{"resourceId":"nvidia-cuda-13-5-vera-rubin-ga-mps-v3-sharing-controls-20261104","versions":[{"version":"external-5e00d42ccb7ae089d5a46d67d3b26e7cf30345e7b9d9616df67ce4d14bf1153a","resource":{"id":"nvidia-cuda-13-5-vera-rubin-ga-mps-v3-sharing-controls-20261104","title":"CUDA 13.5 makes Vera Rubin support generally available and adds MPS v3 GPU-sharing controls: weighted timeslicing, SM borrowing/lending and cgroup-based routing","organization":"NVIDIA","sector":"GPU software platform","geography":"Global","publishedAt":"2026-11-04","publicationDate":"2026-11-04","eventDate":"2026-11-04","sourceName":"NVIDIA Technical Blog","sourceLabel":"NVIDIA Technical Blog: NVIDIA CUDA 13.5 Brings NVIDIA Vera Rubin GA Support","sourceUrl":"https://developer.nvidia.com/blog/nvidia-cuda-13-5-brings-nvidia-vera-rubin-ga-support/","evidenceClass":"vendor-claim","outcomeClass":"emerging","topics":["infrastructure","developers-agents"],"finding":"CUDA 13.5 moves Vera Rubin support from early developer access to general availability and extends Windows on Arm support to DGX Station for Windows. Multi-Process Service v3 adds weighted timeslicing, SM partition load balancing with borrowing and lending, cgroup v2 subtree-to-namespace routing, per-namespace SM partition assignments, and a polling-based fallback for memory accounting when cgroup notifications are missing or unreliable. Other changes cover CUDA Tile multi-GPU communication, cuda.core 1.2 and CCCL 3.5, including reproducible floating-point scans.","sledRelevance":"Shared campus and state GPU clusters often split GPUs among many users. MPS v3 gives administrators finer controls over that sharing, following the CUDA 13.4 Kubernetes memory enforcement covered on Oct 27. Vera Rubin GA matters to institutions planning next-generation purchases.","evidence":"Vendor release notes; no performance or fairness measurements for the new MPS features.","architectureImplications":"Sharing policy can follow the launcher's cgroup groupings, which ties scheduler configuration to GPU partitioning.","governanceImplications":"Weighted timeslicing turns GPU-share priorities into an explicit policy that someone must set.","securityPrivacyImplications":"Not addressed; MPS sharing is not presented as a security isolation boundary.","caveats":"Vendor announcement; feature claims are untested here; borrowing behaviour under contention is not quantified.","streamIds":["nvidia"],"roles":{"sales":"For clients that run shared GPU clusters, CUDA 13.5 matters for two reasons. Vera Rubin support is now generally available, which removes one software-readiness question for buyers planning next-generation systems. MPS v3 also gives administrators more say over how many users share a GPU. Present both as release-note capabilities. NVIDIA gives no utilisation or fairness figures, so savings from denser sharing must be measured on the client's workloads.","engineering":"From an engineering view, MPS v3 lets administrators weight timeslices per MPS server, let SM partitions borrow and lend idle capacity, route clients by cgroup v2 subtree to MPS namespaces, and set SM partitions per namespace. A polling fallback covers unreliable cgroup memory notifications at some per-eviction cost. Test these with the site's Slurm or Kubernetes launcher, and do not treat MPS sharing as a security boundary between tenants.","delivery":"Delivery teams should treat the new sharing controls as policy that needs an owner. Decide who sets timeslice weights and partition sizes, how changes are reviewed, and how users are told. Run a pilot on one partition, compare queue wait and job completion before and after, and keep a rollback configuration. Version upgrades to CUDA 13.5 should follow the site's normal driver and toolkit testing."},"roleDetails":{"sales":{"customerProblem":"Shared GPU clusters struggle to allocate capacity fairly among many users.","stakeholders":["Research computing director","HPC administrators","Faculty user committee"],"discoveryQuestions":["How are GPUs shared today?","Are users waiting while GPUs sit partly idle?","Is a Vera Rubin purchase being planned?"],"valueHypothesis":"Finer sharing controls may raise utilisation on shared GPUs; gains must be measured locally.","potentialEngagement":"A sharing-policy pilot on one partition after a CUDA 13.5 upgrade.","unsupportedClaims":["Any utilisation gain","Tenant security isolation via MPS"]},"engineering":{"fit":"Fits multi-user GPU clusters using MPS.","architectureIntegration":"MPS v3 namespaces mapped from cgroup v2 subtrees; per-namespace SM partitions.","prerequisites":["CUDA 13.5 toolkit and drivers","cgroup v2","Launcher integration"],"constraints":["Feature behaviour under contention unquantified","Driver/toolkit upgrade testing needed"],"security":"MPS is a sharing mechanism, not presented as an isolation boundary.","proofOfValue":["Measure queue wait and throughput before and after on one partition","Test polling memory accounting overhead"]},"delivery":{"work":"Introduce governed GPU-sharing policy.","dependencies":["CUDA 13.5 upgrade","Scheduler configuration"],"ownership":"HPC operations manager.","skillsAdoption":"Administrator training on MPS v3 and cgroups.","governanceCheckpoints":["Policy owner named","User notice before weight changes"],"acceptanceCriteria":["Proposed: documented sharing policy","Proposed: before/after utilisation report"],"risks":["Unfair allocation","Upgrade regressions"]}},"retrievedAt":"2026-11-05T03:56:52Z","enrichedAt":"2026-11-05T03:59:54Z","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Administrators need MPS and cgroup skills to use the new controls.","procurementImplications":"Vera Rubin GA support in CUDA reduces one software-readiness risk for buyers; hardware availability is separate.","operatingModelImplications":"Allocation policy for shared GPUs becomes more configurable and needs an owner.","sourceVerification":{"openedUrl":"https://developer.nvidia.com/blog/nvidia-cuda-13-5-brings-nvidia-vera-rubin-ga-support/","referenceExcerpt":"NVIDIA Vera Rubin support moves from early developer access to general availability (GA), ready for enterprise production.","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}