NVIDIA announced on the 18th (local time) that it will start mass production of 'Vera Rubin', which combines its own designed central processing unit (CPU) and next-generation graphics processing unit (GPU), from the second half of next year. The next-generation product of the cutting-edge GPU, Blackwell, known as 'Blackwell Ultra', will be shipped in earnest starting in the second half of this year.
On the second day of its annual developer conference 'GTC 2025', NVIDIA unveiled the next-generation AI semiconductor Rubin and Blackwell Ultra, which were highlighted as the main attractions of the event. Jensen Huang, CEO of NVIDIA, personally introduced the performance and features of the next-generation products.
First of all, the next-generation GPU 'Vera Rubin' is expected to start shipping in the second half of 2026. There have been many changes in the system, as a custom-designed CPU called Vera and a new GPU called Rubin have combined to transform the design of AI acceleration chips and CPUs. Vera is NVIDIA's first self-designed CPU. It is named after astronomer Vera Rubin, who discovered dark matter.
NVIDIA emphasized that the custom Vera CPU can achieve more than twice the performance of the CPU equipped in 'Grace Blackwell', which was introduced last year. It also added that while Blackwell has a speed of 20 petaflops (1 petaflop is one quadrillion calculations per second), Rubin can achieve a speed of 50 petaflops while performing inference.
The hardware capacity of the GPU will also double. NVIDIA plans to launch Vera Rubin Ultra in the second half of 2027. Vera Rubin Ultra will apply HBM4e 1 terabyte (TB). The die configuration will also change. Instead of a GPU composed of two dies (Die, small square pieces) as before, it is decided to switch to a GPU system composed of four dies. Currently, it combines two GPUs into one for a single GPU, but from Rubin Next, which will be shipped in the second half of 2027, it will integrate four GPUs into one.
NVIDIA also decided to ship the enhanced version of Blackwell, called Blackwell Ultra, in the second half of this year. NVIDIA emphasized that "using Blackwell Ultra enables advanced AI services for time-sensitive applications" and that "large-scale cloud service providers expect up to 50 times the revenue compared to using Hopper semiconductors shipped in 2023."