gpt-oss 120B — B200 vs MI325X
Head-to-head AI inference benchmark comparison of B200 (NVIDIA Blackwell) and MI325X (AMD CDNA 3) on gpt-oss 120B. 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 72 tok/s/user as the target on gpt-oss 120B, B200 produces 25938 tok/s/chip ($0.02 per million tokens) and MI325X produces 1999 ($0.15). B200 is 725% cheaper per token; B200 delivers 1198% more tok/s/chip.
At 86 tok/s/user interactivity on gpt-oss 120B, B200 delivers 22522 tok/s/chip at $0.02 per million tokens; MI325X delivers 1208 tok/s/chip at $0.25. B200 is 1087% cheaper per token; B200 delivers 1765% more tok/s/chip at this point.
B200 posts 19611 tok/s/chip for $0.02 per million tokens at 100 tok/s/user on gpt-oss 120B; MI325X posts 835 tok/s/chip for $0.37. B200 is 1404% cheaper per token; B200 delivers 2250% more tok/s/chip. (Numbers reflect the default 1k/1k · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | B200:25938.2MI325X:1998.6 | B200:22522.4MI325X:1207.8 | B200:19610.5MI325X:834.5 |
| Cost ($/M tok) | B200:$0.019MI325X:$0.154 | B200:$0.021MI325X:$0.253 | B200:$0.024MI325X:$0.367 |
| tok/s/MW | B200:15168535MI325X:1182629 | B200:13170991MI325X:714674 | B200:11468138MI325X:493807 |
| Concurrency | B200:~256MI325X:~60 | B200:~256MI325X:~31 | B200:~233MI325X:~17 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.