Personalized AI for everyone: Nvidia unveils revolutionary Project Digits

      At CES 2025, which kicked off in Las Vegas today, Nvidia unveiled a new personal supercomputer solution for artificial intelligence, a system called Project Digits. This compact supercomputer, very similar to a Mac Mini, runs from a standard wall outlet, is designed to make powerful AI computing available to a wide range of developers.Project Digits is capable of processing AI models with up to 200 billion parameters. For users with higher requirements, the option to combine the two systems is available, allowing it to handle models up to 405 billion parameters - equivalent to the size of today's best models such as Llama 3.1. The device includes 128GB of unified memory, far exceeding the capabilities of standard laptops, and up to 4TB of NVMe storage. "AI will become mainstream in every application for every industry. With Project Digits, the Grace Blackwell superchip will be available to millions of developers," said Nvidia CEO Jensen Huang. At the heart of the device is the GB10 Grace Blackwell superchip, it offers performance of up to 1 petaflop with FP4 accuracy to accelerate complex calculations through the use of approximate calculations. It features the latest generation of CUDA cores and fifth-generation tensor cores, which are connected via NVLink-C2C to a Grace processor with 20 power-efficient cores on Arm architecture. MediaTek also participated in the development of the chip, ensuring its energy efficiency and optimal performance. Project Digits users have access to Nvidia's extensive library of AI software: development kits, management tools and pre-trained models available through Nvidia's NGC catalog. The system runs on Nvidia's DGX OS, built on Linux, and supports popular AI tools such as PyTorch, Python and Jupyter. Developers can customize their models using Nvidia's NeMo framework and accelerate data analysis tasks using RAPIDS libraries. Project Digits provides the ability to develop and test AI models locally using Grace Blackwell's high-performance architecture and Nvidia AI Enterprise software. Once development is complete, the models can be deployed in the cloud or on a server infrastructure using the same software tools, making the system universal and suitable for tasks of any complexity. Project Digits costs from $3000. Read also VCS 2024: Rating of video conferencing systems IT-World conducted a survey of IT directors, experts of the IT Dialog club, about the quality of the most popular VCS systems according to several criteria. This approach is fundamentally different from the subjective view of other editors and is based solely on the practical experience of managers who use these systems in their daily work.