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.
Throughput at 23 tok/s/user on GLM 5/5.1: GB300 NVL72 hits 4083 tok/s/GPU, MI325X hits 161. Per-million costs land at $0.18 and $2.22 respectively. GB300 NVL72 is 1129% cheaper per token; GB300 NVL72 delivers 2434% more tok/s/GPU.
GB300 NVL72 / MI325X on GLM 5/5.1 at 27 tok/s/user: 3994 / 85 tok/s/GPU, $0.18 / $4.34 per million tokens. GB300 NVL72 is 2253% cheaper per token; GB300 NVL72 delivers 4620% more tok/s/GPU.
Toward the upper edge of the 19–34 tok/s/user interactivity band, at 30 tok/s/user on GLM 5/5.1: GB300 NVL72 runs 3869 tok/s/GPU at $0.19/M tokens, MI325X runs 62 at $5.71/M. GB300 NVL72 is 2899% cheaper per token; GB300 NVL72 delivers 6186% more tok/s/GPU. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | GB300 NVL72:4082.8MI325X:161.1 | GB300 NVL72:3993.9MI325X:84.6 | GB300 NVL72:3869.0MI325X:61.6 |
| Cost ($/M tok) | GB300 NVL72:$0.180MI325X:$2.218 | GB300 NVL72:$0.184MI325X:$4.337 | GB300 NVL72:$0.190MI325X:$5.710 |
| tok/s/MW | GB300 NVL72:1944179MI325X:73913 | GB300 NVL72:1901851MI325X:38813 | GB300 NVL72:1842383MI325X:28234 |
| Concurrency | GB300 NVL72:~7423MI325X:~30 | GB300 NVL72:~6952MI325X:~15 | GB300 NVL72:~6351MI325X:~9 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.