NVIDIA’s CEO, Jensen Huang, took the stage at the GTC Conference in San Jose, California, to unveil two new chips: Vera Rubin and Blackwell Ultra.
Imagine a supercomputer in the palm of your hand – that’s what NVIDIA is promising with Vera Rubin, a chip that combines a CPU and GPU into one powerful package. Vera Rubin, named after the renowned female astronomer, is set to hit the market in the second half of 2026. It features a custom-designed CPU, called Vera, based on NVIDIA’s Olympus architecture, which is twice as fast as the CPU in the current Grace Blackwell chip.
When paired together, Vera and the Rubin GPU will deliver AI processing speeds of up to 50 PFLOPs, more than twice as fast as the current Blackwell chip. The Rubin GPU also supports up to 288 GB of memory. What’s unique about Rubin is that it’s essentially two GPUs in one, and NVIDIA plans to take this design even further with the Rubin Next chip, which will combine four chips into one and double the speed again.
But that’s not all – NVIDIA also announced the Blackwell Ultra, a new GPU that will be available later this year. This chip is designed to generate more tokens per second, making it ideal for cloud service providers to offer AI services. According to NVIDIA, Blackwell Ultra will be able to handle tasks up to 50 times faster than the previous Hopper architecture-based chips. The Blackwell Ultra will come in two flavors: a version paired with an Arm-based CPU called GB300, and a standalone GPU version called B300.
Looking ahead, Huang revealed that NVIDIA’s next chip architecture, named after physicist Richard Feynman, is expected to be released by 2028.
For more information, visit CNBC.