gpt-oss 120B · GPU comparison

gpt-oss 120B — GB200 NVL72 vs MI355X

Head-to-head AI inference benchmark comparison of GB200 NVL72 (NVIDIA Blackwell) and MI355X (AMD CDNA 4) 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.

Near the low end of the 71–317 tok/s/user interactivity band, at 132 tok/s/user on gpt-oss 120B: GB200 NVL72 runs 18334 tok/s/GPU at $0.03/M tokens, MI355X runs 7454 at $0.06/M. GB200 NVL72 is 65% cheaper per token; GB200 NVL72 delivers 146% more tok/s/GPU.

Setting 194 tok/s/user as the target on gpt-oss 120B, GB200 NVL72 produces 11222 tok/s/GPU ($0.06 per million tokens) and MI355X produces 3059 ($0.13). GB200 NVL72 is 141% cheaper per token; GB200 NVL72 delivers 267% more tok/s/GPU.

At 256 tok/s/user interactivity on gpt-oss 120B, GB200 NVL72 delivers 5771 tok/s/GPU at $0.11 per million tokens; MI355X delivers 1547 tok/s/GPU at $0.26. GB200 NVL72 is 141% cheaper per token; GB200 NVL72 delivers 273% more tok/s/GPU at this 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.)

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 NVL72:18334.5MI355X:7454.3
GB200 NVL72:11222.5MI355X:3058.7
GB200 NVL72:5771.4MI355X:1546.6
Cost ($/M tok)
GB200 NVL72:$0.034MI355X:$0.055
GB200 NVL72:$0.056MI355X:$0.135
GB200 NVL72:$0.107MI355X:$0.259
tok/s/MW
GB200 NVL72:8730692MI355X:2812942
GB200 NVL72:5344037MI355X:1154210
GB200 NVL72:2748305MI355X:583609
Concurrency
GB200 NVL72:~1178MI355X:~29
GB200 NVL72:~436MI355X:~8
GB200 NVL72:~60MI355X:~7

Inference Performance

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