GLM 5/5.1 — B200 vs MI325X
Head-to-head AI inference benchmark comparison of B200 (NVIDIA Blackwell) and MI325X (AMD CDNA 3) on GLM 5/5.1. 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.
Setting 21 tok/s/user as the target on GLM 5/5.1, B200 produces 1987 tok/s/chip ($0.24 per million tokens) and MI325X produces 173 ($1.79). B200 is 640% cheaper per token; B200 delivers 1050% more tok/s/chip.
At 26 tok/s/user interactivity on GLM 5/5.1, B200 delivers 1054 tok/s/chip at $0.46 per million tokens; MI325X delivers 101 tok/s/chip at $3.04. B200 is 558% cheaper per token; B200 delivers 940% more tok/s/chip at this point.
B200 posts 666 tok/s/chip for $0.73 per million tokens at 30 tok/s/user on GLM 5/5.1; MI325X posts 62 tok/s/chip for $4.91. B200 is 573% cheaper per token; B200 delivers 982% more tok/s/chip. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | B200:1987.1MI325X:172.9 | B200:1053.9MI325X:101.3 | B200:665.7MI325X:61.6 |
| Cost ($/M tok) | B200:$0.242MI325X:$1.789 | B200:$0.462MI325X:$3.037 | B200:$0.729MI325X:$4.907 |
| tok/s/MW | B200:1162021MI325X:102283 | B200:616298MI325X:59939 | B200:389313MI325X:36421 |
| Concurrency | B200:~1254MI325X:~34 | B200:~757MI325X:~16 | B200:~413MI325X:~9 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.