"Nvidia CEO Huang Announces New Blackwell Ultra Chip and New Vera Rubin Chip"

Nvidia CEO Jensen Huang recently made a groundbreaking announcement that has sent ripples through the tech industry. The unveiling of the Blackwell Ultra chip and the Vera Rubin chip marks a significant leap forward in AI and high-performance computing (HPC) capabilities. These new products are set to redefine the landscape of AI and HPC, offering unparalleled performance, efficiency, and scalability.

The Blackwell Ultra chip, slated for release in the second half of 2025, promises a 50% performance uplift compared to its predecessor, the B200-series. This chip will feature up to 288GB of onboard memory, thanks to the use of HBM4E memory technology. This high-capacity memory allows for the handling of larger and more complex AI models, enabling faster data processing and improved computational performance.

One of the standout features of the Blackwell Ultra chip is its use of NVLink 6 switches, which support 3600 GB/s of bandwidth. This high bandwidth enables swift and seamless communication among GPUs within a server cluster, leading to accelerated performance for trillion- and multi-trillion parameter AI models. The NVLink 6 switches also support clusters beyond a single server, ensuring that multi-server clusters can scale GPU communications in balance with increased computing power.

The Vera Rubin chip, set to release in 2026, takes these advancements even further. It will feature eight stacks of HBM4E memory, providing up to 288GB of onboard memory. The Vera Rubin platform will also include a Vera CPU, NVLink 6 switches at 3600 GB/s, CX9 network cards supporting 1,600 Gb/s, and X1600 switches. These specifications indicate a significant leap in performance and efficiency, which is crucial for handling the increasing demands of AI and HPC workloads.



The technological advancements in the Blackwell Ultra and Vera Rubin chips translate into significant tangible benefits for customers in terms of performance, efficiency, and scalability. The use of HBM4E memory not only increases the capacity but also improves the efficiency of data access and processing. The high-speed link between the Blackwell GPU and the NVIDIA Grace CPU, providing 900 gigabytes per second (GB/s) of bidirectional bandwidth, accelerates the full pipeline of database queries, leading to higher performance in data analytics and data science.

The NVLink 6 switches enhance efficiency by providing a high-speed interconnect that reduces latency and improves data transfer rates. This results in more efficient use of computational resources, leading to faster completion of tasks and reduced energy consumption. The NVLink 6 switches enable the scaling of GPU communications in balance with increased computing power, allowing for the creation of larger and more powerful AI systems. The NVLink Switch Chip supports clusters beyond a single server, providing 130TB/s of GPU bandwidth in one 72-GPU NVLink domain (NVL72) and delivering 4X bandwidth efficiency with NVIDIA Scalable Hierarchical Aggregation and Reduction Protocol (SHARP)™ FP8 support. This scalability ensures that customers can easily expand their AI infrastructure to meet growing demands.

In conclusion, the introduction of the Blackwell Ultra chip and the Vera Rubin chip aligns with NVIDIA's long-term strategic goals in the AI and HPC markets. These new products offer several competitive advantages over existing technologies, including superior performance, efficiency, scalability, and security. These competitive advantages position NVIDIA as a leader in the AI and HPC markets, capable of meeting the evolving demands of these industries. The technological advancements in the Blackwell Ultra and Vera Rubin chips provide customers with enhanced performance through faster data processing and improved computational capabilities, increased efficiency through reduced latency and energy consumption, and greater scalability through the ability to handle larger AI models and expand their AI infrastructure.