NVIDIA has unveiled the DGX Station for Windows, a desktop AI supercomputer with capabilities approaching those of data centers. This system is designed to run and work with large AI models locally, including those with up to 1 trillion parameters. This is reported by Ixbt.com reports .
The platform is based on the NVIDIA GB300 Grace Blackwell Ultra architecture. The configuration includes a 72-core Grace processor, Blackwell Ultra graphics, and up to 748 GB of total memory. The system's performance reaches 20 petaflops in FP4 format, making it suitable for training complex AI models.
If necessary, it is possible to install an additional NVIDIA RTX PRO 6000 Blackwell graphics card. The manufacturer states that the DGX Station is capable of serving multiple AI agents simultaneously and analyzing large volumes of data locally without relying on cloud services.
The device is primarily aimed at developers and research teams. Sales are expected to begin in the fourth quarter of 2026. Production will be handled by partner companies such as Asus, Dell, Gigabyte, HP, MSI, and Supermicro.
Read “Zamin” on Telegram!
