From the NVIDIA edition of September 16, 2026
GPU Operator 26.7.0: GPUCluster DRA management and passthrough allocation pitfalls
NVIDIA · Kubernetes GPU platform software · Global (product documentation)
- Publisher
- NVIDIA GPU Operator Documentation
- Original publication
- 2026 (v26.7.0 release notes)
- Source retrieved
- 2026-09-17
What happened
GPU Operator 26.7.0 adds GPUCluster CR to manage DRA Driver for GPUs (full GPU and preconfigured MIG via ResourceClaims), KubeVirt VFIO passthrough with DRA, driver 595.91.07 stack updates, and in-place patch restarts. Requires Kubernetes v1.34.2+, driver 580+, CDI runtime; cannot mix GPUCluster with ClusterPolicy. Known issue: concurrent container and VFIO claims on the same physical GPU can race—submit serially or dedicate nodes.
Why it matters
Critical platform constraint for SLED clusters adopting Dynamo shadow recovery or NIM DRA allocation—both need modern K8s DRA. Mixing VM passthrough and containers without node dedication is a concrete failure mode.
Evidence and measured results
Opened latest release notes for 26.7.0 New Features and Known Issues describing GPUCluster, K8s v1.34.2+ requirement, mutual exclusion with ClusterPolicy, and DRA passthrough double-allocation race.
Limitations and uncertainty
Release notes are version living docs; some DRA capabilities remain alpha/disabled by default.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-09-17; this does not change the original publication date. Each role follows the same checklist so takeaways compare across sources.
Sales
Role takeaway
- Customer problem
- Buyers want DRA-based GPU sharing and VM+container flexibility without knowing version floors.
- Stakeholders
- platform eng
- virtualization team
- security
- Discovery questions
- Kubernetes version today?
- Using ClusterPolicy or planning GPUCluster?
- Do VMs and containers share GPU nodes?
- Value hypothesis
- GPUCluster DRA unlocks modern allocation when version prerequisites are met.
- Potential engagement
- Platform readiness review for 26.7.0 DRA features.
- Unsupported claims
- Works on any K8s 1.28 cluster
- Safe simultaneous container+VFIO claims on one GPU
Pre-sales engineering
Role takeaway
- Fit
- Fit for clusters 1.34.2+ adopting DRA; migration required from ClusterPolicy.
- Architecture and integration
- GPUCluster manages DRA driver, DCGM, validation workload; ResourceClaims select GPUs/MIG.
- Prerequisites
- K8s 1.34.2+
- driver 580+
- CDI runtime
- Constraints
- No ClusterPolicy coexistence
- Passthrough race known issue
- Some DRA alpha
- Security
- Separate node pools for VFIO vs containers when isolation required.
- Proof of value
- ResourceClaim binds
- No dual-prepare failures
- DCGM metrics present
Delivery
Role takeaway
- Work
- Upgrade Operator, cut over to GPUCluster, validate claims, document node roles.
- Dependencies
- K8s upgrade
- workload owners
- KubeVirt team if VMs
- Ownership
- GPU platform team.
- Skills and adoption
- DRA ResourceClaim literacy for app teams.
- Governance checkpoints
- Upgrade CAB
- Coexistence ban confirmation
- Known-issue communication
- Acceptance criteria
- GPUCluster Ready
- Test claims for MIG and full GPU
- Written node dedication policy if VFIO used
- Risks
- Partial migration leaving ClusterPolicy
- Silent claim races
Implementation considerations
Lighthouse Advisory interpretation across the operating dimensions a public-sector buyer must settle before this evidence becomes a design. Each note answers the question under its heading for this specific source.
Architecture and integration
What must connect, and where does the AI sit in the workflow?
Choose GPUCluster or ClusterPolicy—not both; align K8s minor version before promising DRA features used by Dynamo/NIM.
Governance
Who approves, reviews and stays accountable for outcomes?
Platform upgrade boards must approve 1.34.2+ moves in regulated clusters.
Security and privacy
What data, permissions and controls need testing?
VFIO passthrough + containers on one node needs clear tenancy isolation policy.
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
Stable GPU scheduling improves shared research access reliability.
Procurement
What should contracts, pricing and exit terms secure?
Require supported K8s versions in managed Kubernetes contracts when DRA features are in scope.
Operating model
Which teams own the service once it runs?
Platform must own DRA lifecycle separately from app teams.
Publication history
- 2026-09-16NVIDIA · Issue 119 resources
Stable resource ID: nvidia-gpu-operator-26-7-0-dra-constraints-2026