Blackwell Ultra delivers 1.1 exaflops FP4 compute, 288GB HBM3e memory, and 50% higher performance than GB200.
NVIDIA has announced Blackwell Ultra, building on the groundbreaking Blackwell architecture introduced a year ago. The company is bringing the latest AI powerhouse to market with improved supply and hardware offerings for major AI and data center vendors. This next-generation offering includes the NVIDIA GB300 NVL72 rack-scale solution and the NVIDIA HGX B300 NVL16 system, representing a significant evolution from the initial B100 and B200 GPU families.
The B300 chips at the core of Blackwell Ultra expand upon the original Blackwell with increased memory densities featuring up to 12-Hi HBM3E stacks, delivering enhanced compute capabilities for faster AI. These chips are coupled with the latest Spectrum Ultra X800 Ethernet switches with 512-Radix connectivity. The GB300 NVL72 delivers 1.5x more AI performance than the NVIDIA GB200 NVL72, while increasing Blackwell's revenue opportunity by 50x for AI factories compared with those built with NVIDIA Hopper.
The GB300 NVL72 connects 72 Blackwell Ultra GPUs and 36 Arm Neoverse-based NVIDIA Grace CPUs in a rack-scale design, functioning as a single massive GPU built for test-time scaling. This architecture enables AI models to access the platform's increased compute capacity to explore different solutions to problems and break down complex requests into multiple steps, resulting in higher-quality responses and supporting more advanced AI reasoning capabilities.
The GB300 NVL72 will be available on NVIDIA DGX Cloud, an end-to-end, fully managed AI platform on leading clouds that optimizes performance with software, services and AI expertise for evolving workloads. NVIDIA DGX SuperPOD with DGX GB300 systems uses the GB300 NVL72 rack design to provide customers with a turnkey AI factory. Additionally, the NVIDIA HGX B300 NVL16 features 11x faster inference on large language models, 7x more compute and 4x larger memory compared with the Hopper generation, delivering breakthrough performance for the most complex workloads, such as AI reasoning.