GB10 Grace Blackwell Superchip
AI superchip for desktop workstations
The GB10 Grace Blackwell Superchip is a system on a chip designed by Nvidia. It powers the DGX Spark desktop platform, offering up to 128 GB of unified memory and high AI compute performance for local agent development.
- Delivers up to 1 petaFLOP of AI performance at FP4 precision
- Supports up to 128 GB of unified system memory
- Enables fine-tuning of models up to 70 billion parameters
- Features ConnectX-7 networking for multi-node clustering
What it is
The GB10 Grace Blackwell Superchip is a system on a chip created by Nvidia. It combines central processing and graphics capabilities into a single unit. This chip serves as the core engine for the DGX Spark, a compact desktop computer aimed at AI developers. It allows users to run large language models and autonomous agents locally without relying on cloud servers. The design focuses on providing high memory capacity and efficient computing power in a small form factor.
Inside the chip
The chip uses the Grace Blackwell architecture. It features fifth-generation Tensor Cores that support FP4 precision for AI tasks. A key design element is the unified system memory, which can reach up to 128 GB. This large memory pool helps in fine-tuning models with up to 70 billion parameters. The chip also integrates high-performance networking through ConnectX technology. This allows multiple systems to link together for larger workloads. The architecture supports running complex data analytics and machine learning workflows directly on a desk.
Where it went
This superchip powers the Nvidia DGX Spark platform. It is used by AI developers, researchers, and data scientists who need local compute resources. The chip enables the development and testing of autonomous agents and large language models. It supports edge applications in robotics and computer vision using frameworks like Isaac and Metropolis. Users can also connect up to four DGX Spark systems to create a small cluster for more demanding inference tasks.
Versions and family
Systems powered by the GB10 are available with different memory configurations. Options include 64 GB and 128 GB of unified memory. The 64 GB version is offered exclusively through specific original equipment manufacturer partners. These variants allow users to choose the right balance of cost and performance for their specific AI development needs.
Questions people ask about GB10 Grace Blackwell Superchip
What is the GB10 Grace Blackwell Superchip?
It is a system on a chip designed by Nvidia. It powers the DGX Spark desktop platform and provides high AI compute performance with large unified memory for local development.
How much memory does the GB10 support?
The chip supports up to 128 GB of unified system memory. There is also a 64 GB configuration available through certain OEM partners. This memory is shared across the system components.
What kind of AI tasks can it handle?
It can fine-tune models with up to 70 billion parameters. It also supports inference for models up to 200 billion parameters. The chip is designed for running autonomous agents and local AI workloads.
Good reading and sources
There is no English Wikipedia article on GB10 Grace Blackwell Superchip yet, so the facts on this page come from the sources listed above: nvidia.com/en-us/products/workstations/dgx-spark/, servethehome.com/nvidia-dgx-spark-review-the-gb10-machine-is-so-freaki. 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.