Nvidia has rebranded Project Digits as the ‘DGX Spark’ mini AI supercomputer, launching it for reservations starting at $3,000 at GTC 2025. This compact computer, similar in size to a Mac Mini, offers up to 1 petaflop (1,000 TOPS) of AI performance at FP4 precision, powered by the Grace Blackwell GPU with 5th-gen Tensor cores.
The DGX Spark is equipped with 20 Arm CPU cores, and it can be configured with up to 128GB of LPDDR5X unified memory and 4TB of NVMe storage, although it has a limited memory bandwidth of 273 GBps. It supports Wi-Fi 7, Bluetooth 5.3, and consumes up to 170W of power. The device runs on Nvidia’s Linux-based DGX OS.
Additionally, Nvidia announced the larger DGX Station, featuring the new Blackwell Ultra GB300 GPU, aimed at developers, researchers, data scientists, and students for running and training large AI models. This workstation promises “data-center-level performance” in a desktop format.
The DGX Station boasts 784GB of unified memory, with the GPU accessing 288GB of HBM3e memory at 8TBps bandwidth, and the CPU utilizing 496GB of LPDDR5X memory with up to 396GBps bandwidth, powered by the server-class Grace-72 Core Neoverse V2 CPU.
Nvidia is clearly positioning itself against Apple’s recently launched M4 Max and M3 Ultra Mac Studios. Reservations for the DGX Spark can be made on Nvidia’s website, with shipping set to begin this summer. The DGX Spark will also be available through Asus, Dell, HP, and Lenovo, while the DGX Station will launch later this year.
