NVIDIA DGX Spark. Big ideas start here.
Rent UK-hosted AI compute in London Docklands, Wolverhampton or Manchester.
Compact outside. Capable inside.
Grace Blackwell performance. A compact system. Room for your next idea.
128 GB
Coherent memory shared by the CPU and GPU, giving larger models room to work.
20 cores
Grace Arm processing and NVIDIA Blackwell graphics in one GB10 superchip.
Up to 1 PFLOP
Theoretical FP4 performance using sparsity. Application results depend on model and software.
Built for your next AI project
Private assistants, useful agents and models built around your work. Choose a supported recipe for your model, precision and context length.
Discuss your AI projectPrivate LLMs
Prototype and run private assistants on UK-hosted NVIDIA compute.
AI agents and RAG
Build agents and connect them to your approved documents and tools.
Inference and testing
Develop and evaluate models up to 200B parameters with suitable precision and runtime.
Fine-tuning
Adapt models up to 70B parameters using supported fine-tuning workflows.
Your models. Your familiar tools.
Choose a documented DGX Spark recipe for your model, precision and workload.
Qwen 3.8 and more
Explore Qwen, Llama, gpt-oss, DeepSeek, Mistral, Gemma and NVIDIA Nemotron.
Work from your laptop
Connect with NVIDIA Sync or SSH. Launch VS Code or AI Workbench remotely, and use compatible Ollama, vLLM, SGLang or llama.cpp workflows.
More nodes. More possibilities.
Build a same-site cluster with Ray and suitable vLLM workflows. We’ll help plan the connectivity around your workload.
Coordinate workloads
Ray schedules tasks and workers across configured nodes.
Ray coordinates configured resources; it does not automatically pool GPU memory.
Serve larger models
Use suitable vLLM parallelism and model-serving recipes.
Connect within one site
Plan node count, interconnects and software together.
UK DGX Spark hosting
Three locations. Connected by Exascale.
Choose your UK base for DGX Spark rental. Customer-owned colocation is available in Wolverhampton.
RENTAL
London Docklands
DGX Spark rental in London’s connectivity hub.
RENTAL + COLOCATION
Wolverhampton
Rental and customer-owned colocation.
RENTAL
Manchester
DGX Spark rental for teams choosing a northern UK base.
Make it yours. Two ways to host.
A clear quotation sets out your hosting, connectivity and service terms.
Rent a DGX Spark
An Exascale-owned system, hosted for your AI project.
London Docklands · Wolverhampton · Manchester
Discuss rentalColocate your DGX Spark
Bring your own system. We provide its UK hosting environment.
Wolverhampton only
Discuss colocationDGX Spark hosting questions
What Is Data Centre Colocation?
Colocation places customer-owned equipment in a specialist data-centre facility. The provider supplies the environment, while responsibilities for hardware and management…
View What Is Data Centre Colocation?How to Choose a UK Colocation Location
The best colocation location is not automatically the closest or largest facility. It should fit the organisation’s network design, operational… How to Choose a UK Colocation Location
View How to Choose a UK Colocation LocationWhat Is IP Transit and Who Needs It?
IP transit is a service that connects an organisation’s network to the wider internet. It is commonly used by internet… What Is IP Transit and Who Needs It?
View What Is IP Transit and Who Needs It?Yes. Rent an Exascale-owned DGX Spark in London Docklands, Wolverhampton or Manchester. Customer-owned DGX Spark colocation is available in Wolverhampton only.
Yes. NVIDIA Sync and SSH support remote development on DGX Spark, with user access and network connections arranged for your deployment.
Yes. NVIDIA documents Qwen3.8-27B on DGX Spark, plus Llama, gpt-oss and Nemotron configurations. The selected checkpoint, precision, context length and inference engine determine memory requirements and application results.
Ray can coordinate distributed tasks and GPU workers across configured nodes. vLLM also supports Ray as a distributed executor for suitable model-serving configurations. For tightly coupled LLM inference, plan connected DGX Spark nodes at one location; the model recipe and interconnect define how work and model weights are distributed. Read the vLLM scaling guidance.
Tell us whether you want rental or colocation, your preferred location and the workloads you want to run. Your quotation sets out setup and monthly charges, connectivity, support and the contract term.
Let’s plan your UK DGX Spark deployment
Choose rental in London Docklands, Wolverhampton or Manchester, or bring your own DGX Spark to Wolverhampton. Tell us about your application, users and preferred timescale.
Your quotation will clearly set out the hardware arrangement, connectivity, support, setup and monthly charges, and contract term.
Explore our UK data centre services and IP transit.