NVIDIA’s Jensen Huang announcing the DIGITS project at CES 2025
NVIDIA CEO Jensen Huang gave a 1 hour and 45 minute keynote just before the start of CES 2025. He made a lot of announcements, including the announcement of an AI computing machine called Project DIGITS, pictured above, which should be available in May, in partnership with MediaTek.
He also talked about AI computer graphics with the Blackwell GeForce RTX 50 models and AI for RTX AI computers. He also talked about their Cosmos Foundation model platform to accelerate physical AI development, generative physical AI with NVIDIA’s Omniverse, vehicle company partners for next-generation highly automated and autonomous vehicle fleets, and safety milestones of automotive and cybersecurity NVIDIA for AV development. In this article we will talk about storage and memory and their use in some of the products discussed in Jensen’s keynote.
The Blackwell RTX hardware powering the Blackwell GeForce RTX 50 desktop and laptop GPUs is shown below, with Jensen holding the desktop version. It has a memory bandwidth of 1.8 TB/s with up to 32 GB of VRAM (video random access memory for storing graphics data). Jensen noted that the G7 (GDDR7) memory was sourced from Micron, one of two calls for Micron memory products in his presentation (the other being the HBM memory used in the Grace Blackwell NVLink72 shown below).
NVIDIA RTX Blackwell
GeForce RTX 50 Series GPUs run creative AI generation models up to 2x faster in a smaller memory footprint compared to the previous generation product. GPUs support FP4, a quantization method to reduce image model sizes, somewhat similar to file compression. As a result, FP4 uses less than half the memory and twice the performance, and this is done with virtually no loss in image quality.
NVIDIA demonstrated real-time image generation of very high-resolution streaming video using this GPU during the presentation. This GPU also enables 3D and virtual reality (VR) video and AV1 mode with ultra-high quality and greater color depth than available in most consumer cameras.
Jensen showed an image of Grace Blackwell NVLink72. This was introduced by NVIDIA in March of 2024. As shown below, this big chip includes 72 Blackwell GPUs and 36 Grace CPUs. It supports 1.4 exaflops per second, EFLOPS of 4-bit Tensorflow floating point processing.
The 576 high-bandwidth memory chips connected to the GPUs provide about 14 TB of memory with a total bandwidth of 1.2 PB/s. The CPUs have up to 17 TB of LPDDR5X memory with up to 18.4 TB/s performance. These chiplets are mounted in racks that provide the power and connectivity to support massive AI training.
NVIDIA Grace Blackwell NVLink72
At the end of his talk, Jensen introduced NVIDIA Project DIGITS with a new Grace Blackwell Superchip in a package that provides an AI computing engine, as shown below and at the beginning of this article. The GB10 Grace Blackwell Superchip is a system-on-a-chip (SoC) and offers a petaflop of AI FP4 computing performance that can be used to create and run large AI models. The box shown also includes a 4TB SSD and 128GB of LPDDR5X memory.
The press release states that the GB10 Superchip features an NVIDIA Blackwell GPU with latest-generation CUDA cores and fifth-generation Tensor cores connected via NVLink-C2C chip-to-chip interconnect to a high-performance NVIDIA Grace CPU. which includes 20 energy-efficient cores built with the Arm architecture. MediaTek, a market leader in Arm-based SoC designs, collaborated on the design of the GB10, contributing to best-in-class efficiency, performance and power connectivity.
Contents of the NVIDIA DIGITS project box
NVIDIA’s Jensen Huang delivered the first CES 2025 keynote and announced a slew of enterprise as well as consumer GPU products. Memory and storage play an important role in AI training and completion, and his presentation showed how they enable modern AI solutions.
