2U Q-Power PT9W
When demanding AI, rendering and visualisation workloads outgrow conventional workstations, the 2U Q-Power delivers serious graphics performance with up to four NVIDIA RTX PRO GPUs in a compact, rackmount platform. Run large models locally, accelerate photorealistic rendering and keep sensitive workloads on your own network. Q-Power gives studios, engineers and AI teams high-performance GPU compute they can own, manage and scale on their terms.
Built in the UK for organisations that need more compute power without relying on costly cloud infrastructure, it gives you dedicated performance, predictable operating costs and complete control over your data.
Choose a configuration built around your workload. For rendering, VFX, computer vision and broadcast applications, the PT9W combines an AMD Threadripper processor with dual NVIDIA GeForce RTX 5090 GPUs, delivering 64GB of GDDR7 VRAM and 512GB of ECC memory.
For larger AI and LLM deployments, the SE5 pairs AMD EPYC processing with up to four NVIDIA RTX PRO GPUs, up to 384GB of GPU VRAM and 1TB of ECC system memory.
Both system as standard have Dual 10GbE networking, IPMI remote management, PCIe 5 storage support and a powerful 2700W PSU make Q-Power ready for demanding production environments.
Case Specification
- Size (WxDxH): 430x887x88.8mm
- Weight: 18 KG
- Power Supply: Redundant 2.7KW 100-240VAC
- PCIE Slots: Up to 8x PCIe x16
- Front USB: 2x USB3.0
Total Price
£11,488.70 EX. VAT
Estimated lead time
12 working days
Configure
* Return to Base warranty with Telephone support during normal business hours.
*1 Next Business Day Response. A fault must be diagnosed and replacement parts available before an engineer can attend site. Specific terms & conditions apply as set out in our Warranty Service Contract.
Can't Configure What You Need?
We can’t list all of the possible options, if you want something not shown in the configurator then click Custom Quote button. We’ll get in touch the same business day to discuss it.
Request a Call Back
Rental Quote
Request Custom Quote
Specifications
RAID: RAID 0, 1, 10 (NVMe/SATA)
DVD: No DVD drive bay
GFX Output: ASPEED AST2600 with DisplayPort Output(s)
GFX Card: Not Quoted - Options Available
LAN: 2x Intel X710 (10Gb/s)
WLAN: Not Supported
USB Ports Rear: 2x USB3.2, 4x USB3.2 Gen2, 2x USB-C Gen4
Serial Ports: 1 Optional Ports Available
Expansion Slot 1: PCIe x16 Gen5 (Full Height) 75cm (x16 Lanes, Gen 5.0)
Expansion Slot 2: PCIe x16 Gen5 (Full Height) 75cm (x16 Lanes, Gen 5.0)
Expansion Slot 3: PCIe x16 Gen5 (Full Height) 75cm (x16 Lanes, Gen 5.0)
Expansion Slot 4: PCIe x16 Gen5 (Full Height) 75cm (x16 Lanes, Gen 5.0)
Motherboard Remote Management: IPMI (Remote Management)
PSU: 2700W 100-240VAC Redundant Kit
Thunderbolt: Not Available on this Product
Audio Ports: 2x 3.5mm Ports (Stereo Out, Mic In) & Optical Digital
Motherboard I/O Ports:
FAQs
Does my data leave the building if I run AI on this?
No. Everything runs on your own network, so nothing gets sent to a third-party server to process. Response latency comes in under 40ms, against roughly 1.5 seconds for a cloud API call.
What happens if my internet goes down?
Nothing, because the server doesn't need it to keep working. There's no dependency on a cloud provider staying online. If the connection is lost, your rendering or inference jobs carry on regardless.
Is it cheaper than using cloud AI or render farms?
Yes to both. You own the hardware, so there's no per-seat render farm fee eating into every job. And measured against premium cloud AI APIs at consistent load, the Q-Power works out up to 18 times cheaper per million tokens over time.
Can I fine-tune a model on my own data with this?
Yes. You get full access to the model weights rather than renting access to someone else's, so you can fine-tune on your own data and own what you build.
Where's the 2U Q-Power built and is it tested before it ships?
Hampshire, UK. Every configuration runs Cinebench under sustained load before it leaves us. Each unit is tested to prove it can handle full load without errors or overheating.
I'm not sure which build I need.
Tell us what you're rendering or what you're running, and they'll tell you which build is most suitable.
Does my data leave the building if I run AI on this?
No. Everything runs on your own network, so nothing gets sent to a third-party server to process. Response latency comes in under 40ms, against roughly 1.5 seconds for a cloud API call.
What happens if my internet goes down?
Nothing, because the server doesn't need it to keep working. There's no dependency on a cloud provider staying online. If the connection is lost, your rendering or inference jobs carry on regardless.
Is it cheaper than using cloud AI or render farms?
Yes to both. You own the hardware, so there's no per-seat render farm fee eating into every job. And measured against premium cloud AI APIs at consistent load, the Q-Power works out up to 18 times cheaper per million tokens over time.
Can I fine-tune a model on my own data with this?
Yes. You get full access to the model weights rather than renting access to someone else's, so you can fine-tune on your own data and own what you build.
Where's the 2U Q-Power built and is it tested before it ships?
Hampshire, UK. Every configuration runs Cinebench under sustained load before it leaves us. Each unit is tested to prove it can handle full load without errors or overheating.
I'm not sure which build I need.
Tell us what you're rendering or what you're running, and they'll tell you which build is most suitable.
Rack Rail Installation Guide
Please see the attached PDF for Rack Rail installation instructions
Custom Serial Stickers
If you'd rather your units carried your own name than G2's, we produce a custom sticker in place of our standard
label. It carries your logo, your company name and address, your own product and model naming, and a
QR code, alongside the compliance marks. Please see the PDF guide for full details.
Download PDF