NVIDIA DGX Spark
NVIDIA · High-end
A small desktop AI computer built on NVIDIA's GB10 Grace Blackwell chip, with 128GB of unified memory. It brings NVIDIA's software stack to a compact machine aimed at running and prototyping large models locally.
Specifications
| Unified memory | 128GB |
|---|---|
| Memory bandwidth | 273 GB/s |
| Power draw | 240W |
| Type | complete machine |
A complete machine at a price well above a mini PC but below a multi-GPU server. Check current availability and pricing, as supply has been limited.
NVIDIA DGX Spark: common questions
- What models can the NVIDIA DGX Spark run?
- With 128GB of unified memory it can run models up to roughly 210B parameters at a 4-bit quantisation, or smaller models with more context. Use the hardware matrix for specifics; these figures are approximate.
- How much power does the NVIDIA DGX Spark draw?
- About 240W under load, so pair it with a power supply that has real headroom.
Strengths
- 128GB unified memory for large models, with NVIDIA's software stack
- Compact and far lower power than a multi-GPU server
- CUDA compatibility, easing the move to and from data-centre GPUs
Weaknesses
- Memory bandwidth is modest, so generation is slower than a discrete GPU
- Priced above unified-memory mini PCs offering similar memory
- Aimed at developers and prototyping more than raw serving throughput
Entry last verified 30 July 2026. Specifications and especially prices change; verify before buying.