GLM 5.3 744B — GB300 NVL72 vs MI325X
Head-to-head AI inference benchmark comparison of GB300 NVL72 (NVIDIA Blackwell) and MI325X (AMD CDNA 3) on GLM 5.3 744B. 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.
AgentX replays real coding-agent sessions rather than fixed-length prompts, so context grows turn over turn and most of each request is served from cache instead of being recomputed. That turns the comparison into a systems question: KV transfer between nodes, prefix-aware routing, and cache capacity all move the curve alongside raw chip throughput. Fixed-sequence workloads stay the clean baseline for kernel and silicon performance, so the two scenarios answer different questions about the same hardware. Learn more about AgentX →
GB300 NVL72 hits 14788 tok/s/chip for $0.04 per million tokens at 139 tok/s/user on GLM 5.3 744B. No MI325X data at this operating point.
GB300 NVL72: 9451 tok/s/chip, $0.07 per million tokens at 192 tok/s/user on GLM 5.3 744B. MI325X is unmeasured here.
At 245 tok/s/user on GLM 5.3 744B, GB300 NVL72 delivers 6125 tok/s/chip at $0.10 per million tokens; MI325X hasn't been benchmarked at this target. (Numbers reflect the default agentic-traces · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | GB300 NVL72:14788.2MI325X:— | GB300 NVL72:9451.0MI325X:— | GB300 NVL72:6124.6MI325X:— |
| Cost ($/M tok) | GB300 NVL72:$0.043MI325X:— | GB300 NVL72:$0.068MI325X:— | GB300 NVL72:$0.105MI325X:— |
| tok/s/MW | GB300 NVL72:6975580MI325X:— | GB300 NVL72:4458017MI325X:— | GB300 NVL72:2888976MI325X:— |
| Concurrency | GB300 NVL72:~53MI325X:— | GB300 NVL72:~20MI325X:— | GB300 NVL72:~17MI325X:— |