Kimi K3 2.8T · Chip comparison

Kimi K3 2.8T — GB200 NVL72 vs GB300 NVL72

Head-to-head AI inference benchmark comparison of GB200 NVL72 (NVIDIA Blackwell) and GB300 NVL72 (NVIDIA Blackwell) 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 →

At 80 tok/s/user interactivity on Kimi K3 2.8T, GB200 NVL72 delivers 7,321 tok/s/chip at $0.071 per million tokens; GB300 NVL72 delivers 16,781 tok/s/chip at $0.038. GB300 NVL72 is 46% cheaper per token; GB300 NVL72 delivers 129% more tok/s/chip at this point.

GB200 NVL72 posts 3,128 tok/s/chip for $0.17 per million tokens at 127 tok/s/user on Kimi K3 2.8T; GB300 NVL72 posts 10,765 tok/s/chip for $0.060. GB300 NVL72 is 64% cheaper per token; GB300 NVL72 delivers 3.4× the tok/s/chip.

Throughput at 173 tok/s/user on Kimi K3 2.8T: GB200 NVL72 hits 1,880 tok/s/chip, GB300 NVL72 hits 9,392. Costs per million tokens land at $0.27 and $0.068 respectively. GB300 NVL72 is 75% cheaper per token; GB300 NVL72 delivers 5.0× the 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 →

GB300 NVL72 vs GB200 NVL72: full specs & benchmarks across models →

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)
GB200 NVL727,321GB300 NVL7216,781
GB200 NVL723,128GB300 NVL7210,765
GB200 NVL721,880GB300 NVL729,392
Cost ($/M tok)
GB200 NVL72$0.071GB300 NVL72$0.038
GB200 NVL72$0.17GB300 NVL72$0.060
GB200 NVL72$0.27GB300 NVL72$0.068
tok/s/MW
GB200 NVL723,914,710GB300 NVL727,915,762
GB200 NVL721,672,889GB300 NVL725,077,683
GB200 NVL721,005,268GB300 NVL724,430,117
Concurrency
GB200 NVL72~47GB300 NVL72~115
GB200 NVL72~10GB300 NVL72~40
GB200 NVL72~5GB300 NVL72~31