DeepSeek R1 — H100 vs MI300X
Head-to-head AI inference benchmark comparison of H100 (NVIDIA Hopper) and MI300X (AMD CDNA 3) on DeepSeek R1. Latency, throughput, and cost across LLM workloads. Use the chart controls below to switch sequences, precisions, and metrics — same interactions as the main inference chart.
At 36 tok/s/user interactivity on DeepSeek R1, H100 delivers 279 tok/s/GPU at $1.31 per million tokens; MI300X delivers 268 tok/s/GPU at $1.16. MI300X is 12% cheaper per token; H100 delivers 4% more tok/s/GPU at this point.
H100 posts 205 tok/s/GPU for $1.77 per million tokens at 47 tok/s/user on DeepSeek R1; MI300X posts 186 tok/s/GPU for $1.66. MI300X is 6% cheaper per token; H100 delivers 10% more tok/s/GPU.
Throughput at 58 tok/s/user on DeepSeek R1: H100 hits 137 tok/s/GPU, MI300X hits 113. Per-million costs land at $2.59 and $2.74 respectively. H100 is 6% cheaper per token; H100 delivers 21% more tok/s/GPU. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | H100:279.4MI300X:267.6 | H100:205.0MI300X:186.4 | H100:137.5MI300X:113.3 |
| Cost ($/M tok) | H100:$1.305MI300X:$1.163 | H100:$1.766MI300X:$1.663 | H100:$2.589MI300X:$2.737 |
| tok/s/MW | H100:161495MI300X:149525 | H100:118518MI300X:104145 | H100:79475MI300X:63313 |
| Concurrency | H100:~666MI300X:~31 | H100:~293MI300X:~17 | H100:~140MI300X:~8 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.