{"resourceId":"signal65-vera-agentic-cpu-1-64x-20260915","versions":[{"version":"external-9f522938491b1278b32fa6ff6c4eb4c3f9488cc66de6cc6356fe4f9bc6abbb57","resource":{"id":"signal65-vera-agentic-cpu-1-64x-20260915","title":"Signal65: Vera 1.64× per-core on agentic CPU tasks vs EPYC 9755 (not whole-server)","organization":"Signal65","sector":"Independent AI infrastructure evaluation","geography":"Lab comparison (NVIDIA Vera vs AMD EPYC 9755)","publishedAt":"September 15, 2026","publicationDate":"2026-09-15","eventDate":"2026-09-15","sourceName":"Signal65","sourceLabel":"Independent evaluation","sourceUrl":"https://signal65.com/research/ai/validating-nvidia-vera-on-agentic-cpu-workloads/","evidenceClass":"independent-research","outcomeClass":"mixed","topics":["developers-agents","infrastructure","operating-model"],"finding":"Signal65 validated NVIDIA Vera on agentic host CPU work using Terminal-Bench 2 oracle (reference-solution) runs that exclude the LLM. Across 69 matched tasks under full load (one sandbox per physical core), Vera’s geometric-mean per-core task lifecycle was 1.64× faster than AMD EPYC 9755; 61/69 tasks favored Vera. Signal65 reproduced NVIDIA’s Vera pass profile hands-on, but the 1.64× x86 comparison remains based on NVIDIA’s paired measurements. Explicitly not measured: whole-server throughput, power, price, or memory capacity (88 Vera cores vs 128 EPYC cores).","sledRelevance":"Newly relevant as SLED teams evaluate Vera Rubin agentic stacks: GPU AgentX headlines omit CPU-bound sandbox time—this study bounds per-core CPU gains while warning against socket-level capacity claims.","evidence":"Date Sep 15, 2026; Terminal-Bench 2.0 oracle agent; 69 matched tasks geo-mean 1.64×; 61/69 favor Vera; 34 tasks ≥2×; 8 favor x86; hands-on Vera pass-profile match; 1.64× vs x86 still NVIDIA paired data; no whole-server/power/cost; 88 vs 128 cores.","architectureImplications":"Size agent fleets on per-core sandbox latency and core count together; do not translate 1.64× into rack throughput without socket math and power.","governanceImplications":"Require disclosure when vendor-paired measurements underpin independent write-ups; ask for whole-server PINNACLE follow-ons before procurement.","securityPrivacyImplications":"Bare-metal sandbox methodology; no confidential-computing or multi-tenant isolation claims.","caveats":"Oracle agent removes model variance; NVIDIA provided hardware and paired x86 data; subsetting to most agent-like 20 tasks (1.87×) is author judgment; PINNACLE CPU suite still developing.","streamIds":["nvidia"],"roles":{"sales":"When buyers chase AgentX GPU efficiency, introduce Vera CPU diligence: Signal65’s 1.64× per-core agentic sandbox result is real but narrow. Bound the pitch—no whole-server, power, or cost figures—and ask whether their agent loops are CPU-bound on tools and tests. Offer a scoped host sizing study rather than a rack-level guarantee tied to GPU marketing.","engineering":"Fit for agent platforms with heavy sandbox and tool loops. Architecture implication is per-core pinning and concurrency equal to physical cores in the study’s model, separate from GPU inference. Prerequisites: representative terminal tasks and fair core-count math (88 versus 128). Proof of value: reproduce a Terminal-Bench subset on candidate hosts with matched methodology and published limits.","delivery":"Work is a CPU host bake-off with oracle tasks—not an inference benchmark. Ownership sits with platform performance engineering. Acceptance: geo-mean and pass/fail tables plus explicit non-claims on whole-server throughput and power. Risks include over-indexing on the 1.87× “most agent-like” subset, ignoring core-count disadvantage, and treating NVIDIA-paired x86 data as fully independent."},"roleDetails":{"sales":{"customerProblem":"Agent fleets stall on CPU sandbox work even when GPUs are fast.","stakeholders":["AI platform","CFO","infra architects"],"discoveryQuestions":["Are agent loops CPU-bound on tools/tests?","Do you size by per-core or socket?","Will you accept paired-lab comparisons?"],"valueHypothesis":"Per-core gains can matter if tool loops dominate and socket math still closes.","potentialEngagement":"Two-week host bake-off on a Terminal-Bench subset.","unsupportedClaims":["1.64× whole-server","Lower $/agent guaranteed","Power win proven"]},"engineering":{"fit":"Strong for sandbox-heavy agent hosts; weak for inference-only nodes.","architectureIntegration":"One sandbox per core pinning model used in study; separate from GPU inference path.","prerequisites":["Task corpus","Bare-metal or equivalent isolation","Core-count-aware capacity model"],"constraints":["Paired NVIDIA x86 data","No power/cost","88 vs 128 cores"],"security":"Sandbox isolation assumed; CC not tested.","proofOfValue":["Subset geo-mean reproduced","Pass/fail tables","Socket math documented"]},"delivery":{"work":"Run oracle Terminal-Bench subset on Vera and x86 candidates.","dependencies":["Hardware access","Harbor/Terminal-Bench tooling"],"ownership":"Platform performance engineering.","skillsAdoption":"Perf engineers learn agent-host metrics beyond GPU.","governanceCheckpoints":["Disclose paired-data basis","No whole-server claim","PINNACLE follow-up planned"],"acceptanceCriteria":["Geo-mean reported","Core-count normalized view","Limitations published"],"risks":["Subset cherry-picking","Model variance excluded","Procurement overclaim"]}},"retrievedAt":"2026-09-23T03:13:00Z","enrichedAt":"2026-09-23T03:21:35Z","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Faster agent tool loops can improve developer assistant responsiveness; no a11y features claimed.","procurementImplications":"Hold Vera CPU purchases pending whole-server/power/cost evidence and competitive PINNACLE runs.","operatingModelImplications":"Add CPU-host KPIs to agent platform SLOs alongside GPU tokens/s.","sourceVerification":{"openedUrl":"https://signal65.com/research/ai/validating-nvidia-vera-on-agentic-cpu-workloads/","referenceExcerpt":"Vera finished the agentic task lifecycle 1.64x faster per core than the x86 processor","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}