Telio
Workstations · AI Supercomputers · GPU Servers

Silicon built for the heaviest lifts.

T-rex is a family of workstations, AI supercomputers, and rack-scale GPU servers — from a walnut-slat desktop tower to a 748GB coherent-memory Grace Blackwell deskside, to a 10.8kW dual-EPYC cluster node with eight GPUs. Rendering, simulation, on-prem LLM training, and every workload that eats lesser machines for breakfast.

AI Performance
20 PFLOPS
Coherent Memory
748GB
Discrete GPUs
8 max
Maximus AI supercomputer
The Lineup

Eight machines. One mission.

Four categories. Workstations, a dynamic tower unit, rack-scale GPU clusters, and an AI supercomputer. Pick the machine that fits the workload.

Configurator

Build your T-rex.

Pick a chassis. Pick a finish. We spec every T-rex to your workload during the quote — this is the starting canvas.

T-rex configurator preview
Select a family
Select a system
Finish
Selected

2W

Threadripper PRO desktop workstation

Desktop tower with walnut-slat premium fascia.

Request a Quote →

Every T-rex is a custom order. Configured to your workload.

Request a Quote

Tell us what you're building.

Rendering farm? On-prem LLM training? Sim workstation? We'll spec the CPU, GPUs, RAM, and storage to fit.

  • · One business-day response
  • · Custom order — configured to your workload
  • · No pressure — we'd rather get the spec right
Selected Build
2W — Black / Walnut
Change your selection using the configurator above.

We only use your info to reply about your T-rex build. No spam, no lists.

Why T-rex

Built for workloads that don't wait.

Threadripper PRO & Xeon

Up to 96 Zen cores in a desktop tower, or dual Intel Xeon Scalable in a whisper-quiet workstation. Real workstation platforms — not consumer silicon reheated.

Grace Blackwell, deskside

Maximus packs an NVIDIA GB300 Superchip, 748GB coherent memory, and 20 PFLOPS of FP4 AI performance into a tower that runs on a standard office outlet.

8-GPU EPYC rack servers

Ultra 4 and Ultra 5 deliver eight double-width PCIe 5.0 GPUs, dual AMD EPYC 9005, and up to 10.8kW of redundant Titanium power for on-prem training and inference at scale.

ECC-always, PCIe 5.0 everywhere

DDR5 ECC memory across the entire lineup. PCIe Gen 5 to every slot, every M.2. Bit flips and bottlenecks don't ship on our machines.

Custom order, every time

Tell us the workload — rendering, AI training, GPU cluster, simulation. We spec the machine to fit and ship it to you. No mystery configurations, no compromise.

Serviceable, forever

Standard components, standard connectors, open BMC. You own the machine — not a warranty lock.

T-rex

Ready to spec your T-rex?

Start in the configurator or just tell us the workload — we'll build the machine that fits it.