Muse Spark 1.3/Open 816B · Chip comparison

Muse Spark 1.3/Open 816B — B200 vs Vera Rubin NVL72

Head-to-head AI inference benchmark comparison of B200 (NVIDIA Blackwell) and Vera Rubin NVL72 (NVIDIA Vera Rubin) on Muse Spark 1.3/Open 816B. 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 →

Setting 200 tok/s/user as the target on Muse Spark 1.3/Open 816B, B200 produces 19,742 tok/s/chip ($0.024 per million tokens) and Vera Rubin NVL72 produces 82,243 ($0.012). Vera Rubin NVL72 is 50% cheaper per token; Vera Rubin NVL72 delivers 4.2× the tok/s/chip.

At 293 tok/s/user interactivity on Muse Spark 1.3/Open 816B, B200 delivers 11,802 tok/s/chip at $0.041 per million tokens; Vera Rubin NVL72 delivers 58,124 tok/s/chip at $0.017. Vera Rubin NVL72 is 58% cheaper per token; Vera Rubin NVL72 delivers 4.9× the tok/s/chip at this point.

B200 posts 7,065 tok/s/chip for $0.068 per million tokens at 386 tok/s/user on Muse Spark 1.3/Open 816B; Vera Rubin NVL72 posts 45,063 tok/s/chip for $0.022. Vera Rubin NVL72 is 67% cheaper per token; Vera Rubin NVL72 delivers 6.4× 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 →

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)
B20019,742Vera Rubin NVL7282,243
B20011,802Vera Rubin NVL7258,124
B2007,065Vera Rubin NVL7245,063
Cost ($/M tok)
B200$0.024Vera Rubin NVL72$0.012
B200$0.041Vera Rubin NVL72$0.017
B200$0.068Vera Rubin NVL72$0.022
tok/s/MW
B20011,544,865Vera Rubin NVL7224,922,241
B2006,901,569Vera Rubin NVL7217,613,280
B2004,131,832Vera Rubin NVL7213,655,433
Concurrency
B200~13Vera Rubin NVL72~46
B200~7Vera Rubin NVL72~30
B200~8Vera Rubin NVL72~24