AI hardware
AI servers, GPUs and workstations
From one professional GPU to multi-GPU rack servers, with the CPUs, boards and ECC memory around them.
Where to start
Three ways into AI hardware
We stock 198 GPUs suited to AI work, 283 servers and server barebones, and 346 server and workstation CPUs. Most projects start in one of three places.
1
Upgrade a workstation
Add a professional or high-memory GPU to a tower you already own.
Suits: prototyping, fine-tuning smaller models, local inference.
Compare GPUs2
Buy a GPU server
2U and 4U rack servers designed to hold several GPUs, with the airflow and power they need.
Suits: shared team use, training, inference services.
See GPU servers3
Build your own platform
Choose the CPU, server board and ECC memory, then add GPUs and storage.
Suits: integrators and teams with specific requirements.
Pick a platformGPUs for AI, side by side
Professional and data centre cards from NVIDIA and AMD, plus 32 GB GeForce RTX 5090 cards for development machines.
Memory is shown where the product name states it; otherwise see the product page. Server editions are passively cooled and need a chassis with suitable airflow.
Rack servers
GPU servers and AI systems
Barebone and configured systems built around multiple GPUs. Barebones need CPUs, memory and storage adding.
Gigabyte
Gigabyte G292-Z46 rev. A00 2U GPU Server Dual Socket AMD EPYC
SKU 6NG292Z46MR-00
£4,227.41View
Platform
CPUs, boards and ECC memory
Current AMD EPYC, AMD Threadripper and Intel Xeon processors, with server boards and registered DDR5 to match.
Processors
Server and workstation boards
ECC DDR5 memory
Ready-built workstations
Towers and compact workstations with professional graphics, ready to deploy.
Questions we often get
Do I need a professional GPU, or will a GeForce card do?
GeForce cards are a cost-effective way to develop and test models on a single workstation. For servers that run around the clock, or several GPUs in one chassis, professional and data centre GPUs are built for that duty, and server editions are passively cooled to suit rack airflow.
How much GPU memory do I need?
It depends on the models you plan to run. The model and its working data need to fit in GPU memory unless you split the work across several GPUs, so larger models need more memory. Check the requirements published for the models you use, and allow room to grow.
Can you quote a complete server with GPUs, memory and storage?
Yes. Send us your requirements or a bill of materials and we will quote the full build, from the server and GPUs to memory, storage and networking.
Can we buy on account?
Business customers can request quotes and proforma invoices, and apply for a trade account. Contact our sales team to set this up.
Planning an AI build?
Send us the models you want to run or a parts list and we will quote the complete system.
Prices are as listed on networkdistri.co.uk in September 2026 and can change; the product page always shows the current price, VAT and delivery. Product names and trademarks belong to their owners.
