Qwen 3.5 397B-A17B · GPU comparison

Qwen 3.5 397B-A17B — GB200 NVL72 vs MI300X

Head-to-head AI inference benchmark comparison of GB200 NVL72 (NVIDIA Blackwell) and MI300X (AMD CDNA 3) on Qwen 3.5 397B-A17B. 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.

Near the low end of the 34–71 tok/s/user interactivity band, at 43 tok/s/user on Qwen 3.5 397B-A17B: GB200 NVL72 runs 4910 tok/s/GPU at $0.12/M tokens, MI300X runs 346 at $0.89/M. GB200 NVL72 is 629% cheaper per token; GB200 NVL72 delivers 1320% more tok/s/GPU.

Setting 53 tok/s/user as the target on Qwen 3.5 397B-A17B, GB200 NVL72 produces 1977 tok/s/GPU ($0.32 per million tokens) and MI300X produces 190 ($1.63). GB200 NVL72 is 412% cheaper per token; GB200 NVL72 delivers 943% more tok/s/GPU.

At 62 tok/s/user interactivity on Qwen 3.5 397B-A17B, GB200 NVL72 delivers 1328 tok/s/GPU at $0.47 per million tokens; MI300X delivers 123 tok/s/GPU at $2.52. GB200 NVL72 is 432% cheaper per token; GB200 NVL72 delivers 982% more tok/s/GPU at this point. (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.)

View performance-per-dollar view →

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/gpu)
GB200 NVL72:4910.3MI300X:345.8
GB200 NVL72:1976.7MI300X:189.5
GB200 NVL72:1327.7MI300X:122.7
Cost ($/M tok)
GB200 NVL72:$0.123MI300X:$0.893
GB200 NVL72:$0.319MI300X:$1.633
GB200 NVL72:$0.473MI300X:$2.517
tok/s/MW
GB200 NVL72:2338245MI300X:193162
GB200 NVL72:941273MI300X:105886
GB200 NVL72:632228MI300X:68573
Concurrency
GB200 NVL72:~1438MI300X:~34
GB200 NVL72:~416MI300X:~15
GB200 NVL72:~146MI300X:~8

Inference Performance

Inference performance metrics across different models, hardware configurations, and serving parameters.

Vendor:
Aggregation:
Spec Decoding: