NVIDIA Unveils Next-Gen Rubin AI Accelerators Set to Launch in September, Just Six Months Post Blackwell Ultra Release

NVIDIA appears to be accelerating its product development timeline, with reports indicating that its new Rubin AI chips and Vera CPUs may arrive as early as September this year.

Upcoming Launch of NVIDIA’s Rubin R100 AI GPUs for Customer Sampling in September

Team Green continues to push the boundaries of innovation at an impressive rate, spurred in part by the rapid advancements in artificial intelligence. NVIDIA has publicly stated its commitment to a “yearly”production schedule, promising to release a new architecture every twelve months. However, recent insights suggest that this timeline is being condensed, with the gap between the Blackwell Ultra and Rubin platforms shrinking to just six months. According to a report from Ctee, the sampling phase for both Rubin GPUs and Vera CPUs is set to begin in September, with tape-out processes already underway.

So, what can we expect from NVIDIA’s Rubin series? The new R100 GPUs will leverage next-gen HBM4 chips, marking a significant evolution from the current HBM3E technology. Additionally, NVIDIA plans to utilize TSMC’s 3nm (N3P) process and CoWoS-L packaging technologies, setting a new benchmark in performance standards. Notably, the Rubin architecture will feature an innovative chiplet design, making it the first implementation of its kind by NVIDIA, utilizing a 4x reticle design compared to Blackwell’s 3.3x design.

Performance efficiency is a key focus for the Rubin GPUs, especially as the growing energy demands of data centers necessitate more sustainable solutions. Furthermore, NVIDIA plans to replace its renowned Grace CPUs with the new Vera series, which is based on ARM’s next-generation cores and is expected to deliver substantial generational improvements in performance. This evolving hardware stack that integrates Rubin and Vera is anticipated to lead to a significant enhancement in computational power.

Although the upcoming Rubin release generates excitement, one must ponder the feasibility of NVIDIA rolling out distinct architectures so close together. Given the minimal time frame for the supply chain to adopt and implement newer technologies, it’s crucial to consider past challenges faced with the initial batches of AI products, such as Blackwell and Blackwell Ultra. It remains to be seen how NVIDIA will manage the introduction of the Vera and Rubin lineups, which have undergone intensive redesigns.

Source & Images