NVIDIA Vera Rubin benchmarks show 35x cheaper agentic coding tokens
brianryhuang · x · 2026-08-25
NVIDIA released the first on-silicon benchmarks for the Vera Rubin NVL72, measured against the SemiAnalysis AgentX workload using the DeepSeek V4 Pro model. The results show up to 30x higher throughput per megawatt and 35x lower token costs compared to GB300 NVL72.
Key drivers include:
- Full-stack Co-design: Enhanced TensorCore FLOPs, HBM4 bandwidth, 6th Gen NVLink/Spectrum-6 SPX networking, and Vera CPUs for faster tool calls.
- Disaggregated Serving: Separating prefill and decode with rate matching, optimized for the Rubin+LPX architecture.
- Expert Parallelism: Spreading experts across many GPUs to scale capacity.
Related event: NVIDIA Shows Vera Rubin Silicon: Up to 30x Agent Throughput over GB300(5 posts)→
More from coding & agent
- raylib 6.2 ships with 300+ improvements and 100 new contributors, 20 new API functions — jsuarez · 2026-10-16
- "Vibecoding" is as misleading a label as "AI" and "SI," dev complains — eschadiol · 2026-10-16
- Dev runs Pstack as a Grok bot to audit his own pipeline — it surfaces a pile of hidden bugs — alexcovo_eth · 2026-10-16
- Doc2Harness: New benchmark finds best coding agent passes only 53% across 329 cases — _akhaliq · 2026-10-16
- Perplexity open-sources SPACE, the Firecracker microVM sandbox behind Perplexity Computer — perplexity_ai · 2026-10-16
- Blackstone goes all-in on Factory for autonomous software development — matanSF · 2026-10-16