Edge Compute
Hardware

GPU servers that are
ready on day one.

Pre-built, burn-in tested RTX 5090 nodes, configured for our network before they ship. Plug it in, boot the included installer, and it's online. Use it yourself, host it with us, or both.

2× RTX 509064 GB GDDR7 total
~210FP32 TFLOPS combined
72 hburn-in before shipping
1 USBto install everything
Reference build

Edge Node · 2× RTX 5090

A workstation-class node balanced for two 575 W GPUs: full PCIe 5.0 bandwidth, twice as much RAM as VRAM, and power and cooling headroom for running at 100% around the clock.

ComponentSpecification
GPU2× ASUS ProArt GeForce RTX 5090 OC
32 GB GDDR7 each · Blackwell · 5th-gen tensor cores
CPUAMD Ryzen 9 9950X
16 cores / 32 threads
MotherboardASUS ProArt X870E-Creator WiFi
Two PCIe 5.0 slots (×8/×8), 10 GbE
Memory128 GB DDR5-5600
2× 64 GB · ≥ total VRAM, as large models need
Storage4 TB Crucial T705 PCIe 5.0 NVMe
Fast model loading, room for datasets and images
PowerSeasonic Prime PX-1600, 1600 W Platinum
ATX 3.1, native 12V-2×6 for both GPUs
Cooling360 mm AIO + high-pressure 140 mm fans
Lian Li GA II Trinity · tuned for sustained load
ChassisLian Li LANCOOL III full tower
Room for two large GPUs with airflow between them
SoftwareUbuntu Server 24.04 + host agent
Driver, Docker, security hardening, auto-updates

Why this configuration

  • ✓
    No bandwidth bottleneckPCIe 5.0 ×8 per GPU equals the bandwidth of a PCIe 4.0 ×16 slot.
  • ✓
    RAM ≥ VRAMLoading and quantising large models needs host memory at least equal to GPU memory.
  • ✓
    Power headroom~1.4 kW peak draw on a 1.6 kW Platinum PSU, so it stays efficient and cool.
  • ✓
    Quiet enough for an officeTower format, no rack required. Wired Ethernet recommended.

You own it

The node is your hardware. Use it for your own work, rent it out on our network when it's idle, take it off the network or sell it whenever you want. Hosting income depends on demand and is never guaranteed.

How hosting works
How it works

From order to online

1

Tell us what you need

Pre-built node, bare GPUs or a custom build, quantity and timeline. We reply with a quote within 1–2 business days.

2

We build & certify

Assembly, BIOS tuning, OS install and a full certification run before anything ships.

3

Plug in & go

Connect power and Ethernet, boot, and the node registers itself. Approve it and it's live.

Certified

Every node is tested before it ships

You receive the certification report with your node.

🔥

72-hour burn-in

Both GPUs and the CPU at full load for three days.

🧠

VRAM validation

Complete memory pattern test on every GPU — no defective cells.

📊

Benchmark scoring

Training and inference throughput compared against our reference node.

🌡️

Thermal profile

Temperatures and clocks recorded under sustained load.

Plus our automatic security self-check: network isolation, user namespaces and GPU isolation.

Get a quote

Tell us what you need

Pricing depends on GPU availability, quantity and whether you want us to build it. We'll come back with options within 1–2 business days.

Building your own? Follow the host setup guide.