NVIDIA has unveiled two supercomputers: the compact DGX Spark and the powerful DGX Station for running large-scale AI models.
NVIDIA has unveiled two new supercomputers for the development and broad application of artificial intelligence - DGX Spark and DGX Station.
Here's What We Know
These are mini-computers that NVIDIA CEO and co-founder Jensen Huang said will "define the future of computing." He believes that AI agents will become an integral part of human life in the near future and will require powerful computers to operate them.
TheDGX Spark is the world's smallest AI supercomputer, ideal for researchers, developers and students, as well as creators of various AI models. It features a GB10 Grace Blackwell superchip that delivers up to 1 quadrillion operations per second and performance of up to 1,000 teraflops (or 1 petaflops). The device will get 128GB of unified system memory.
TheDGX Station is designed for training and developing large-scale AI models. NVIDIA says the computer uses the GB300 Grace Blackwell Ultra superchip and will get 784GB of unified memory. It supports data transfer rates of up to 800Gbps thanks to ConnectX-8 SuperNIC technology. Its performance data in teraflops has not been specified yet.
Both systems allow easy switching between on-premises and cloud networks.
DGX Spark is available for order now, while DGX Station will be available later in 2025.
Source: NVIDIA
