Ce guide n'est disponible qu'en anglais pour le moment.
Dell AI servers and workstations for on-premises AI
Dell builds AI hardware at every size: a workstation under a desk, a 2U server with a couple of GPUs, eight-GPU NVIDIA and AMD servers, and whole liquid-cooled racks. Dell calls the full set the Dell AI Factory. This page explains the range in plain words, so you can see which part of it fits your work, and how we put it into service on your site.
Dernière vérification : 10 Nov 2026
Vous hésitez ? Demandez à notre conseiller IA
Tell it the models you run, how many people use them and what power and cooling your room has. Il ne propose que des machines que nous vendons, avec des prix et des délais de livraison réels.
Why run AI on your own premises
- Your data stays with you: prompts, documents and models never leave your building, which makes privacy and regulation simpler.
- Predictable cost: no per-token or per-hour fees. Once the hardware is paid for, heavy daily use costs only power.
- No queue: your team gets the GPUs whenever it needs them.
Dell's own studies argue that on-premises inference costs less over a few years than public cloud for steady workloads. That depends on how busy the hardware is, so we size it to your real usage before you buy.
The Dell range, from desk to data centre
| Where it runs | Dell systems | Good for |
|---|---|---|
| Under a desk | Dell Pro Max with GB10 and GB300, Dell Precision towers | One person or a small team: prototyping, private assistants, fine-tuning small models |
| An existing rack, 2U | PowerEdge R760xa, R770, R7725 | Adding inference to a server room you already have, next to your other workloads |
| A GPU server, 4U | PowerEdge XE7745 (AMD EPYC), XE7740 (Intel Xeon) | Up to eight standard PCIe GPUs, air cooled: the flexible choice for most companies |
| An eight-GPU platform | PowerEdge XE9780 / XE9780L (NVIDIA HGX B200, B300), XE9785 / XE9785L (AMD Instinct MI355X) | Training and serving large models |
| A whole rack | PowerEdge XE9712 (NVIDIA GB300 NVL72), XE8712 (GB200 NVL4) | The largest models and supercomputing-scale work |
| The edge | PowerEdge XR7620, XR5610, XR8000 sleds | Machine vision and inference on a factory floor or field site |
| Everything else | PowerEdge R, T and XR servers (R660 to R960, R6725, R7715, T160 to T560) | Virtualisation, databases and storage, with room for a few GPUs on many models |
NVIDIA or AMD?
Dell offers both, as two families:
- Dell AI Factory with NVIDIA: NVIDIA GPUs, networking and software (NVIDIA AI Enterprise) tested together. The widest software support and the safest choice if your team already uses CUDA.
- Dell AI Platform with AMD: AMD Instinct GPUs and EPYC processors on the open ROCm software stack, with no per-GPU software licence. Instinct GPUs carry more memory per GPU (up to 288 GB), so large models need fewer of them. PyTorch, vLLM and the main open models run on ROCm.
If you're not sure, tell the advisor which models and tools you use. Most open models now run well on both.
Air or liquid cooling
- Air cooled (XE9780, XE9785, XE7745, R-series): fits a normal server room with enough power and airflow. Check the room before ordering an eight-GPU server: each can draw more than 10 kW.
- Liquid cooled (the L models, XE9712, XE8712): more GPUs per rack and less fan noise, but the room needs liquid cooling and Dell's IR7000 or IR9000 racks.
We check power, cooling, floor loading and access routes before we quote a data-centre system.
How we deliver it
- Size it: we work out GPU memory, users and storage from your models and data, and pick the smallest Dell system that does the job.
- Quote it: Dell systems are configured to order, so we confirm the exact configuration, price and Dell lead time in writing.
- Build it: we install the operating system, GPU drivers, CUDA or ROCm and your AI software, then test it under load.
- Support it: our UK engineers are your first contact, with Dell's warranty behind the hardware.
Dell AI workstations
For one person or a small team, under the desk.
- CPU: Intel Xeon W3/W5/W7
- Form factor: Tower
- NVIDIA GPU options: NVIDIA RTX 6000, 5000, 4500, 4000 or 2000 Ada; RTX A2000/A1000; A800; T400
- CPU: AMD Ryzen Threadripper PRO 7000 WX
- Form factor: Tower
- NVIDIA GPU options: NVIDIA RTX 6000, 5000, 4500, 4000 or 2000 Ada; RTX A4000/A2000/A1000; A800; T400
- CPU: Intel Xeon W3/W5/W7
- Form factor: Tower
- NVIDIA GPU options: NVIDIA RTX 6000, 5000, 4500, 4000 or 2000 Ada; RTX A4000/A2000/A1000; A800; T400
- CPU: Intel Xeon Silver/Platinum
- Form factor: Rack
- NVIDIA GPU options: NVIDIA RTX 6000, 5000, 4500, 4000 or 2000 Ada; A800; T400
- CPU: Intel 14th Gen Core i3 – i9
- Form factor: Small form factor
- NVIDIA GPU options: NVIDIA RTX 4000 SFF Ada, RTX 2000 Ada, RTX A2000/A400, T1000, T400
Questions fréquentes
Can I buy a Dell server from you in the UK?
Yes. We specify the system with you, order it configured from Dell, install and test the AI software, and support it from the UK. You get one written quote covering the hardware and our work.
Does a Dell server come with Dell's warranty?
Yes. It carries Dell's hardware warranty, and we can add Dell ProSupport. Our engineers handle the first call and work with Dell on hardware faults.
What's the difference between the XE9780 and the XE9785?
Both are eight-GPU AI servers. The XE9780 uses NVIDIA HGX B200 or B300 GPUs with Intel Xeon processors; the XE9785 uses AMD Instinct MI355X GPUs with AMD EPYC processors. The L versions of each are liquid cooled.
Can I use AMD Instinct GPUs in a server I already have?
The MI350P is a standard PCIe card, so it can go in many existing servers if they have the power (up to 600 W per card), airflow and slots. Send us the model and we'll check it. The larger Instinct GPUs only come as eight-GPU platforms inside a server.
How long does a Dell AI server take to arrive?
It depends on the model and on GPU supply. Each product shows the lead time from our catalogue, and your quote confirms Dell's current lead time before you order.
Obtenez une suggestion adaptée à vos besoins
Tell it the models you run, how many people use them and what power and cooling your room has. Il ne propose que des machines que nous vendons, avec des prix et des délais de livraison réels.