Head-to-head AI inference benchmark comparison of B300 (NVIDIA Blackwell) and H100 (NVIDIA Hopper) on DeepSeek R1. 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 20–122 tok/s/user interactivity band, at 45 tok/s/user on DeepSeek R1: B300 runs 6086 tok/s/chip at $0.10/M tokens, H100 runs 739 at $0.44/M. B300 is 326% cheaper per token; B300 delivers 723% more tok/s/chip.
Setting 71 tok/s/user as the target on DeepSeek R1, B300 produces 4027 tok/s/chip ($0.16 per million tokens) and H100 produces 281 ($1.16). B300 is 643% cheaper per token; B300 delivers 1335% more tok/s/chip.
At 97 tok/s/user interactivity on DeepSeek R1, B300 delivers 1933 tok/s/chip at $0.32 per million tokens; H100 delivers 154 tok/s/chip at $2.11. B300 is 548% cheaper per token; B300 delivers 1152% more tok/s/chip at this point. (Numbers reflect the default 8k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | B300:6086.1H100:739.5 | B300:4027.2H100:280.6 | B300:1932.5H100:154.3 |
| Cost ($/M tok) | B300:$0.103H100:$0.439 | B300:$0.156H100:$1.158 | B300:$0.325H100:$2.106 |
| tok/s/MW | B300:3203218H100:539778 | B300:2119595H100:204852 | B300:1017116H100:112644 |
| Concurrency | B300:~358H100:~132 | B300:~103H100:~41 | B300:~52H100:~17 |
Agentic inference metrics from the AgentX scenario and fixed-sequence inference metrics across models, hardware configurations, and serving parameters.
TCO $/chip/hr:
Source: SemiAnalysis Market July 2026 Pricing Surveys & AI Cloud TCO Model
Note: Disaggregated inference configurations (e.g., MoRI SGLang, Dynamo TRTLLM) calculate cost per million tokens per decode chip or per prefill chip, rather than per total chip count. This makes direct token cost comparison with aggregated configs not an apples-to-apples comparison.
Note: Disaggregated inference configurations (e.g., MoRI SGLang, Dynamo TRTLLM) calculate input throughput per decode chip or per prefill chip, rather than per total chip count. This makes direct input throughput comparison with aggregated configs not an apples-to-apples comparison.
Note: Disaggregated inference configurations (e.g., MoRI SGLang, Dynamo TRTLLM) calculate output throughput per decode chip or per prefill chip, rather than per total chip count. This makes direct output throughput comparison with aggregated configs not an apples-to-apples comparison.
Note: Disaggregated inference configurations (e.g., MoRI SGLang, Dynamo TRTLLM) calculate power per decode chip or per prefill chip, rather than per total chip count. This makes direct power comparison with aggregated configs not an apples-to-apples comparison.
Note: Disaggregated inference configurations (e.g., MoRI SGLang, Dynamo TRTLLM) calculate Joules per decode chip or per prefill chip, rather than per total chip count. This makes direct Joules per token comparison with aggregated configs not an apples-to-apples comparison.
Matching measurements exist, but their chip series are hidden.
Shift+Scroll to zoom • Drag to pan • Double-click to reset • Click a point to pin tooltip
DeepSeek R1 0528 671B 8K / 1K Total Tokens per $1 TCO vs. Interactivity