Nvidia’s Generative AI Supercomputer: Complete Details with Specifications and Uses

Nvidia has revealed computers with a new GB10 Grace-Blackwell super chip and 128GB of memory. These machines are intended to provide artificial intelligence (AI) developers, researchers, and students with the capabilities they need to run huge models on the desktop.

The $3,000 system, which was built in partnership with MediaTek and was revealed at the annual CES super-event in Las Vegas today, is powered by an Arm-based Grace CPU and Blackwell GPU that, according to renderings shared by Nvidia, seem to live in a single system-on-chip (SoC). The device was given the codename Project Digits. A customized version of Ubuntu Linux that has been pre-configured to make the most of the hardware will be included in the box when it is shipped.

Despite this, the system is far more powerful than an artificial intelligence personal computer (AI PC) that is powered by CPUs from Intel, AMD, or Qualcomm. However, it will have a difficult time competing with a workstation that is equipped with Nvidia’s current top workstation card, the RTX 6000 Ada. The performance of the accelerator is 1.45 petaFLOPS for sparse FP/INT8 operations, which is almost three times the performance that we anticipate Project Digits to produce (500 teraFLOPS) at that level of accuracy.

There are 128 gigabytes of LPDDR5x memory that is feeding those flops. As stated by Allen Bourgoyne, who is the director of product marketing for enterprise platforms at Nvidia, the choice to equip the machine with this much RAM was deliberate in order to make the process of working with huge AI models more accessible.

The dimensions of Project Digits are comparable to those of an Intel NUC mini-PC in a general sense. Although Nvidia has not provided a comprehensive list of the GB10’s specifications, the company has said that the machine it powers is capable of delivering a complete petaFLOP of artificial intelligence capability. However, before you get too excited about the possibility of a compact form factor PC surpassing Nvidia’s A100 tensor core GPU, you should be aware that the performance of the machine was evaluated based on sparse 4-bit floating point workloads.

According to the specifications that we have seen, the GB10 is equipped with a Grace central processing unit (CPU) that has 20 cores and a graphics processing unit (GPU) that manages to achieve a performance that is forty times that of the twin Blackwell GPUs that are used in Nvidia’s GB200 AI server.

A new compact generative artificial intelligence supercomputer has been unveiled by NVIDIA. This supercomputer offers improved performance at a reduced price, and it also has upgraded software.

Specifications of the Nvidia AI Supercomputer: Project Digits

Everything that customers get with the Digits and Consumer PC supercomputers is detailed in the following paragraphs.

Efficient Use of Power

The power and cooling requirements of high-performance personal computers (PCs) are often higher, particularly while gaming or doing intense workloads. Although it has a high level of performance, the DIGITS is designed to be power-efficient and can take electricity from a conventional wall socket to power it.

Functioning System (OS)

Windows and macOS, which are not specifically designed for AI-related activities, are often used on consumer personal computers. DIGITS is a Linux-based operating system that is geared for artificial intelligence research and implementation.

Pricing

Consumer personal computers cost less than DIGITS, depending on the setup, but they do not have the specialized hardware and software that is required for artificial intelligence. With a price tag of $3,000, DIGITS is aimed at professionals and universities.

Performance for Artificial Intelligence Workloads

Consumer PCs have limited memory and processing power, which makes it difficult for them to handle huge AI models. As a result, they often need cloud-based solutions. AI models with up to 200 billion parameters may be executed locally using the DIGITS platform. It is suited for large-scale artificial intelligence activities since it can manage 405 billion parameters when two units are coupled together.

Integrated Ecosystem of Software

It is necessary to manually install artificial intelligence frameworks and tools on consumer personal computers, which may not be as optimized or integrated. The DIGITS package comes pre-installed with NVIDIA’s AI Enterprise software stack, which includes frameworks such as PyTorch and TensorFlow, as well as tools for artificial intelligence development such as NeMo and RAPIDS.

Structure and Portability of the Device

Consumer personal computers are often bigger, with desktops needing more room and laptops being portable but producing less power for artificial intelligence activities.

Compact, with a size comparable to that of a Mac Mini, the DIGITS is meant to be able to fit on a desk or even in a bag.

Specifications of the Hardware

Although powerful, graphics processing units (GPUs) for consumer computers like the RTX 5070 or RTX 5090 are not suited for AI-specific tasks. Generally speaking, it contains 8–32 GB of RAM and 1–2 TB of solid-state storage.

It provides artificial intelligence performance of one petaFLOP at FP4 accuracy, and it is suited for machine learning workloads. We have 128 gigabytes of unified LPDDR5X memory and 4 terabytes of NVMe storage for managing huge datasets and artificial intelligence models. High-speed data transport is enabled by the GB10 Grace Blackwell Superchip, which combines a Blackwell graphics processing unit (GPU) and a 20-core Grace central processing unit (CPU) with an NVLink-C2C interface.

Objectives and the Intended Readership

The consumer personal computer is designed to do general-purpose activities such as gaming, office work, and the consumption of multimedia. Not optimized for tasks related to artificial intelligence. For the purpose of prototyping, fine-tuning, and running big AI models locally, DIGITS was developed for artificial intelligence researchers, data scientists, and students. A specialist tool for artificial intelligence creation and inference tasks, it is.

Also know: Computer Runs Slowly? Here Are the Tips to Speed Up Windows PC

Featured Elements of the Project Digits:

• Project DIGITS will be made accessible beginning in May 2025 via NVIDIA and its partners, and it will include a price tag of $3,000.

• Designed for artificial intelligence (AI) researchers, data scientists, and students, this software gives users the ability to create and execute AI models on their desktop computers without having to depend on costly and power-hungry platforms.

• Users have the ability to develop and test models locally on Project DIGITS, and then deploy them in a seamless manner to either the NVIDIA DGX Cloud or the infrastructure housed in data centers.

• For the purpose of fine-tuning models, it supports NVIDIA NeMo, and for accelerating data science procedures, it supports RAPIDS.

• The NVIDIA AI Enterprise software stack, which includes frameworks such as PyTorch, Python, and Jupyter notebooks, comes pre-installed on the device.

• The design is compact, comparable to that of a Mac Mini, and it is fueled by a regular electrical socket.

• The presence of 128 gigabytes of unified memory and up to four terabytes of NVMe (Non-Volatile Memory Express storage) storage makes it possible to handle massive datasets and models in an effective manner.

• This configuration, in a nutshell, indicates that the system is equipped with a strong graphics processor (GPU) and a clever, multi-tasking central processor (CPU) that are able to interact with each other in an expedient manner. As a result, the whole system is highly effective for artificial intelligence activity.

• The NVLink-C2C interconnect technology is comparable to a super-fast highway that links the central processing unit (CPU) and the graphics processing unit (GPU). This technology enables the GPU and CPU to collaborate and transfer information at a considerably quicker rate than is possible with conventional connections.

Also know: How to Fix Windows Resource Protection Could Not Start The Repair Service

What the Future Holds: Summing Up

The 20-core Grace central processing unit, which acts as the “manager” of the system, is responsible for handling general duties and organizing them all. The term “20-core” refers to the fact that it is equipped with twenty distinct mini-processors (cores) that collaborate to create an extremely powerful system.

Blackwell graphics processing unit (GPU), which serves as the “brain” of the system and is particularly adept at handling visuals and sophisticated computations, particularly for artificial intelligence applications. This will help the designers and developers offer better results paired with generative AI.