Dedicated workstations
A real workstation. Hosted in your region. Streamed to your browser.
Not a VM with a sliced GPU. A physical machine: server-grade CPU, direct PCIe, all yours. Same browser-streamed experience as the rest of Canvex, with the hardware locked to you instead of pooled.
The pain
- Shipping a $12k workstation to a new hire on day one, then shipping it back when they leave eight months later. Imaging, asset tracking, the whole logistics tail.
- An EDA synthesis run that needs 64 cores and 512 GB RAM for two hours. Buying that workstation for the engineer who only uses it 6% of the year.
- Catia and SolidWorks license servers locked to a specific physical machine. The seat is on the workstation. The workstation is on the desk. The engineer is anywhere except the desk.
- Shared GPU slices that are perfect for ten engineers doing SolidWorks but fall apart when one of them opens a 4000-part assembly with realtime ray tracing.
With Canvex
- A physical workstation in the region you pick: server-grade Xeon W or EPYC, 96 to 1024 GB RAM, full-card NVIDIA RTX 6000 Ada / RTX 4090 / L40S / A100 / H100, no virtualization layer between the OS and the hardware.
- Reserved 1:1 to one user (or one license seat). Nobody else's workload lives on this box.
- Stream it to whatever the engineer is on: Chromebook, MacBook, a Windows laptop, a tablet. The pixels come from the workstation; the laptop doesn't need to know what a GPU is.
- Monthly reservation pricing instead of per-day metered. Predictable budget line on the project P&L.
Recommended config
Typical fit
CAD / AEC seat: Xeon W-3400 series, 128-256 GB RAM, RTX 6000 Ada (48 GB) or RTX 4090 (24 GB). Comfortable for Revit, Catia, NX, large SolidWorks assemblies, Lumion, V-Ray. Roughly $700-900/mo.
EDA workstation: dual EPYC 9554 (128 cores total), 512-1024 GB RAM, optional one or two H100s for hardware emulation runs. For synthesis, place-and-route, full-chip simulation. Roughly $1,800-3,500/mo.
AI / ML workstation: 32-64 core EPYC or Xeon, 256-512 GB RAM, 1-4x H100 or H200. For training-adjacent work that's too big for a shared GPU slice but too small for a full cluster. Quote on request.
What's working for you under the hood
Direct PCIe
No virtualization tax on GPU, NVMe, or USB. The OS sees the actual hardware. Driver behavior, performance counters, and ISV-cert testing all behave like a workstation under your desk.
Locked to your tenant
Nothing else runs on this machine. No noisy neighbors, no scheduling jitter, no cross-tenant performance variance. The serial number is yours for the term of the rental.
BMC console access
Out-of-band management for the rare moment the OS won't boot. Recover, reinstall, or just watch POST. Same affordance you'd have on a workstation in your own rack.
Workspace, not a hardware lock-in
Move your workspace (the OS image, tools, files, settings) onto a bigger SKU for a project, back down between projects. The workspace doesn't care which physical machine it's running on this month.
Monthly reservation
Flat monthly rate per workstation. Predictable budget, no per-minute metering. Annual commitments available for orgs that want a lower rate.
Same UX as the shared tierBusiness+
Browser streaming, file transfer, clipboard, session recording, audit log. Everything the shared cloud-desktop product does, the dedicated workstation does too.
We don't think dedicated workstations are the right answer for most workstations. Most teams are better served by a shared GPU tier that scales up and down with the work. Dedicated is for the specific cases where shared physics breaks down: huge assemblies, ISV-cert workflows, EDA runs that monopolize a host for hours, workloads that genuinely need every PCIe lane. When you're in one of those cases, "buy a workstation, then ship it" stops being a sensible answer and this becomes one.
Not shipping yet. Get in line.
We're standing up the first machines now. Drop your email and what you'd run on one; we'll reach out when there's hardware to allocate.