Nvidia has reportedly scrapped plans for a four-die version of its upcoming Rubin Ultra GPU, opting instead for a simpler dual-die design, according to a report highlighted by Tom's Hardware.
Modern high-end AI chips are often built from multiple pieces of silicon, called dies, stitched together to act as one giant processor. The more dies you pack together, the more computing power you can theoretically deliver — but the harder the chip becomes to manufacture reliably. Nvidia's original Rubin Ultra was said to use four dies; the new plan reportedly calls for two.
According to Tom's Hardware, the change was driven by "manufacturing execution concerns" — in plain terms, worries about whether the more ambitious quad-die design could actually be built at scale without costly problems. A separate summary from Let's Data Science described the same move: Nvidia dropping the quad-die Rubin Ultra in favor of a dual-die version.
It's important to note that these are report-based claims rather than official confirmations from Nvidia. Tom's Hardware frames the story as "reportedly" and "purportedly" throughout, so the details could shift.
Why it matters: Nvidia's chips power much of the world's AI boom, so even a design pullback on a next-generation product signals how hard cutting-edge manufacturing has become — and hints that the company may be prioritizing reliable production over maximum on-paper performance.