NVIDIA DGX Spark™: designed to build and run autonomous agents.
NVIDIA DGX Spark™, with the NVIDIA Agent Toolkit and OpenShell, provides the ideal platform for autonomous AI agent development and deployment, providing the compute, large GPU memory, and power efficiency required for always-on agents.
The NVIDIA DGX Spark™ belongs to a new class computers designed to build and run AI.
Powered by the NVIDIA GB10 Grace Blackwell Superchip, DGX Spark delivers up to 1 petaflop of AI performance to power workloads for long-running agents that generate large numbers of concurrent tokens for communication between models and agents. With 128 GB of coherent unified system memory, autonomous agents can run larger models of up to 200 billion parametersii locally and bring up additional models as required, helping to minimize per-token cloud costs using external models. Built-in NVIDIA ConnectX™ networking can connect up to four DGX Spark systems to run larger, more complex agents and models.
Main features of the NVIDIA DGX Spark
Based on NVIDIA Grace Blackwell architecture
New GB10 Grace Blackwell Superchip, based on the Grace
Blackwell architecture and optimized for a desktop form factor.
128 GB of Coherent Unified System memory
Run large AI models of up to 200 billion parameters locally—and bring up additional models as needed, without the latency or cost of offloading to the cloud.
ConnectX-7 NIC
Connect up to four DGX Spark systems to scale seamlessly to larger, more complex agents
and models, turning individual units into a powerful compute cluster.
NVIDIA AI software stack
NVIDIA AI Enterprise for DGX Spark
Provides software tools, libraries, and frameworks validated on DGX Spark, with enterprise-class support during business hours and optional 24/7 support. Available on DGX Spark and OEM GB10 systems that have completed NVIDIA-Certified Systems™ testing.
