Nvidia Presents Ambitious Plan for Future GPU and CPU Releases

In a presentation at Computex 2024, Nvidia’s head unveiled a plan to release innovative products over the next three years. These will include state-of-the-art graphics and central processing units (CPUs), as well as other chips and interfaces for artificial intelligence and machine learning. The company aims to annually release flagship solutions that push the bounds of manufacturing technologies.

Nvidia’s Three-Year Plan

During the Computex 2024 conference, Nvidia’s head, Jensen Huang, detailed a three-year plan for products designed for data centers. Nvidia aspires to release new flagship data center solutions annually, constantly enhancing their performance and functionality.

By 2025, the release of the successor to the current flagship GPU, Blackwell — Blackwell Ultra — is expected. This new GPU will feature HBM3e memory with 12 stackable chips, increasing memory by 50% in comparison to Blackwell, which uses 8 stackable chips.

Major Announcements expected in 2026

In 2026, Nvidia plans to make several significant announcements. Firstly, the new graphics chip Rubin featuring HBM4 memory will be unveiled. Secondly, the new CPU, Vera, will replace Grace. Hence, the flagship product of 2026, Vera Rubin, will be an amalgam of GPU and CPU, similar to the existing model, Grace Hopper. Additionally, the bandwidth of Nvidia’s patented interconnect technology, NVLink, will double to 3.6 terabits/sec (NVLink 6 versus NVLink 5).

Plans for 2027

In 2027, an enhanced version of Rubin Ultra featuring 12 stackable HBM4 chips, up from 8, is planned for release. Huang emphasized that Nvidia aims to exploit current manufacturing processes optimally to introduce flagship solutions every year. According to Jensen Huang, this strategy of continuous improvement will help the company maintain its leadership in data center solutions.