Nvidia 940-54242-0000-100 2 DGX Spark bundle +1 Connecting Cable,GB10 Grace Blackwell,128GB LPDDR5x,4TB NVMe SSD,WiFi 7
NvidiaNVIDIA DGX Spark · 2-Unit Cluster
Two Sparks.
One cluster.
Two NVIDIA DGX Spark systems, linked over ConnectX-7 with the included cable — 256 GB of combined unified memory to fine-tune and serve models a single unit can’t hold.
- 2× GB10 Grace Blackwell
- 256 GB combined memory
- 2× 1 PetaFLOP FP4
- ConnectX-7 linked
The silicon
GB10 Grace Blackwell Superchip
A 20-core Arm CPU and a Blackwell GPU share one pool of memory over NVLink-C2C — no PCIe bottleneck between them. It is the architecture NVIDIA ships in DGX systems, on your desk.
- 20-core Arm CPU
- Blackwell GPU
- NVLink-C2C
Capacity
Two nodes, one memory budget
256 GB combined
Two 128 GB pools linked over ConnectX-7 — fine-tune and serve models that don’t fit on a single unit, entirely on-prem.
Scales to four
ConnectX networking connects up to four DGX Spark systems — NVIDIA notes models up to 700B parameters across a four-unit cluster.
Who it's for
Built for teams doing real AI work
Owned, local compute for the people who were paying for it by the hour.
Developers & ML researchers
Prototype and fine-tune in the 30–70B range; trade per-run cloud cost for a fixed, owned asset.
Regulated industries
Healthcare, finance and government teams that need on-prem, air-gapped-capable processing.
Research & academia
Capex-budget compute for imaging, genomics, astrophysics and coursework — no recurring cloud bill.
Robotics & edge vision
A local target for NVIDIA Isaac, Metropolis and Holoscan, from sensor processing to assistive tech.
Specifications
Technical specifications
- Configuration
- 2× NVIDIA DGX Spark + connecting cable
- Superchip (each)
- NVIDIA GB10 Grace Blackwell
- Combined memory
- 256 GB LPDDR5X unified (128 GB × 2)
- Combined AI performance
- Up to 2 petaFLOPS (FP4)
- Interconnect
- NVIDIA ConnectX-7 — 200 Gbps QSFP direct link
- CPU (each)
- 20-core Arm (10× Cortex-X925 + 10× Cortex-A725)
- GPU (each)
- NVIDIA Blackwell — 5th-gen Tensor Cores, FP4
- Storage (each)
- 4 TB NVMe M.2
- Networking (each)
- 10 GbE · Wi-Fi 7
- Operating system
- NVIDIA DGX OS (Ubuntu Linux)
- Dimensions (each)
- 150 × 150 × 50.5 mm · 1.2 kg
Questions
Frequently asked questions
Who is the 2-unit DGX Spark cluster for?
Teams whose models or context windows outgrow a single 128 GB unit — pairing two DGX Spark systems for 256 GB of combined memory to fine-tune and serve larger models on-prem.
How are the two units connected?
Directly, over each unit’s built-in NVIDIA ConnectX-7 NIC at 200 Gbps, using the connecting cable included in this bundle — no switch required.
How much bigger a model can two units run?
Two units pool 256 GB of unified memory, so you can fine-tune and serve models beyond a single unit’s ~200B-parameter reach. NVIDIA notes ConnectX networking scales up to four units for models up to 700B parameters.
How does it compare to cloud GPUs on cost?
Cloud GPU sessions run roughly $2–$15 each and never stop. This turns that recurring opex into a fixed, owned asset — and keeps sensitive data on-prem.
What’s included in the bundle?
Two NVIDIA DGX Spark systems (GB10, 128 GB each, 4 TB NVMe, DGX OS) plus one connecting cable for the ConnectX-7 link. Ask us to confirm exact in-box contents.
What AI applications does it support?
Generative AI (fine-tuning and serving larger LLMs), computer vision, predictive analytics, and simulation and research — scaled across two nodes.