Source URL: https://www.theregister.com/2025/01/07/nvidia_project_digits_mini_pc/
Source: The Register
Title: Nvidia shrinks Grace-Blackwell Superchip to power $3K mini PC
Feedly Summary: Tuned for running chunky models on the desktop with 128GB of RAM, custom Ubuntu
CES Nvidia has announced a desktop computer powered by a new GB10 Grace-Blackwell superchip and equipped with 128GB of memory to give AI developers, researchers, and students the tools they need to run large models on the desktop.…
AI Summary and Description: Yes
Summary: Nvidia’s newly announced Project Digits, a desktop powered by the GB10 Grace-Blackwell superchip, aims to equip AI developers and researchers with robust capabilities to run large AI models. With features including a powerful CPU/GPU combination and high memory capacity, it positions itself as a strong alternative for AI workloads at a competitive price.
Detailed Description: The unveiling of Nvidia’s Project Digits at CES marks a significant development in desktop computing tailored for AI applications. This desktop computer is poised to provide researchers and developers with potent tools for running extensive AI models efficiently. Here are the major points of significance:
– **Powerful Hardware**:
– Powered by the new GB10 Grace-Blackwell superchip.
– Features a 20-core Arm-based Grace CPU and Blackwell GPU.
– Supports running models with up to 200 billion parameters when compressed to 4-bits; potential for 405 billion parameters when two units are connected.
– **Memory and Performance**:
– Equipped with 128GB of LPDDR5x memory for enhanced performance with large AI models.
– Claimed to deliver a full petaFLOP of AI performance, although performance metrics largely depend on the 4-bit floating point workloads.
– Expected memory bandwidth might approach 825GB/s, providing substantial bandwidth for processing large datasets.
– **Storage and Connectivity**:
– Contains 4TB of NVMe storage, sufficient for many AI models, particularly those downscaled to 4-bits.
– Onboard ConnectX networking enables multiple systems to work together for larger-scale AI tasks.
– **Market Positioning**:
– With a price tag starting at $3,000, it is marketed as a competitive choice against existing high-performance desktop systems powered by Nvidia’s top GPUs.
– While it resembles Nvidia’s previous Jetson dev kits, Project Digits aims to ease users into utilizing Nvidia’s stronger Grace-Blackwell superchips.
– **Applications**:
– Nvidia anticipates that this machine will support a range of tasks including model experimentation, fine-tuning, data science, and edge applications.
This development is highly relevant for professionals working in AI, cloud infrastructure, and security, as it indicates a strengthening of desktop capabilities to support complex AI tasks, making these technologies more accessible and manageable in various operational environments.