GLM 5/5.1 — B200 vs MI355X
Head-to-head AI inference benchmark comparison of B200 (NVIDIA Blackwell) and MI355X (AMD CDNA 4) 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.
Throughput at 32 tok/s/user on GLM 5/5.1: B200 hits 1756 tok/s/chip, MI355X hits 1369. Per-million costs land at $0.27 and $0.31 respectively. B200 is 13% cheaper per token; B200 delivers 28% more tok/s/chip.
B200 / MI355X on GLM 5/5.1 at 51 tok/s/user: 1287 / 957 tok/s/chip, $0.37 / $0.43 per million tokens. B200 is 17% cheaper per token; B200 delivers 35% more tok/s/chip.
Toward the upper edge of the 12–90 tok/s/user interactivity band, at 71 tok/s/user on GLM 5/5.1: B200 runs 1004 tok/s/chip at $0.48/M tokens, MI355X runs 709 at $0.58/M. B200 is 21% cheaper per token; B200 delivers 41% more tok/s/chip. (Numbers reflect the default 8k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | B200:1756.3MI355X:1368.5 | B200:1287.2MI355X:956.5 | B200:1003.6MI355X:709.4 |
| Cost ($/M tok) | B200:$0.274MI355X:$0.309 | B200:$0.370MI355X:$0.433 | B200:$0.480MI355X:$0.580 |
| tok/s/MW | B200:1027090MI355X:654793 | B200:752761MI355X:457666 | B200:586911MI355X:339417 |
| Concurrency | B200:~240MI355X:~19 | B200:~23MI355X:~9 | B200:~13MI355X:~4 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.