Nvidia Unveils Blackwell Ultra B300: A Leap Forward in AI Computing

Curated by THEOUTPOST

On Wed, 19 Mar, 12:09 AM UTC

9 Sources

Share

Nvidia announces the Blackwell Ultra B300 GPU, offering 1.5x faster performance than its predecessor with 288GB HBM3e memory and 15 PFLOPS of dense FP4 compute, designed to meet the demands of advanced AI reasoning and inference.

Nvidia Introduces Blackwell Ultra B300: A New Era in AI Computing

Nvidia has unveiled its latest advancement in AI computing technology, the Blackwell Ultra B300 GPU, during CEO Jensen Huang's keynote at GTC 2025 in San Jose, California. This new offering represents a significant leap forward in AI processing capabilities, particularly for reasoning and inference tasks 1.

Key Features and Performance Improvements

The Blackwell Ultra B300 boasts impressive specifications that set it apart from its predecessors:

  • 1.5x faster performance compared to the B200
  • 288GB of HBM3e memory, a 50% increase from the previous generation
  • 15 PFLOPS of dense FP4 compute power
  • Designed for advanced AI reasoning tasks

These improvements enable the B300 to handle more sophisticated AI models, such as DeepSeek R1, which go beyond simple information regurgitation to perform complex reasoning tasks 12.

Scalable Solutions for Enterprise AI

Nvidia is not just offering a single GPU but a range of solutions to meet various enterprise needs:

  • B300 NVL16 server rack solutions
  • GB300 DGX Station
  • GB300 NV72L full rack solutions

The pinnacle of these offerings is the Blackwell Ultra DGX SuperPOD, which combines:

  • 288 Grace CPUs
  • 576 Blackwell Ultra GPUs
  • 300TB of HBM3e memory
  • 11.5 ExaFLOPS of FP4 compute power 13

Performance in Real-World Applications

Nvidia has provided some performance comparisons, albeit primarily against the older Hopper architecture:

  • An NV72L rack can deliver 30x more inference performance than a similar Hopper configuration
  • Blackwell Ultra can process up to 1,000 tokens/second with the DeepSeek R1-671B model, compared to Hopper's 100 tokens/second
  • This translates to reducing query response times from 1.5 minutes to just 10 seconds 12

Market Positioning and Competition

The introduction of Blackwell Ultra comes at a time when Nvidia is looking to extend its lead over competitors like AMD and Intel in the AI accelerator market. The increased memory capacity of 288GB per chip allows for running substantially larger models, which is crucial for memory-bound inference workloads 24.

Future Roadmap and AI Vision

Nvidia's announcement also included glimpses into future developments:

  • Vera Rubin GPUs, expected in 2026, promising further performance improvements
  • Rubin Ultra, slated for 2027, with even more ambitious specifications
  • "Feynman" architecture, anticipated for 2028 3

Jensen Huang outlined Nvidia's vision of data centers as "AI factories," emphasizing the growing importance of computational power in AI advancements 35.

Availability and Industry Impact

The Blackwell Ultra B300 products are expected to start shipping in the second half of 2025. This timeline puts Nvidia in direct competition with AMD's upcoming Instinct MI355X accelerators 24.

As AI continues to evolve rapidly, Nvidia's latest offerings are positioned to meet the increasing demands for more powerful and efficient AI computing solutions across various industries, from research and development to enterprise applications.

Continue Reading
NVIDIA's GB300 'Blackwell Ultra' AI Servers: A Leap Towards

NVIDIA's GB300 'Blackwell Ultra' AI Servers: A Leap Towards Fully Liquid-Cooled AI Clusters

NVIDIA is set to unveil its GB300 'Blackwell Ultra' AI GPUs at GTC 2025, featuring fully liquid-cooled AI clusters. The new servers promise significant performance improvements and mark a shift in cooling technology for AI infrastructure.

TweakTown logoWccftech logo

3 Sources

TweakTown logoWccftech logo

3 Sources

NVIDIA Unveils Roadmap for Next-Gen AI GPUs: Blackwell

NVIDIA Unveils Roadmap for Next-Gen AI GPUs: Blackwell Ultra and Vera Rubin

NVIDIA CEO Jensen Huang confirms the company's plans for Blackwell Ultra and Vera Rubin AI architectures, promising significant advancements in GPU technology for AI and data centers.

TechSpot logoTom's Hardware logoTweakTown logoDataconomy logo

7 Sources

TechSpot logoTom's Hardware logoTweakTown logoDataconomy logo

7 Sources

NVIDIA Unveils Blackwell AI GPUs: A Leap Forward in AI and

NVIDIA Unveils Blackwell AI GPUs: A Leap Forward in AI and Data Center Technology

NVIDIA showcases its next-generation Blackwell AI GPUs, featuring upgraded NVLink technology and introducing FP4 precision. The company also reveals its roadmap for future AI and data center innovations.

Wccftech logoTweakTown logoTom's Hardware logo

4 Sources

Wccftech logoTweakTown logoTom's Hardware logo

4 Sources

NVIDIA Unveils Next-Gen AI Powerhouses: Rubin and Rubin

NVIDIA Unveils Next-Gen AI Powerhouses: Rubin and Rubin Ultra GPUs with Vera CPUs

NVIDIA announces its upcoming Rubin and Rubin Ultra GPU platforms, along with Vera CPUs, set to revolutionize AI computing in 2026-2027 with unprecedented performance and memory capabilities.

TechCrunch logoTom's Hardware logoTweakTown logoWccftech logo

4 Sources

TechCrunch logoTom's Hardware logoTweakTown logoWccftech logo

4 Sources

NVIDIA Rebrands Blackwell Ultra to B300 Series, Featuring

NVIDIA Rebrands Blackwell Ultra to B300 Series, Featuring Advanced HBM3E Memory and TSMC Packaging

NVIDIA has renamed its upcoming Blackwell Ultra products to the B300 series, incorporating HBM3E 12-Hi memory and TSMC's CoWoS-L packaging. The new lineup aims to meet diverse performance demands in the AI chip market.

TweakTown logoWccftech logo

2 Sources

TweakTown logoWccftech logo

2 Sources

TheOutpost.ai

Your one-stop AI hub

The Outpost is a comprehensive collection of curated artificial intelligence software tools that cater to the needs of small business owners, bloggers, artists, musicians, entrepreneurs, marketers, writers, and researchers.

© 2025 TheOutpost.AI All rights reserved