DeepSeek R1 — B300 vs MI355X
Head-to-head AI inference benchmark comparison of B300 (NVIDIA Blackwell) and MI355X (AMD CDNA 4) 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 66 tok/s/user interactivity on DeepSeek R1, B300 delivers 1436 tok/s/chip at $0.45 per million tokens; MI355X delivers 807 tok/s/chip at $0.50. B300 is 11% cheaper per token; B300 delivers 78% more tok/s/chip at this point.
B300 posts 408 tok/s/chip for $1.54 per million tokens at 116 tok/s/user on DeepSeek R1; MI355X posts 295 tok/s/chip for $1.43. MI355X is 8% cheaper per token; B300 delivers 38% more tok/s/chip.
Throughput at 166 tok/s/user on DeepSeek R1: B300 hits 208 tok/s/chip, MI355X hits 130. Per-million costs land at $2.96 and $3.64 respectively. B300 is 23% cheaper per token; B300 delivers 60% more tok/s/chip. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | B300:1436.5MI355X:807.1 | B300:408.4MI355X:295.1 | B300:207.9MI355X:130.1 |
| Cost ($/M tok) | B300:$0.451MI355X:$0.503 | B300:$1.541MI355X:$1.426 | B300:$2.964MI355X:$3.637 |
| tok/s/MW | B300:756042MI355X:386158 | B300:214963MI355X:141218 | B300:109432MI355X:62242 |
| Concurrency | B300:~567MI355X:~115 | B300:~159MI355X:~10 | B300:~5MI355X:~4 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.