Nvidia unveils 64GB version of DGX Spark desktop AI system

Nvidia has introduced a 64GB version of DGX Spark, its desktop computer built for running AI models and agents without sending work to a cloud server. The company says the new configuration gives developers a single machine to start on, with the option to add more power later.

The 64GB DGX Spark will be sold by six hardware partners: Acer, ASUS, Dell, Gigabyte, HP and MSI. Nvidia did not say in its announcement when the machines go on sale or what they will cost.

The system runs on Nvidia’s GB10, a chip that combines a Grace CPU with a Blackwell GPU, the company’s current graphics architecture for AI work. Nvidia says the setup delivers up to 1 petaFLOP of performance, a measure equal to roughly one quadrillion calculations per second, aimed at tasks like running local models, fine-tuning them with new data, and operating AI agents on the desktop.

Nvidia is also pitching the 64GB unit as a building block rather than a ceiling. According to the company, two 64GB DGX Spark machines can be connected to form a cluster with 128GB of memory shared across both systems, along with added compute. That is meant to let a developer begin with one desktop and scale up without moving to a data center or a cloud subscription.

Why memory matters here

Memory capacity determines how large a model a desktop system can hold and run directly, rather than offloading it to a remote server. By offering a 64GB option alongside clustering, Nvidia is giving developers a way to size their setup to the model they are working with, then expand if that model or workload grows. The company frames this as useful for practitioners who want to keep data and computation on their own hardware instead of relying on cloud infrastructure.

Nvidia’s announcement, as reported by MarkTechPost, does not include independent benchmarks, pricing from the six partner companies, or a specific release date. Those details will determine how the 64GB DGX Spark compares with other ways developers already run AI models locally, including earlier DGX Spark configurations.

What happens next depends on the six hardware partners. Acer, ASUS, Dell, Gigabyte, HP and MSI will set their own pricing and ship dates for the 64GB DGX Spark, and those listings, along with any independent testing of the clustered 128GB setup, will show whether Nvidia’s performance claims hold up outside the company’s own announcement.