek V4 Flash on 8×H100 SXM with vLLM FP8

Llama 3.3 70B on 8× A100 SXM4 80GB with vLLM

Llama 4 Scout on 8×H100 SXM with vLLM (public benchmark)

Qwen2.5-Coder 32B on 4× L40S with vLLM (FP8)

DeepSeek V4 Flash with disaggregated prefill (H100) + decode (H200) via Mooncake

GLM-5.1 on 8× H200 SXM with vLLM BF16

Gemma 4 26B on 4× H100 SXM with FP8