NVIDIA Launches Blackwell Ultra GPU to Power Next-Generation AI Factories

Details

• NVIDIA has introduced the Blackwell Ultra GPU, the latest addition to the Blackwell architecture family, designed for large-scale AI workloads.
• The chip features a dual-reticle design with 208 billion transistors, offering 2.6 times more transistors than the prior Hopper lineup.
• It includes 288 GB of HBM3e memory, fifth-generation Tensor Cores, and the NVFP4 precision format, enabling significant leaps in processing power and memory efficiency.
• Blackwell Ultra delivers 15 PetaFLOPS of dense NVFP4 compute, a 1.5x increase in compute over the base Blackwell GPU, and doubles the speed of attention-layer operations for complex AI reasoning tasks.
• Engineered for deployment in AI factories and cloud data centers, this chip sets a new standard for supporting large-scale AI services and next-generation machine learning models.

Impact

The Blackwell Ultra GPU marks a major advancement in AI hardware, positioning NVIDIA ahead of rivals in both speed and efficiency for demanding generative and reasoning workloads. As AI services grow rapidly, this leap consolidates NVIDIA’s role as a leader in the infrastructure behind large-scale, production-ready AI, challenging competitors like AMD and Google to close the performance gap.