Home /
Built to render. Built to reason. One 2U server.
The 2U Q-Power is our rack-mount GPU server, engineered for studios that render and teams that run AI — often the same team, on the same box. Specified by engineers, built in Hampshire, tested on the benchmarks that actually matter.
2700 W
PSU
Up to 1TB ECC
Memory
2x M.2 + PCIe 5
Storage
Dual 10GbE + IPMI
Network
2U – 890mm
Chassis
UK-Built
Origin
Meet the Machine.
Take a closer look at the 2U Q-Power. Hear from our Head of Sales and Marketing team as they talk through its capabilities, and the thinking behind its design.
Pick the build for the workload in front of you.
Both ship in the same 2U chassis. The GPU and memory fit-out is the only thing that changes — because a render farm and an LLM server ask different things of the same hardware.
PT9W – Workstation Spec
|
GPU 2× NVIDIA GeForce RTX 5090 |
CPU AMD Threadripper |
Total Memory 512GB 5600MHz RDIMM ECC |
Total GPU VRAM 64GB GDDR7 |
|
Features IPMI, Dual Intel X710 10G NICs, RHLX, Windows 11 Pro / Workstation |
AI TOPS (Per card) ~3,352 TOPS |
Best for Cost-effective speed. CV inference, broadcast, rendering. |
SE5 – Server Spec
|
GPU 4× NVIDIA RTX PRO 4000 / 5000 / 6000 |
CPU AMD EPYC |
Total Memory 1TB LRDIMM ECC |
Total GPU VRAM 96GB / 192GB / 384GB |
|
Features IPMI, Dual Broadcom 10G NICs, RHLX, Windows Server OS |
AI TOPS (Per card) ~1,600 TOPS (4000)~2,000 TOPS (5000)~4,000 TOPS (6000) |
Best for PRO 4000: models to 128GB, edge AI.PRO 6000: 70B+ LLM, max VRAM density. |
Two disciplines. Equal billing
The 2U Q-Power wasn’t built for AI and adapted for rendering, or the other way round. Both were the brief from day one, and both get validated before the server ships.
512GB ECC workstation memory
Handle large projects, complex simulations and heavy creative workloads with reliable, error-correcting memory.
Data never leaves the building
Sensitive data stays on your network — supports data requirements including GDPR, HIPAA and SOC 2.
Built for photorealistic 3D at studio speed
V-Ray CUDA and V-Ray RTX scores lead the field — the two industry-standard tests for archviz and photoreal rendering.
Under 40ms latency vs 1.5 seconds on cloud APIs
A 97% cut in response time for real-time workflows that can’t wait on a round trip to someone else’s data centre. Nucamp, 2025
OctaneBench-proven
Validated for high-end visualisation and VFX pipelines, not just synthetic benchmarks.
Keeps working if the internet doesn’t
No dependency on a cloud provider staying up. Lose the connection and the server keeps running.
Owned outright
Render around the clock without per-seat cloud render farm fees eating into every job.
You own the model, not a subscription to it
Full access to the weights, so you can fine-tune on your own data instead of renting someone else’s.
Tested before it leaves Hampshire
Every configuration runs Cinebench under sustained load before it ships.
Up to 18x cheaper per million tokens
Measured against premium cloud AI APIs, at consistent load, over time.
The rendering numbers, not just the AI ones.
Dual RTX 5090 compared to a 4x RTX 4000 ADA build across the four benchmarks studios use to spec hardware.
≤85°C
GPU temperature stayed within healthy limits across every test at full load, with no thermal throttling.
0 Errors
Both configurations completed the full stress-test suite without a single error.
~218W
Combined power draw during live AI inference across the 4× RTX 4000 Ada build—efficient enough to run 24/7.
Higher-concurrency figures for the RTX PRO 5000 and PRO 6000 Blackwell builds are extrapolated from single-GPU benchmarks and independently verified — early numbers point to even more headroom for large-model AI work, with Cinebench 2026 scores already ahead of the 5090 pairing.
Built for the people who spec their own hardware.
Computer Vision
Real-time video inference at the point of capture, with no cloud round-trip. Runs industrial image comparison, grading and classification at line speed, and the same CUDA stack used on drone and autonomous systems and subsea ROV deployments — so what’s built on the desk matches what ships to the field.
Studios and visualisation teams
Archviz houses, VFX studios and anyone running V-Ray, OctaneBench or Blender pipelines who’s tired of paying by the render.
Teams running models on their own data
Organisations that need a 70 billion parameter model running locally, on hardware they own, with data that never leaves the network.
Engineers who spec, not configure
Buyers who know exactly what a workload demands and want a server built to that spec, not a generic box with a spreadsheet of options.
Arrive with a partial spec. Leave with the right answer.
Tell us what you’re rendering, or what you’re running, and we’ll tell you which build gets it done.
Enter your email to receive updates for the 2U Q-Power below.