Artificial intelligence has rapidly become one of the most talked-about things of the 21st century, managing to disrupt many industries ranging from healthcare, entertainment, finance, and even changing the way we interact with technology.
When I think about this, only one organization comes to mind: NVIDIA. Their investments in AI were earlier than those of other big names, but the returns they have generated since have been nothing short of remarkable, propelling them to eye-watering evaluations.
At the recent CES event, they unveiled a new AI supercomputer that fits on a desk and runs on powerful hardware with a Linux-powered operating system making it all happen.
Let's take a closer look. 😃
NVIDIA Project DIGITS:
Source: NVIDIA
Offering up 1 petaflop of FP4 AI compute performance, Project DIGITS is powered by a new SoC that was made in collaboration with MediaTek. Developers can use this to develop and run inference on AI models locally.
All of this is meant to be done on DGX OS, an Ubuntu 22.04-based Linux operating system that has been tailored by NVIDIA to include various optimizations, additional drivers, and diagnostic/monitoring tools.
Additionally, developers will have access to a comprehensive suite of tools and resources for AI development, including SDKs, orchestration tools, AI frameworks like PyTorch, and a wide range of pre-trained models and other resources from the NVIDIA NGC Catalog.
Tools like the NVIDIA NeMo framework and RAPIDS libraries are also available. These utilities, alongside others, help developers deploy their work to cloud and data center infrastructure.
Source: NVIDIA
That was all about the software side of things, now let's dive into the juicy hardware aspect of it. The key specs of Project DIGITS include:
SoC: NVIDIA GB10 Grace Blackwell Superchip.
CPU: NVIDIA Grace CPU with 20 power-efficient Arm-based cores.
GPU: NVIDIA Blackwell with latest-gen CUDA cores and fifth-gen Tensor cores.
RAM: 128 GB of low-power DDR5X, which is used by both the CPU and GPU.
Storage: Up to 4 TB NVMe.
Connectivity: Wi-Fi, Bluetooth, and USB; not sure on the specifics yet.
Performance: Capable of running 200-billion-parameter LLMs. When linked with another Project DIGITS via NVIDIA ConnectX, this can scale up to 405-billion-parameter models.
The press release has more details for those interested in learning more.
If you are enthusiastic about getting this, then you will have to wait a bit as Project DIGITS is scheduled to be shipped sometime in May 2025 with a price tag of $3,000. Not sure how many countries will see the availability at launch time.
You can sign up on NVIDIA's official website to receive updates when the product becomes available through their authorized partners.
Abhishek Prakash
