News
EveryChip
GB300 Grace Blackwell

GB300 Grace Blackwell Ultra Superchip

AI superchip combining Grace CPUs and Blackwell Ultra GPUs

The GB300 Grace Blackwell Ultra Superchip is a platform by NVIDIA that pairs Arm-based Grace central processing units with Blackwell Ultra graphics processors. It targets artificial intelligence reasoning and inference tasks in data centres, using liquid cooling and high-speed interconnects to boost throughput.

Share on WhatsApp

  • Integrates 72 Blackwell Ultra GPUs and 36 Grace CPUs
  • Uses liquid-cooled rack-scale architecture
  • Features Micron SOCAMM LPDDR5X memory modules
  • Delivers 50x AI factory output boost over Hopper
  • Manufactured by Wistron in Texas

What it is

The GB300 Grace Blackwell Ultra Superchip is a computing platform designed for artificial intelligence workloads. It integrates thirty-six Arm-based Grace central processing units and seventy-two Blackwell Ultra graphics processing units into a single system. NVIDIA states this architecture is built for AI reasoning and test-time scaling. The system uses a fully liquid-cooled rack design to manage heat and maintain performance during intensive calculations.

How it came about

NVIDIA developed the GB300 as part of its Grace Blackwell product line. The company positions it as a successor to earlier Hopper-based platforms for AI factories. Manufacturing support comes from Wistron, which opened a facility in Texas to produce these systems. This move represents a significant investment in American manufacturing for advanced AI hardware.

Inside the chip

The platform relies on fifth-generation NVLink technology to connect its many processors. This interconnect allows seamless communication between the GPUs. Each graphics unit features Tensor Cores with enhanced attention-layer acceleration. The system also includes ConnectX-8 SuperNICs for high-speed networking via InfiniBand or Ethernet. Memory subsystems use HBM3E for the GPUs and modular SOCAMM LPDDR5X for the CPUs.

How it was made

Wistron manufactures the GB300 systems at its plant in Texas. The graphics processors utilize High Bandwidth Memory 3E (HBM3E) to handle large data contexts. The central processing units employ Micron SOCAMM memory modules. These small form factor memory sticks allow for modular capacity upgrades without replacing the entire circuit board assembly.

Where it went

This superchip targets data centre environments focused on AI inference and training. NVIDIA claims it accelerates real-time video generation and large language model reasoning. The platform supports workflows that require high throughput and low latency. It is designed to integrate with NVIDIA Mission Control software for managing infrastructure and orchestrating workloads across the rack.

Versions and family

The primary configuration described is the GB300 NVL72, which houses seventy-two GPUs. A desktop version exists as the DGX Station, powered by the GB300 Grace Blackwell Ultra Desktop Superchip. This smaller variant allows developers to run similar workloads on a deskside unit before scaling to the full data centre rack.

Why it matters

NVIDIA states the GB300 offers a fifty-fold increase in AI factory output compared to Hopper-based systems. It aims to improve user responsiveness and throughput per megawatt. By enabling faster video generation and reasoning tasks, the platform seeks to reduce the time required for complex AI simulations and content creation.

Questions people ask about GB300 Grace Blackwell Ultra Superchip

What processors are inside the GB300?

The system contains thirty-six Arm-based NVIDIA Grace central processing units and seventy-two NVIDIA Blackwell Ultra graphics processing units. These components work together to handle AI reasoning and inference tasks.

How is the GB300 cooled?

The platform uses a fully liquid-cooled rack-scale architecture. This design helps manage the heat generated by the high density of processors and ensures stable performance during heavy AI workloads.

What type of memory does it use?

The Blackwell Ultra GPUs use HBM3E memory for high bandwidth. The Grace CPUs use Micron SOCAMM LPDDR5X memory modules, which are modular and can be replaced or upgraded independently of the main board.

Who manufactures the GB300 systems?

Wistron manufactures the NVIDIA Grace Blackwell Ultra systems. The company operates a production facility in Texas specifically for these advanced AI computing platforms.

Good reading and sources

    Where this page comes from

    There is no English Wikipedia article on GB300 Grace Blackwell Ultra Superchip yet, so the facts on this page come from the sources listed above: nvidia.com/en-us/data-center/gb300-nvl72/, servethehome.com/micron-socamm-memory-powers-next-gen-nvidia-grace-gb3. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryChip wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.

    It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.

    Photo credits on this page

    The small pictures above come from Wikimedia Commons under free licences. Some are cropped to fit.

    • Tensor Processing Unit: Norman P. Jouppi, George Kurian, Sheng Li, Peter Ma, Rahul Nagarajan,, CC BY 4.0, source

    The weekly digest

    Every Monday: the chip of the day, new pages on EveryChip, and the conferences and anniversaries of the week ahead.