NVIDIA DGX Spark / GB10 desktop systems
Your Desktop
Agent Computer.
Designed to build and run autonomous agents. Bring large local models, powerful AI compute and the NVIDIA AI software stack to your desk.
The platform behind the possibilities
NVIDIA GB10 Grace Blackwell Superchip
Large local memory
Give your models room to work.
Coherent unified memory gives the CPU and GPU access to one memory pool for demanding local AI workloads.
NVIDIA ConnectX networking
Think beyond one system.
Connect compatible systems to support larger models and multi-agent workflows as your requirements grow.
NVIDIA AI software stack
Build on the tools behind AI.
Develop with NVIDIA tools, frameworks, libraries and models, with support for CUDA, NIM and familiar AI development workflows.
NVIDIA Sync
Your laptop. Your desktop AI.
Connect from your laptop or desktop to manage your DGX Spark, launch development tools and work with supported multi-system setups.
NVIDIA reference-platform maximums. FP4 performance is theoretical using sparsity. OEM configurations, software and included components vary; check the selected product.
Build. Run. Refine.
Accelerate All Your AI Workloads
Agentic AI
Build, run, and manage always-on AI agents from your desk.
Develop and run AI agents using compatible models, frameworks and tools. Choose orchestration and evaluation software for your workflow, then configure it on your system. Whether an agent runs entirely locally depends on the models, tools and external services you choose.
Prototyping
Develop, test, and validate AI models and applications.
With the NVIDIA AI software stack, DGX Spark provides a platform for developers to create, test, and validate AI models, and build AI agents, AI-augmented applications, and solutions. For final tuning or deployment, conveniently evaluate the work for eventual migration to the NVIDIA DGX™ cloud or other NVIDIA-accelerated data centers or cloud infrastructures.
Fine-Tuning
Fine-tune AI models up to 70 billion parameters.
Improve pre-trained model performance by fine-tuning on DGX Spark. With up to 128 GB of unified system memory, fine-tune models up to 70 billion parameters to customize AI models and solutions for specific needs and use cases.
Inference
Test, validate, and inference with AI models up to 200 billion parameters.
Fifth-generation Tensor Cores with support for FP4 deliver up to 1 petaFLOP of AI computing performance, combined with up to 128 GB of system memory, accelerating inference of state-of-the-art AI models to test, validate, and deploy from your DGX Spark.
Data Science
High-performance data science at your desk.
DGX Spark’s combination of up to 128 GB of unified memory and 1 petaFLOP of parallel throughput maximizes performance of large, computationally complex data analytics and machine learning workflows at your desk.
Edge Applications
Develop edge applications with NVIDIA AI frameworks, including NVIDIA Isaac™, Metropolis, and many others.
DGX Spark offers an exceptional platform for developing robotics, smart city, and computer vision solutions. NVIDIA frameworks include Isaac, Metropolis, and Holoscan, enabling developers to leverage the power of DGX Spark to develop edge applications quickly.
NVIDIA DGX Spark reference-platform capabilities. Model-size figures depend on the configuration, quantisation and workload; manufacturer-included software depends on the selected model. FP4 performance is theoretical using sparsity.
GB10 / room to grow
Scale to Larger Models, Faster Inference, and Agentic Workloads
Start with a desktop AI system. Build towards a connected setup as your models and workloads grow.
Make room for more.
Large local memory supports more demanding models. Connect compatible GB10 systems to extend capacity for larger supported models and projects.
Put Blackwell to work.
Use NVIDIA’s accelerated software stack for local inference. Speed depends on the model, precision, software and system configuration.
Build beyond the chatbot.
Develop local agents and tool-using applications, with a desktop platform for prototyping and running your workflows.
OEM configuration
1 × DGX Spark
64 GB
Up to 100Bmodel parameters
Single system
1 × DGX Spark
128 GB
Up to 200Bmodel parameters
Two-system configuration
2 × 64 GB systems
128 GB total
Up to 200Bmodel parameters
Two-system configuration
2 × 128 GB systems
256 GB total
Up to 400Bmodel parameters
Four-system configuration
4 × 128 GB systems
512 GB total
Up to 700Bmodel parameters
NVIDIA DGX Spark reference configurations. Model capacity depends on quantisation, context, software and workload; multi-system memory figures are aggregate totals. 64 GB configurations are offered through participating OEMs. Compatible networking/software are required, with an external switch for four systems.
GB10 systems / VRS Computing
Start at your desk.
Build from there.
Explore GB10 desktop systems built on the Grace Blackwell platform. Choose the model around your applications, memory needs, storage and support requirements.
Check the selected listing for the exact configuration and availability. Connected systems require suitable networking and customer configuration; extra licences may be needed.