DGX Spark

AI workstation, GB10 superchip

Short answer. The DGX Spark is a compact NVIDIA artificial intelligence workstation built around the GB10 superchip. It runs models of up to 200 billion quantised parameters locally on a single node; NVIDIA claims up to 405 billion by linking two units and up to 700 billion with four. It puts a private model on the desk of a very small business, without cloud, with llama.cpp or vLLM.

Hardware sheet · QDNA sovereign AI platform · Published on 30 July 2026 · Updated on 13 October 2026

Indicative price	€7,000 excl. VAT, the price used from October 2026 after the observed increase (LDLC.pro reading of 26 September 2026: €5,083 excl. VAT, PNY edition, 15-day delivery), i.e. €8,400 incl. VAT at 20% VAT; €5,290 excl. VAT at Nelpx; NVIDIA list price of $4,699 since February 2026, up from $3,999 at launch (Tom’s Hardware, price guide)
Memory	128 GB unified LPDDR5X, 273 GB/s
Compute	1 PFLOP FP4 with sparsity (1,000 TOPS)
Chip	GB10 Grace Blackwell, 20 Arm cores
Network	ConnectX-7 NIC at 200 Gb/s (two QSFP ports), up to 4 units in a cluster according to NVIDIA
Power draw	240 W power supply, 140 W GB10 chip TDP (NVIDIA datasheet), wall outlet; 240 W is the upper-bound assumption used in the site’s calculations, ServeTheHome measuring 60 to 200 W under load
Form factor	150 mm desktop

What the machine runs

Models up to 200 billion parameters quantised on a single node. With two nodes directly connected by a QSFP cable, NVIDIA claims up to 405 billion parameters (Project DIGITS announcement, January 2025); beyond that, the ConnectX-7 NICs go through an Ethernet switch, up to four units and 700 billion parameters according to the NVIDIA datasheet re-read on 2 September 2026. Recommended runtime: llama.cpp or vLLM locally.

Positioning

DGX Spark: NVIDIA's compact AI workstation on the GB10 superchip, 128 GB of unified memory, quantised models up to 200 billion parameters on one node.

Manufacturers

NVIDIA offers the DGX Spark through its partners, including Acer, ASUS, Dell, Gigabyte, HP, Lenovo and MSI.

GPUDirect Storage

On platforms fitted with ConnectX cards, GPUDirect Storage technology establishes a direct path between NVMe or NVMe over Fabric storage and GPU memory, bypassing the CPU buffer. Compatible storage arrays, such as NetApp, VAST, DDN or WEKA, feed training and large-scale RAG at full speed.

Support

Open stack, maintained by its communities. Optional NVIDIA AI Enterprise (NIM, NeMo, Triton) with a service-level agreement, licensed per GPU.

Which models fit on DGX Spark?

The machine offers 128 GB of memory. 1 of the 7 open models in the catalogue load on it, in the format shown.

Model	Billion parameters	Most precise format that fits	Sizing page
Qwen 3.8 27B	27	FP16	Qwen 3.8 27B on DGX Spark

See the full sizing matrix.

Frequently asked questions

How much does a configured dgx spark cost for an LLM?

Pricing depends on configuration and supplier. The page shows a dated public price; a scoping call helps compare supplier quotes.

Which LLM fits in a dgx spark?

Depends on the quantization format and the input context. The table on the page indicates the recommended maximum model size.

Which runtimes support the dgx spark?

vLLM, llama.cpp, Triton Inference Server, depending on the chosen framework. The model, format and GPU combinations measured by QDNA are published in the measurements section.