NVIDIA Unveils $4,999 DGX Spark 64GB Desktop AI Supercomputer for Local Inference
NVIDIA announces the 64GB DGX Spark desktop supercomputer powered by GB10 Grace Blackwell silicon, starting at $4,999 with 200GbE QSFP clustering support.
NVIDIA has unveiled a 64GB unified memory configuration of the DGX Spark desktop AI supercomputer, launching October 23, 2026, starting at $4,999 across six manufacturing partners. Powered by the GB10 Grace Blackwell Superchip, the compact workstation runs models up to 100 billion parameters locally and supports 200GbE QSFP clustering for combined 128GB workloads.
GB10 Grace Blackwell Superchip and Unified Memory Architecture
On October 20, 2026, details emerged regarding NVIDIAs new 64GB DGX Spark workstation configuration, built in partnership with Acer, ASUS, Dell, Gigabyte, HP, and MSI. Built around the GB10 Grace Blackwell Superchip, the system combines a 20-core Arm CPU, Blackwell-architecture GPU, and 273 GB/s of shared memory bandwidth. The unified memory architecture allows seamless tensor execution for models up to 100 billion parameters—such as Qwen3.8-27B, Meta Muse Glimmer, and Nemotron 3.5 Lightning—directly on-device without per-token cloud inference charges.
Dual-Node QSFP Clustering via NVIDIA Sync
The DGX Spark 64GB incorporates ConnectX-7 high-speed networking, enabling two standalone systems to interconnect directly via a QSFP cable over a 200GbE link. This setup pools memory into an effective 128GB unified workspace capable of running 200-billion-parameter models with up to a 1.7x performance gain over single-system execution. The integrated NVIDIA Sync Cluster Assistant automatically handles node discovery, network routing, and distributed memory fabric configuration without manual switch management.
Enterprise Software Readiness and Developer Desktop Deployment
Shipping with DGX OS, the 64GB desktop workstation includes preconfigured access to CUDA-X AI libraries, TensorRT, Agent Toolkit, and native acceleration for vLLM, Ollama, and PyTorch. An upcoming Model Launcher utility scheduled for late October will streamline browser-based coding through OpenCode, positioning the $4,999 workstation as an accessible local execution environment for researchers developing continuous coding agents, vision models, and proprietary fine-tuning pipelines.