NVIDIA Vera CPU runs agentic tasks 1.64x faster per core across 69 Terminal-Bench 2 tasks
ryanshrout · x · 2026-09-16
Agentic AI spends much of each session's wall clock on the CPU — compiling, running tests, cloning repos and moving files between model calls — leaving GPUs idle. Signal65, working with NVIDIA, benchmarked the Vera CPU against a leading x86 server processor using one agent sandbox per core, no model in the loop, timed from sandbox creation to teardown.
Key results:
- Vera completed the agentic task lifecycle 1.64x faster per core across 69 matched Terminal-Bench 2 tasks
- 61 of 69 tasks favored Vera; 34 by 2x or better
- On the 20 tasks closest to daily agent work (git repair, database recovery, vulnerability fixes, server setup), the per-core advantage was 1.87x
- Poster Ryan Shrout's own hands-on runs on the same Vera system were even faster than NVIDIA's figures
Takeaway: when planning agent capacity, the CPU — not the GPU — decides how many concurrent sessions a rack can keep busy.
Related event: NVIDIA Vera CPU Hits 1.64x Per-Core Lead in Agent Workloads(2 posts)→
More from Infra
- Lithos: stop treating AI benchmarks as proof, define your own metrics — JiaZhihao · 2026-09-16
- Cheap third-party open LLM providers ranked, coral bricks tops the list — opensourcecolumbus · 2026-09-16
- Could an Xbox Series X cluster with 160GB unified memory host local LLMs? A Reddit thought experiment — RecursiveCTE · 2026-09-16
- Rob Mulla marks one year at Google scaling TPU inference with vLLM — Rob_Mulla · 2026-09-16
- Jev claims new frontier model 40-400x cheaper and 20-200x faster — albelfio · 2026-09-16
- Nvidia B200 Compute Prices Jump 21% in a Month as AI Demand Outstrips Supply — JOBhakdi · 2026-09-16