Stop renting compute. Own it.
Rendering, simulation and model training all need dedicated CPU and GPU resources — and cloud instances bill you every month for hardware you never keep. We design workstations and rackmount systems around your actual workload, then support them out of Utah for as long as you own them.
Four shapes. One conversation.
Every system is configured per order, so start with the shape that fits your rack — or your desk — and we'll scope the rest with you.
Workstation
Maximum single-seat performance, at a desk.
- Threadripper & Threadripper PRO
- Up to 4 double-width GPUs
- Quiet enough to sit next to
Rack workstation
Workstation hardware, moved off the desk and into the rack.
- Same silicon as the tower
- Remote-access ready
- Lower cost than a true server
Rackmount server
Shared compute your whole team pulls from.
- EPYC core density
- Expansive memory capacity
- IPMI out-of-band management
GPU compute
Training and inference that needs every GPU you can feed.
- Multi-GPU in a single chassis
- RTX PRO 6000 Blackwell class
- Power and cooling designed for sustained load
What are you running?
Form factor follows the workload. Here's where each kind of work usually lands.
AI training & inference
Bound by GPU count and VRAM
Model size decides your floor: it has to fit in VRAM before throughput matters at all. After that, more GPUs in one chassis beats more chassis.
Usually GPU compute (4U)
Rendering & VFX
Bound by Cores, then GPU
CPU renderers scale almost linearly with cores; GPU renderers care about VRAM ceilings. Which engine you use changes the machine completely.
Workstation for a seat, server for a farm
Simulation & CAE
Bound by Memory bandwidth
Solvers are frequently starved by memory bandwidth long before they run out of cores, so populating every channel matters more than raw core count.
Usually workstation or rack workstation
Virtualization & shared seats
Bound by Core density and RAM
Many concurrent users, none of them saturating the machine. Density and headroom win; peak single-thread speed rarely does.
Usually a rackmount server
Not sure which applies? Describe the work in the quote form and an engineer will tell you what it actually needs — including when the answer is a smaller system than you expected.
One business
day. No SDRs.
Fill this in once. A real Xidax engineer — not an offshore SDR — will email you a PO-ready quote within 24 hours. If we need clarification, we'll call you. If we don't, you'll just get the quote.
Fleet quote request
All fields except "Use case" are optional, but more detail means a tighter quote. Average completion: 3 minutes.
The procurement
FAQ.
If your question isn't here, email business@xidax.com — Devin replies the same day.