DeepSeek R1 — H100 vs MI325X
Head-to-head AI inference benchmark comparison of H100 (NVIDIA Hopper) and MI325X (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.
H100 posts 274 tok/s/GPU for $1.33 per million tokens at 39 tok/s/user on DeepSeek R1; MI325X posts 341 tok/s/GPU for $1.03. MI325X is 29% cheaper per token; MI325X delivers 24% more tok/s/GPU.
Throughput at 50 tok/s/user on DeepSeek R1: H100 hits 168 tok/s/GPU, MI325X hits 225. Per-million costs land at $2.20 and $1.57 respectively. MI325X is 40% cheaper per token; MI325X delivers 34% more tok/s/GPU.
H100 / MI325X on DeepSeek R1 at 60 tok/s/user: 131 / 132 tok/s/GPU, $2.74 / $2.67 per million tokens. MI325X is 3% cheaper per token; MI325X delivers 1% more tok/s/GPU. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | H100:274.3MI325X:341.4 | H100:168.0MI325X:225.5 | H100:130.6MI325X:132.2 |
| Cost ($/M tok) | H100:$1.328MI325X:$1.027 | H100:$2.197MI325X:$1.566 | H100:$2.739MI325X:$2.672 |
| tok/s/MW | H100:158541MI325X:156612 | H100:97100MI325X:103433 | H100:75490MI325X:60648 |
| Concurrency | H100:~624MI325X:~39 | H100:~203MI325X:~19 | H100:~128MI325X:~9 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.