Silicon Investor (SI) -- The First Internet Community

STOCKTALK

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor. We ask that you disable ad blocking while on Silicon Investor in the best interests of our community. If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.

Technology Stocks : Semi Equipment Analysis -- Ignore unavailable to you. Want to Upgrade?

To: Sam who wrote (94914)	8/15/2025 12:12:40 PM
From: Sam	2 Recommendations Recommended By alwaysbmiki Return to Sender Respond to of 95383

NVIDIA Reportedly Denies Rubin Delay as AMD MI450 Spurs Redesign Speculation 2025-08-14 With its China-focused H20 still facing uncertainty despite U.S. approval, NVIDIA’s next-gen AI chip Rubin has drawn fresh attention amid delay rumors. But the company told Barron’s and Seeking Alpha the claims are false, adding Rubin remains on schedule. The reports, citing Fubon Research, warn that NVIDIA’s Rubin could be delayed as it is redesigned to counter AMD’s upcoming MI450. Although Rubin was first taped out in June, a second tape-out expected in late September or October points to limited 2026 volumes, according to the research note cited. As per wccftech, if the rumors hold, Rubin could be delayed by four to six months. NVIDIA, however, denies the claims, saying Rubin and its GPUs remain on track under the company’s annual release cycle. According to NVIDIA’s roadmap, Rubin is expected in 2026, followed by Rubin?Ultra in 2027, as per Tom’s Hardware. Notably, the Tom’s Hardware report suggests that Rubin will have an HBM upgrade, transitioning from HBM3/HBM3e to HBM4, while Rubin Ultra will adopt HBM4e. The memory capacity will remain at 288GB per GPU, the same as the B300, but bandwidth will increase from 8 TB/s to 13 TB/s. Moreover, NVLink speed will double, the report adds. As TrendForce observes, NVIDIA is advancing its AI roadmap at a remarkable pace. While the current Blackwell Ultra (GB300) has only just entered production, the next-generation “Vera Rubin” architecture is already nearing design completion, with a market launch expected in the second half of 2026. Rubin is poised to mark a significant technological leap, incorporating TSMC’s 3nm process, HBM4, and NVIDIA’s first-ever chiplet design. However, TrendForce notes that such a compressed product cycle presents a substantial challenge to the supply chain’s ability to adapt. AMD’s MI400: A Tougher Challenger On the other hand, as Barron’s points out, AMD’s upcoming MI400 series, including the MI450, will mark AMD’s first rack-scale AI server with 72 processors—potentially a far tougher challenger to NVIDIA than its previous models. According to TechPowerUp, each MI400 is expected to pack twelve HBM4 stacks, totaling a massive 432?GB of on-package memory and nearly 19.6?TB/s of bandwidth. Wccftech adds that AMD’s modular design, using Infinity Fabric over Ethernet, also enables easier rack deployment—pushing NVIDIA to consider new Rubin configurations, though that remains speculative. Read more [News] TSMC Gears Up for SoIC, Reportedly Secures Orders from NVIDIA’s Rubin After AMD and Apple trendforce.com