NVIDIA DGX Spark Specs: Full GB10 Specifications Explained (and vs RTX Spark)

Full specification table

DGX Spark alongside the two announced RTX Spark N1X configurations.

What the specs mean in practice

The 128 GB of memory is the headline: it lets one small box load models that would need several consumer GPUs. The trade-off is bandwidth — about 273 GB/s is well below high-end discrete GPUs — so large dense models generate tokens slowly while MoE models and prompt processing shine.

• Capacity: load 70B dense or ~120B MoE models with room for context.
• Compute: strong prompt processing and fine-tuning of small models.
• Bandwidth: the main limit on tokens per second for big dense models.

Ports and connectivity

The Founders Edition includes USB-C ports (one for power), HDMI, 10 GbE Ethernet, Wi-Fi 7 and a ConnectX-7 NIC for high-speed linking. Partner units may vary slightly — check the specific model page.

Power, size and noise

DGX Spark is a palm-sized desktop (about 150 × 150 × 50 mm) powered by an external adapter of around 240 W. It is designed to sit on a desk; sustained noise depends on the chassis design, which is one reason partner units differ.

DGX Spark vs RTX Spark: the key differences

Both use Grace CPU and Blackwell GPU technology with unified memory. DGX Spark is Linux, 128 GB only, and built for AI developers. RTX Spark runs Windows on Arm, comes in laptops and desktops from 24 GB to 128 GB, and adds RTX graphics features for games and creative apps.

Common spec misconceptions

A few claims show up repeatedly in forums and are worth correcting.

• '1 petaflop' is a sparse FP4 peak — real workloads see a fraction of that.
• More memory does not make small models faster; bandwidth does.
• DGX Spark is not a gaming PC and does not ship with Windows.

How RTXsparks sources and checks these numbers

RTXsparks is independent and not affiliated with NVIDIA or any OEM. Prices come from manufacturer launch announcements and retailer listings (US MSRP before tax unless stated). Specs come from official product pages. Performance figures are either our own lab runs, labelled community submissions, or clearly marked estimates derived from memory bandwidth. Anything a vendor has not publicly confirmed is labelled 'unconfirmed'. See the methodology page for the full SCS, ARS and TLS scoring rules.

• Prices re-checked weekly during the launch window.
• Estimates are always labelled as estimates — never presented as measured results.
• Corrections are welcome via the benchmark submission page.

Connectivity at a glance

DGX Spark is built to sit on a desk or in a small rack and be reached over the network. Key connectivity is listed below; exact port counts can differ on partner models.

• ConnectX-7 high-speed networking for linking two units.
• USB-C ports for peripherals and external storage.
• Ethernet and Wi-Fi for remote access.
• HDMI for a local display.

Specs that matter most for AI

Three numbers decide how DGX Spark feels for local AI: 128 GB memory (which models fit), about 273 GB/s bandwidth (how fast text is generated), and the Blackwell GPU's FP4 support (how fast prompts are processed). Everything else is secondary for most AI work.