Kimi K3 2.8T · Chip comparison

Kimi K3 2.8T — GB300 NVL72 vs MI355X

Head-to-head AI inference benchmark comparison of GB300 NVL72 (NVIDIA Blackwell) and MI355X (AMD CDNA 4) on Kimi K3 2.8T. 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 / MI355X on Kimi K3 2.8T at 39 tok/s/user: 9084 / 5708 tok/s/chip, $0.07 / $0.07 per million tokens. GB300 NVL72 is 3% cheaper per token; GB300 NVL72 delivers 59% more tok/s/chip.

Around the middle of the 13–120 tok/s/user interactivity band, at 66 tok/s/user on Kimi K3 2.8T: GB300 NVL72 runs 7247 tok/s/chip at $0.09/M tokens, MI355X runs 2831 at $0.15/M. GB300 NVL72 is 66% cheaper per token; GB300 NVL72 delivers 156% more tok/s/chip.

Setting 93 tok/s/user as the target on Kimi K3 2.8T, GB300 NVL72 produces 5702 tok/s/chip ($0.11 per million tokens) and MI355X produces 1858 ($0.22). GB300 NVL72 is 99% cheaper per token; GB300 NVL72 delivers 207% more tok/s/chip. (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.)

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/chip)
GB300 NVL72:9083.8MI355X:5707.6
GB300 NVL72:7246.8MI355X:2831.5
GB300 NVL72:5702.2MI355X:1857.9
Cost ($/M tok)
GB300 NVL72:$0.071MI355X:$0.073
GB300 NVL72:$0.089MI355X:$0.147
GB300 NVL72:$0.113MI355X:$0.224
tok/s/MW
GB300 NVL72:4284821MI355X:2730892
GB300 NVL72:3418310MI355X:1354763
GB300 NVL72:2689703MI355X:888930
Concurrency
GB300 NVL72:~60MI355X:~14
GB300 NVL72:~58MI355X:~5
GB300 NVL72:~43MI355X:~3

Inference Performance

Agentic inference metrics from the AgentX scenario and fixed-sequence inference metrics across models, hardware configurations, and serving parameters.