DeepSeek V4 Pro 1.6T — B200 vs MI355X
Head-to-head AI inference benchmark comparison of B200 (NVIDIA Blackwell) and MI355X (AMD CDNA 4) on DeepSeek V4 Pro 1.6T. 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.
Throughput at 17 tok/s/user on DeepSeek V4 Pro 1.6T: B200 hits 2522 tok/s/GPU, MI355X hits 1609. Per-million costs land at $0.21 and $0.25 respectively. B200 is 20% cheaper per token; B200 delivers 57% more tok/s/GPU.
B200 / MI355X on DeepSeek V4 Pro 1.6T at 30 tok/s/user: 1440 / 321 tok/s/GPU, $0.38 / $1.32 per million tokens. B200 is 248% cheaper per token; B200 delivers 348% more tok/s/GPU.
Toward the upper edge of the 5–54 tok/s/user interactivity band, at 42 tok/s/user on DeepSeek V4 Pro 1.6T: B200 runs 1348 tok/s/GPU at $0.40/M tokens, MI355X runs 149 at $2.78/M. B200 is 587% cheaper per token; B200 delivers 806% more tok/s/GPU. (Numbers reflect the default 1k/1k · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | B200:2521.9MI355X:1608.8 | B200:1440.3MI355X:321.3 | B200:1347.8MI355X:148.7 |
| Cost ($/M tok) | B200:$0.212MI355X:$0.254 | B200:$0.378MI355X:$1.316 | B200:$0.404MI355X:$2.776 |
| tok/s/MW | B200:1162144MI355X:607113 | B200:663717MI355X:121244 | B200:621092MI355X:56112 |
| Concurrency | B200:~608MI355X:~402 | B200:~206MI355X:~47 | B200:~149MI355X:~15 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.