The questions our customers ask most often. Searchable. Categorised. If yours isn't here, email us.
A single-cluster install on existing Kubernetes takes under one day. Bare-metal greenfield, including K8s install, takes 2–3 days. Federated multi-cluster: 1–2 weeks per added cluster. Hybrid Cloud: a few hours after credentials are configured.
Three tiers: Academic (free for accredited universities and 501(c)(3) labs, up to 32 GPUs), Professional (per-GPU annual subscription), Enterprise (custom, includes 24×7 support and SLA). Talk to us for a quote.
NVIDIA H200, H100, A100, L40S, L4, A40, A30, RTX 6000 Ada — driver 535+, CUDA 12.x. AMD MI300/MI250 (production). Intel Gaudi 3 on the 2026 H2 roadmap.
No. PyTorch, TensorFlow, JAX, NeMo, Megatron — all run unchanged. CNLab intercepts CUDA at the driver level; your model.cuda() call gets a slice instead of a full GPU.
Only if you allow burst-out and the workload is not pinned. Sensitive datasets and projects can be marked on-prem-only with one click; the scheduler then never bursts those workloads.
SOC 2 Type II is on the 2027 Q1 roadmap. Type I attestation in H2 2026. ISO 27001 Stage 2 audit Q3 2026. K-ISMS-P (Korea) Q4 2026. More on security.
Standard tier: business-day email response. Premier tier: 24×7 phone + Slack, 1-hour SEV-1 response, 4-hour SEV-2. Enterprise tier: dedicated CSM and quarterly architecture reviews.
Yes. CNLab ships as a fully offline-capable Helm chart with a private container registry mirror. No outbound connectivity required. Updates delivered as signed bundles.
Targeted 2026 H2. Currently in private preview with two design partners. Same scheduler abstractions, different device plugin.
Pricing structure stays the same. The pricing page goes public with the billing module in Phase 3 — Q4 2026 target. Existing customers grandfathered for the rest of their contract term.
Yes. Email roadmap@cnlab.ai from a corporate domain — we'll send a current PDF under NDA.
Direct line to the team. Most questions get a same-day reply.