A desktop system is aimed at local model and agent work
NVIDIA says DGX Spark systems with 64GB of unified memory are becoming available through Acer, ASUS, Dell, Gigabyte, HP and MSI on October 23, with an announced price of $4,999. The company describes the machine as a local platform for AI models and agent workflows.
Memory capacity can determine whether a model or workload fits on a device, but 64GB does not imply every large model runs quickly or without quantization. Actual throughput depends on model size, precision, software and the task.
Two systems can pool memory
NVIDIA’s announcement describes linking two Sparks through NVIDIA Sync to pool 128GB of memory for supported workloads. This targets builders who want to experiment locally with larger models or connect an agent environment to local tools.
The source video walks through the DGX Spark developer workflow, including local agent setup. It is a vendor demonstration, not a comparative benchmark against cloud services or workstation GPUs.
A purchase decision depends on the workflow
The practical appeal is a compact machine that keeps model experimentation close to local files and applications. The trade-off is its announced $4,999 price, plus the need to confirm that the exact model, context length and software stack fit the hardware.
Before choosing one, builders should compare memory needs, model speed, power consumption and the cost of cloud alternatives for their actual usage. The launch gives a clear hardware option; it does not prove that every agent task is better or cheaper locally.
Source published October 2, 2026. Coverage is based on the maker’s announcement and demonstration.