NVIDIA DGX Spark
GB10 Grace Blackwell
- VRAM
- 128 GB unified
- CPU
- 20-core Arm (10× Cortex-X925 + 10× Cortex-A725)
- Memory
- 128 GB LPDDR5x unified system memory
- Location
- United Kingdom
Best for: Real-time LLM serving, high-concurrency inference (vLLM, TGI, TensorRT-LLM), diffusion models, mixed AI + rendering/VDI.
Inference with workload-specific precision; larger models need quantization and memory validation.
Typical guidance; actual fit depends on context length, batch size and precision.
| Model size | Starting guidance |
|---|---|
| ≤7B | Inference starting point; reserve KV-cache headroom |
| 7–34B | Precision-dependent; larger models require quantization |
| 70B+ | 70B-class 4-bit only after memory validation; larger models need sizing |
Indicative guidance, not a fit or performance guarantee. Context length, batch size, precision, quantization, KV cache, framework overhead and training method affect memory needs. Validate your exact model and workload with an engineer.
€1,520 per month · EUR · fixed monthly server price.
Order via the HostingB2B client area.
Related
GB10 Grace Blackwell
2× V100S 32 GB
1× A100 40 GB HBM2
FAQ
Yes. Every server comes with full root (or administrator) access.
Ubuntu, Proxmox or a custom OS on the V100S, A100 and L40S servers. DGX Spark runs NVIDIA DGX OS.
Standard configurations are typically provisioned quickly. Custom and multi-GPU builds take longer — please confirm timelines with our sales team.
Yes. You can add more servers or ask us for a multi-GPU or custom build. Talk to an engineer about your scaling path.
24/7 support via live chat, tickets and MS Teams.
Each configuration has a fixed monthly price in EUR. Orders are placed through the HostingB2B client area. Confirm the applicable commercial terms with sales before ordering.
Fixed monthly server prices in EUR.
Order via the HostingB2B client area.
Dedicated GPUs from €326/mo
Pick a server and order online, or talk to a GPU engineer about multi-GPU and custom builds.
€1,520/mo
Order now