Kimi K3 2.8T · Chip comparison

Kimi K3 2.8T — GB200 NVL72 vs H200

Head-to-head AI inference benchmark comparison of GB200 NVL72 (NVIDIA Blackwell) and H200 (NVIDIA Hopper) 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

GB200 NVL72 posts 9890 tok/s/chip for $0.05 per million tokens at 2 tok/s/user on Kimi K3 2.8T; H200 posts 180 tok/s/chip for $1.88. GB200 NVL72 is 3507% cheaper per token; GB200 NVL72 delivers 5399% more tok/s/chip.

Throughput at 3 tok/s/user on Kimi K3 2.8T: GB200 NVL72 hits 9890 tok/s/chip, H200 hits 118. Per-million costs land at $0.05 and $2.86 respectively. GB200 NVL72 is 5379% cheaper per token; GB200 NVL72 delivers 8253% more tok/s/chip.

GB200 NVL72 / H200 on Kimi K3 2.8T at 4 tok/s/user: 9890 / 91 tok/s/chip, $0.05 / $3.74 per million tokens. GB200 NVL72 is 7059% cheaper per token; GB200 NVL72 delivers 10814% 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)
GB200 NVL72:9889.9H200:179.9
GB200 NVL72:9889.9H200:118.4
GB200 NVL72:9889.9H200:90.6
Cost ($/M tok)
GB200 NVL72:$0.052H200:$1.884
GB200 NVL72:$0.052H200:$2.862
GB200 NVL72:$0.052H200:$3.740
tok/s/MW
GB200 NVL72:5288707H200:131281
GB200 NVL72:5288707H200:86423
GB200 NVL72:5288707H200:66144
Concurrency
GB200 NVL72:~96H200:~14
GB200 NVL72:~96H200:~7
GB200 NVL72:~96H200:~5

Inference Performance

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