NVIDIA adds a 64GB DGX Spark for local AI developers
The new lower-capacity model ships Oct. 23 and gives teams a cheaper entry point while still allowing two units to be clustered for larger models.
NVIDIA is adding a 64GB DGX Spark configuration for local AI developers, extending the same Grace Blackwell-based stack to a lower-capacity model that can still run agents and other workloads on-device. The system will ship through Acer, ASUS, Dell, Gigabyte, HP and MSI on Friday, Oct. 23, with pricing starting at $4,999. NVIDIA says two units can be linked through Sync Cluster Assistant to pool memory to 128GB, support models up to 200 billion parameters, and reach up to 1.7x the performance of a single machine in its Qwen 3.8 27B test. The box is meant to arrive ready for common local-AI workflows with DGX OS, CUDA-accelerated software, NVIDIA Agent Toolkit, and runtimes such as Ollama, vLLM and PyTorch already supported.
Why it matters
The 64GB model broadens access to NVIDIA’s local AI box without changing the basic setup. For developers, it means a single machine can handle smaller on-device workloads, while two units can be linked when memory or model size needs exceed one system. That gives teams a clearer path from individual local work to larger clustered runs using the same hardware family.
Keep or strike?
Does this story matter, or is it hype? Mark it before you see what everyone else did.
Sources
- NVIDIA Blog