NVIDIA GB10 Grace Blackwell Powers New Wave of Miniature Supercomputers
Berlin, Germany – NVIDIA is ushering in a new era of accessible AI computing with the GB10 Grace Blackwell processor, featured in the Acer Veriton GN100 AI Mini Workstation unveiled at IFA. The compact system delivers up to 1 PFLOPS FP4 performance for artificial intelligence tasks.
The Veriton GN100 pairs the GB10 with 20 ARM processor cores, 128 GB of unified system memory, and 4TB of SSD storage. Utilizing the NVIDIA ConnectX-7 network card, two units can be linked to support AI models with up to 405 billion parameters.
Acer’s workstation includes NVIDIA’s AI software package,offering compatibility with popular frameworks like Pytorch,Jupyter,and Ollama for local development,fine-tuning,and deployment of large language models,with scalability to cloud infrastructure.
NVIDIA is employing a strategy similar to its graphics card approach, providing manufacturers with a motherboard reference design – allowing for customization with cooling solutions – or the option to create entirely new designs.
The entire system is powered via USB-C, with a maximum consumption of 200W (140W for the processor itself). The universal CX-7 interface ensures compatibility between manufacturers, meaning systems from Acer and Asus will share core components.
The Acer Veriton GN100 AI Mini Workstation will be available for 4,000 euros. NVIDIA believes this approach, combined with established distribution and service networks, will make powerful AI computing more attainable than previous solutions like the NVIDIA DGX Spark.
