NVIDIA Blackwell Ultra “B300” Launching in 2H 2025: 1.1 Exaflops FP4 Compute, 288 GB HBM3e Memory, 50% Performance Increase Over GB200

NVIDIA has officially unveiled its latest advancement in AI computing technology—the NVIDIA Blackwell Ultra—marking a significant enhancement in memory capacity and AI compute capabilities designed specifically for data centers.

NVIDIA Blackwell Ultra: A Leap Forward with Enhanced Performance and Massive Memory Capacity

After a somewhat rocky start during the initial launch of the Blackwell series, NVIDIA has made substantial improvements to the availability of its cutting-edge AI solution. This is aimed at supporting leading AI and data center providers with top-tier hardware innovations and solutions. The original B100 and B200 GPU series set a high bar with their robust AI computational power, but now, the forthcoming Blackwell Ultra is poised to redefine industry benchmarks.

The upcoming B300 GPU series will not only feature enhanced memory configurations, including advanced memory stacks of up to 12-Hi HBM3E, but will also deliver superior AI compute performance for expedited processing. These new chips will integrate seamlessly with the latest Spectrum Ultra X800 Ethernet switches, which boast a 512-radix capability, providing unmatched network performance for data-heavy applications.

Overview of the NVIDIA Data Center GPU Roadmap

For more detailed insights about the NVIDIA Blackwell Ultra, please refer to the source.

Related Articles:

AMD Radeon RX 9070 BIOS Modification Boosts Performance: Outpaces RX 9070 XT Reference GPU with Increased Clocks and TGP

7:07April 7, 2025

NVIDIA GeForce RTX 5060 Ti Surfaces in FurMark Database, Suggesting Mid-April Launch Date

16:04April 6, 2025

AMD Increases CPU Market Share by 16.6% While Intel Experiences 10% Decline According to Early 2025 CPU-Z Data

21:14April 5, 2025

MSI RTX 5090 Fails as 12VHPWR Connector Melts: Plastic Bulges from Excessive Heat

18:43April 5, 2025