Nvidia just released two powerful AI computers for home use. The DGX Station and DGX Spark let AI experts run big models right at their desks instead of always connecting to cloud services. These machines make local AI work much easier for people who need serious computing muscle.
The DGX Station comes ready for the toughest AI jobs you can throw at it. It packs the GB300 Grace Blackwell Ultra Desktop Superchip that cranks out 20 petaflops of AI power. You'll find 784GB of unified memory inside, making it breeze through giant datasets and complex models. The ConnectX-8 SuperNIC handles network speeds reaching 800 Gbit/s, perfect for heavy AI number-crunching.
Looking for something smaller but still mighty? The DGX Spark fits that bill exactly. This compact machine runs on the GB10 Grace Blackwell Superchip with fifth-generation Tensor cores supporting FP4 to make calculations faster. Nvidia built it with their NVLink-C2C system, which keeps the CPU talking smoothly to the GPU, cutting down on data processing delays.
The Spark carries 128GB of unified memory and works with up to 4TB of NVMe storage for quick data movement. It handles an impressive 1000 Tops of operations and supports many AI reasoning models, including those needed for robotics work. Nvidia first showed this machine at CES 2025 as Project Digits before giving it its final name. They created it specifically for AI developers, researchers, and students needing reliable local processing power.
The DGX Station comes ready for the toughest AI jobs you can throw at it. It packs the GB300 Grace Blackwell Ultra Desktop Superchip that cranks out 20 petaflops of AI power. You'll find 784GB of unified memory inside, making it breeze through giant datasets and complex models. The ConnectX-8 SuperNIC handles network speeds reaching 800 Gbit/s, perfect for heavy AI number-crunching.
Looking for something smaller but still mighty? The DGX Spark fits that bill exactly. This compact machine runs on the GB10 Grace Blackwell Superchip with fifth-generation Tensor cores supporting FP4 to make calculations faster. Nvidia built it with their NVLink-C2C system, which keeps the CPU talking smoothly to the GPU, cutting down on data processing delays.
The Spark carries 128GB of unified memory and works with up to 4TB of NVMe storage for quick data movement. It handles an impressive 1000 Tops of operations and supports many AI reasoning models, including those needed for robotics work. Nvidia first showed this machine at CES 2025 as Project Digits before giving it its final name. They created it specifically for AI developers, researchers, and students needing reliable local processing power.
