At the GTC 2025 conference, Nvidia's CEO Jensen Huang introduced new AI-accelerating GPUs, most notably the Vera Rubin, scheduled for late 2026. The GPU boasts advanced specifications like tens of terabytes of memory and a bespoke CPU named Vera. It promises major performance boosts for AI applications, delivering 50 petaflops of FP4 inference per chip, significantly surpassing its predecessor. The Rubin Ultra, set for 2027, will push boundaries further with 15 exaflops of compute capacity per rack and advanced memory capabilities, making it a game-changer for AI workloads.
Nvidia’s Vera Rubin GPU, set for release in 2026, features significant performance upgrades for AI tasks, leveraging tens of terabytes of memory and custom CPUs.
The unveiling of Rubin Ultra for 2027 promises even greater advancement with 100 petaflops of precision per chip, and an overall rack capability of 15 exaflops for inference.
Collection
[
|
...
]