CNLab.ai Solutions

Three deployment models. One operating model.

Choose the cluster topology that matches your scale today. Expand without rewrites tomorrow. Every CNLab deployment shares the same scheduler, the same APIs, and the same admin console — only the resource boundary changes.

Supported Environments

Your code asks for a GPU. We route the rest.

Auto-detection of deep-learning GPU resource allocation timing — across every environment your researchers already use.

📓

Jupyter Notebook

Open a notebook, import torch, call .cuda() — CNLab routes the request to a sliced or full GPU on the closest cluster. No code changes, no kernel surprises.

💻

VS Code & IDE Forks

Cursor, Windsurf, JetBrains — the CNLab agent runs as a remote runtime. Workspace files, secrets, and SSH keys are forwarded transparently.

⌨️

CLI

cnlab run train.py --gpu h100-25 queues your script on the next available 25%-block H100. Logs stream back; the GPU releases on exit.

Decision Helper

Platform Selection Matrix

Choose by org size, GPU pool, and security posture. Every CNLab tier is forward-compatible — Single grows into Multi grows into Hybrid.

Dimension	On-Premise Single	Multi On-Premise	Hybrid Cloud
Recommended for	Labs, small AI teams	Multi-campus universities, mid-sized AI	Enterprise, AI scale-ups
Typical pool size	4–32 GPUs	32–256 GPUs	128+ GPUs + cloud burst
Initial cost	Lowest	Medium	Medium + cloud opex
TCO	Lowest	Low	−50%+ vs. cloud-only
Failover	Single cluster	Cross-cluster live migration	Cross-cluster + cloud
Heterogeneous GPU	—	NVIDIA + AMD	NVIDIA + AMD + cloud
Data residency	100% on-prem	100% on-prem	On-prem first, encrypted egress
Min. allocation unit	1% Block	1% Block	1% Block
Setup time	Days	Weeks	Weeks