Archive — US — techcrunch
Blackwell Ultra, Vera Rubin, and Feynman are Nvidia’s next GPUs
At Nvidia's GTC 2025, CEO Jensen Huang unveiled new GPUs, notably the Vera Rubin set for 2026 and Blackwell Ultra for late 2025, improving AI performance with substantial memory and processing power. Vera Rubin can achieve 50 petaflops, while Blackwell Ultra increases memory and maintains 20 petaflops. Future GPUs, Rubin Ultra and Feynman, are also in the pipeline.
Key points
- Nvidia announced new GPUs at GTC 2025.
- Vera Rubin, releasing in H2 2026, will include tens of terabytes of memory and a custom Nvidia CPU, promising more than double the performance on AI tasks compared to Blackwell.
- Vera Rubin can achieve up to 50 petaflops in inference performance.
- Blackwell Ultra is set for release in H2 2025, with the same AI performance as Blackwell but enhanced memory (288GB).
- Rubin Ultra, coming in 2027, will house four GPUs delivering up to 100 petaflops.
- Feynman GPUs, with a Vera CPU, are expected by 2028.
Why it matters
- Nvidia's advancements reflect the ongoing trend towards hyper-converged infrastructure and AI-first architectures in the IT sector.
- The move to enhance GPU capabilities aligns with the increasing demand for AI-driven solutions and complex data processing in various industries.
- Nvidia's competition with other chip makers highlights the race in AI accelerator technologies as we approach 2025.
Nvidia continues to innovate with its GPU roadmap, aiming to significantly enhance AI computing capabilities, reflecting the industry's need for powerful processors in an increasingly AI-centric future.