Nvidia Unveils Next-Gen GPUs: Blackwell Ultra, Vera Rubin, Feynman

# Nvidia # GTC # GPUS

At Nvidia's GTC 2025 conference in San Jose on Tuesday, CEO Jensen Huang unveiled a lineup of upcoming GPUs set to hit the market in the coming months. The star of the show? The Vera Rubin GPU, slated for a second-half 2026 release. This beast boasts tens of gigabytes of memory and comes with a custom Nvidia-designed CPU called Vera. Nvidia claims that Vera Rubin offers a huge leap in performance over its predecessor, Grace Blackwell, especially in AI inferencing and training tasks. When you pair Vera Rubin — which is essentially two GPUs in one — with the Vera CPU, it can crank out up to 50 petaflops during AI model inference, doubling the 20 petaflops of Nvidia's current Blackwell chips. And Vera? It's about twice as fast as the CPU in the Grace Blackwell GPU. Following Vera Rubin, Nvidia plans to launch Rubin Ultra in the second half of 2027. This powerhouse packs four GPUs into one package and promises up to 100 petaflops of performance. Looking a bit closer, Nvidia's got Blackwell Ultra lined up for release in the second half of 2025. This GPU will come in various configs, with a single Ultra chip matching Blackwell's 20 petaflops of AI performance but with a beefy 288GB of memory, up from 192GB in the standard Blackwell. On the distant horizon, we've got the Feynman GPUs. Huang didn't spill much tea about Feynman's architecture during the keynote — just that it'll feature a Vera CPU and is named after the legendary physicist Richard Feynman. Nvidia aims to roll out Feynman, the successor to Rubin Vera, sometime in 2028. *Updated 3/18 3:07 p.m. Pacific: An earlier version of this story mistakenly stated that Vera Rubin had "tens of terabytes" of memory. It actually has "tens of gigabytes." We apologize for the error.*

Related article

Comments (20)