GLM 5/5.1 · Chip comparison
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 16 tok/s/user as the target on GLM 5/5.1, B200 produces 3,682 tok/s/chip ($0.13 per million tokens) and MI325X produces 413 ($0.74). B200 is 82% cheaper per token; B200 delivers 8.9× the tok/s/chip.
At 20 tok/s/user interactivity on GLM 5/5.1, B200 delivers 3,682 tok/s/chip at $0.13 per million tokens; MI325X delivers 282 tok/s/chip at $1.08. B200 is 88% cheaper per token; B200 delivers 13× the tok/s/chip at this point.
B200 posts 2,822 tok/s/chip for $0.17 per million tokens at 25 tok/s/user on GLM 5/5.1; MI325X posts 180 tok/s/chip for $1.70. B200 is 90% cheaper per token; B200 delivers 16× the tok/s/chip. (Numbers reflect the default 8k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
Interpolated from real benchmark data. Edit target interactivity values below to compare at different operating points.
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | B2003,682MI325X413 | B2003,682MI325X282 | B2002,822MI325X180 |
| Cost ($/M tok) | B200$0.13MI325X$0.74 | B200$0.13MI325X$1.08 | B200$0.17MI325X$1.70 |
| tok/s/MW | B2002,153,308MI325X244,630 | B2002,153,308MI325X166,947 | B2001,650,028MI325X106,283 |
| Concurrency | B200~560MI325X~24 | B200~560MI325X~13 | B200~250MI325X~7 |