GPU workstations
The workstation has a real GPU. Your laptop doesn't have to.
NVIDIA and Intel passthrough, shared, or vGPU. Stream the workstation to a browser tab; run it from a Chromebook, a Mac, or the laptop you wish you hadn't bought a discrete GPU for.
The pain
- CAD admin time spent imaging machines, deploying license servers, and shipping hardware to remote engineers.
- Engineers waiting 40 minutes for an Inventor session to be responsive after opening a 200-part assembly over corporate VPN.
- Buying $4,500 workstations that sit idle 80% of the week, then thermal-throttling during the 20% that matters.
- Pinned licensing: the seat is on the machine, the machine is on the desk, the engineer is anywhere except the desk.
With Canvex
- Real NVIDIA (L4, RTX 4090, A40, A6000, A100, H100) or Intel (Flex 140, Flex 170) GPUs attached as passthrough or sliced as vGPU.
- 1–80 GB VRAM tiers — pick what the workload needs, change it next week when the workload changes.
- Hosted close to where the engineer actually sits via regional routing. <50ms typical browser latency in most US cities.
- Pay by the day while it's running. Stop the workstation and the GPU rate stops too.
Recommended config
Most engineering teams
Start with the Performance tier (5–8 vCPU, 9–16 GB RAM) plus a 12 GB GPU. That's roughly a Flex 140 full-card or a vGPU slice of an L40 — comfortable for Solidworks, AutoCAD, Revit, mid-size Catia assemblies, and Inventor sessions up to a few hundred parts.
For heavy rendering, simulation, and CFD: Power or Workstation tier with a 24 GB GPU (L4 / RTX 4090) or 48 GB (A40 / A6000 / L40). Bump up for the project, bump back down between projects.
What's working for you under the hood
Passthrough or vGPU
Whole card pinned to one workstation, or a vGPU slice for cheaper per-workstation pricing. NVIDIA's licensed mediated devices supported.
Workspace templates
Office, dev, CAD, EDA, data-science presets. Each one is an image with the canonical tooling for the workload already installed.
Per-tenant network
Each org gets its own isolated private network. License servers, CAD vault, NFS shares — drop them on the workstation's network.
Resize without rebuild
Move from a 12 GB Flex 140 to a 48 GB L40 between sessions — the workspace, files, and tooling stay put. Stop the workstation, change the spec, restart.
Pause when idle
Auto-pause stops the compute meter when nobody's connected. A 9-to-5 workstation bills like a 9-to-5 workstation.
Snapshot before risky workBusiness+
Snapshot before a driver update, a Plugin install, or a Catia version bump. Restore in seconds if it goes sideways.
We host real GPUs because they're what the work needs. Software rasterizers and "vGPU-lite" CPU-rendering tiers exist on most cloud-desktop platforms — they look like GPU rows on a pricing page, but they don't render anything an engineer recognizes as CAD. Canvex's GPU tier is hardware attached to your workstation's PCI bus.
Your first desktop, in five minutes.
3-day free trial. Cancel anytime before it ends.