GLM 5/5.1 — GB300 NVL72 vs MI325X
Head-to-head AI inference benchmark comparison of GB300 NVL72 (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.
GB300 NVL72: 6679 tok/s/chip, $0.10 per million tokens at 69 tok/s/user on GLM 5/5.1. MI325X is unmeasured here.
At 121 tok/s/user on GLM 5/5.1, GB300 NVL72 delivers 811 tok/s/chip at $0.79 per million tokens; MI325X hasn't been benchmarked at this target.
GB300 NVL72 hits 151 tok/s/chip for $4.24 per million tokens at 173 tok/s/user on GLM 5/5.1. No MI325X data at this operating point. (Numbers reflect the default 1k/1k · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | GB300 NVL72:6679.2MI325X:— | GB300 NVL72:810.6MI325X:— | GB300 NVL72:151.3MI325X:— |
| Cost ($/M tok) | GB300 NVL72:$0.096MI325X:— | GB300 NVL72:$0.792MI325X:— | GB300 NVL72:$4.242MI325X:— |
| tok/s/MW | GB300 NVL72:3150557MI325X:— | GB300 NVL72:382341MI325X:— | GB300 NVL72:71346MI325X:— |
| Concurrency | GB300 NVL72:~1221MI325X:— | GB300 NVL72:~166MI325X:— | GB300 NVL72:~23MI325X:— |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.