NVIDIA's DGX Spark is getting a cheaper configuration. A 64GB unified memory version goes on sale Friday, Oct. 23 from Acer, ASUS, Dell, Gigabyte, HP and MSI, starting at $4,999, the company says. It keeps the GB10 Grace Blackwell Superchip, DGX OS and the full NVIDIA AI software stack found in the 128GB model, and NVIDIA says it can run models with up to 100 billion parameters entirely on the device.
The 64GB unit is sold only through those manufacturer partners, which NVIDIA frames as a way to hold the platform at a lower price while keeping the same software stack. Buyers get NVIDIA Agent Toolkit, CUDA-X AI libraries, Nemotron open models and runtimes including Ollama, vLLM and PyTorch with CUDA out of the box, according to the announcement. Blender is among the first major creator applications to support the platform, with a downloadable installer coming soon.
Read more: Nvidia Announces RTX Spark Processor Family for Laptops and Desktops at Computex 2026
Two 64GB units can be linked directly with a QSFP cable, pooling memory to 128GB and lifting model support to as much as 200 billion parameters, with NVIDIA claiming twice the memory bandwidth and up to 1.7x the performance of a single system. That figure comes from an internal test on the Qwen 3.8 27B model. Each unit ships with a ConnectX-7 NIC, and the NVIDIA Sync app's Cluster Assistant feature detects connected machines, validates their configuration and sets up the network so developers do not have to.
NVIDIA also says a Sync Model Launcher arrives at the end of the month, letting users download and start Qwen3.8 27B on one system or a cluster with a few clicks, and configure OpenCode to use it from a browser. The company lists round-the-clock coding or research agents, offloading inference from everyday laptops and desktops, and scaling to a second unit as the three main ways developers are expected to use the hardware.
The launch extends a local AI push NVIDIA has been building through the year. At IFA 2026 the company and Microsoft announced RTX Spark, a line of compact Windows PCs arriving in October from Lenovo and Acer, alongside PAIR, a free open-source beta that routes inference jobs across machines on a local network and supports DGX Spark. NVIDIA said then that its llama.cpp optimizations delivered as much as 1.9 times higher throughput on a GeForce RTX 5090. The 64GB Spark arrives as new Windows PCs powered by NVIDIA RTX Spark are also due this month from Acer, ASUS, Dell, HP, Lenovo, Microsoft and MSI.
What the announcement does not include is any benchmark comparing the 64GB configuration against the existing 128GB model beyond the two-unit clustering figures, nor any indication of whether the 128GB version's price changes. NVIDIA says playbooks for serving LLMs with vLLM, running OpenClaw with a local LLM and connecting multiple Sparks for distributed workloads are coming soon to 64GB devices.
