NVIDIA’s CEO, Jensen Huang, announced that the company is advancing into the next generation of computing with a promising new AI architecture known as Rubin. This initiative is anticipated to bring about a revolutionary transformation within the compute marketplace.
NVIDIA Tapes Out Six Rubin Chips for TSMC, Indicating a Comprehensive Tech Stack Revamp
As the tech landscape evolves rapidly, NVIDIA finds itself in a relentless product cycle. Just months after unveiling the ‘Blackwell Ultra’ GB300 AI servers, the company is now shifting focus to the Rubin architecture. Currently in Taiwan, Huang is concentrating on the progress of Rubin at TSMC, where he revealed to local media that NVIDIA has taped out six novel Rubin chips. These include various CPUs and GPUs, which are now set to enter trial production under TSMC’s oversight.
“My main purpose is to visit TSMC as you know we have a next generation architecture called Rubin, and Rubin is very advanced and we have now taped out six brand new chips to TSMC, so all of these chips are now in TSMC’s fabs.”
The forthcoming chips encompass a dedicated CPU, GPU, scale-up NVLink Switch, and an advanced silicon photonics processor, heralding a significant upgrade across the entire tech stack. NVIDIA’s Rubin architecture is poised to redefine computing benchmarks, with foundational enhancements in high-bandwidth memory (HBM), process node advancements, and innovative design elements.
The upcoming R100 GPUs will capitalize on the innovative HBM4 memory chips, representing a significant upgrade over the existing HBM3E standard. Additionally, NVIDIA plans to implement TSMC’s cutting-edge 3nm (N3P) process technology and CoWoS-L packaging methods. Notably, Rubin will also introduce a chiplet design—marking NVIDIA’s first adoption of this technology—and a revised 4x reticle design, compared to the 3.3x design of the Blackwell architecture. This substantial advancement is expected to mirror the transformative impact seen with the Hopper architecture.
Regarding the market debut of Rubin, projections suggest it may emerge between 2026 and 2027, contingent upon the completion of trial production. The heightened anticipation surrounding NVIDIA’s Rubin architecture signifies its potential to be a monumental release for the company.
Related Articles:
NVIDIA RTX 50 Founders Edition GPUs Still in Production, Not Discontinued or EOL
4:12
Intel Arc Pro B50 Emerges as Best-Selling Workstation GPU on Newegg
22:30
Top Four Reasons Gamers Prefer the RTX 5070 to RX 9070 Despite Lower VRAM and Slightly Slower Raster Performance
19:10
Leave a Reply