NVIDIA DGX Spark: what it's for and who it suits

Most teams building with AI hit the same wall. The models worth working with are too big for a normal workstation, and renting cloud GPUs gets expensive quickly, especially when your data cannot leave the building. The NVIDIA DGX Spark Founders Edition is NVIDIA's answer: an AI computer about the size of a hardcover book that sits on your desk and runs from a normal power point.

What's inside

• NVIDIA GB10 Grace Blackwell Superchip
• 128GB of unified memory, shared by the CPU and GPU
• Up to 1 petaFLOP of AI performance (FP4)
• 4TB of storage
• NVIDIA's AI software stack, installed and ready to use

What teams use it for

• Prototyping with large models. Run models of up to around 200 billion parameters locally, well beyond what fits on a typical graphics card.
• Fine-tuning on your own data. Adapt open models to your documents, products or field without uploading anything to the cloud.
• Building AI agents and apps. Develop and test before you deploy to bigger infrastructure. The software is the same stack NVIDIA uses in its data centre systems, so your work carries over.
• Keeping sensitive data in-house. It suits legal, health, finance and government work, where data privacy is not optional.
• Scaling when you need to. Link units together with a stacking cable to work with even larger models.

Who it suits

• AI developers and data scientists who want their own hardware instead of a queue or a cloud bill.
• Startups proving out a product before committing to servers.
• University labs and researchers who need real capability without a server room.
• Businesses exploring private, on-premises AI for their teams.

When something else fits better

If you already own a suitable workstation, want the fastest training on one machine, or also do graphics or simulation work, a workstation GPU may suit you better. See DGX Spark or a workstation GPU? To check which models fit in its memory, see Will it run?

Where to start

• One unit: the DGX Spark Founders Edition.
• Two units with a stacking cable: the DGX Spark bundle.
• Not sure yet: our Prototype on your desk solution, or book a free consultation.