Regardless of little or no details about Nvidia’s upcoming next-generation Blackwell GPU, there already appears to be plans for the era after that, based on a brand new report.
The knowledge comes from TF Securities analyst Ming-Chi Kuo, (reported on by TechPowerUp). These new R100 chips have been codenamed Rubin (named after astronomer Vera Rubin). They’re primarily centered on reducing the ability draw, as high-end Blackwell GPUs are already reaching the best kilowatt vary and may’t preserve growing.
The opposite greatest takeaway is that Rubin shall be powering future AI-based GPUs, being constructed with the TSMC-N3 node and utilizing HBM4 stacked reminiscence. These specs are key to reducing energy consumption, which is significant contemplating that Nividia is investing extra within the AI market, which is infamous for its excessive power prices. It’s nice information for normal customers, because it means the GPU shall be siphoning much less power and hopefully offer you a break in your electrical invoice.
Nevertheless, avid gamers shall be sorely disillusioned as, based on the report, Rubin isn’t made for gaming however quite wholly dedicated to AI as an alternative. What helps that is the talked about 4x reticle design and Chip-On-Wafer-On-Substrate-L (CoWoS-L) know-how in Ming-Chi Kuo’s put up. More than likely, we may very well be a big and costly chip that might simply go for $50,000. This wouldn’t be the primary time such an enormous price ticket has been placed on a chip.
Thus far, we solely have a mass manufacturing window of This fall 2025, that means that customers received’t be getting their palms on it till no less than 2026.
Who’s Rubin for?
As talked about earlier than, the leaks reveal that the Rubin GPU isn’t made for avid gamers however for devoted AI use. And, judging by the variety of chiplets and the ultimate dimension of the GPU, we’re chips made for the skilled market.
More than likely one of many fundamental targets can be information facilities, as they’ve been struggling to keep up themselves by working the present GPUs, and it’ll solely worsen the upper the ability consumption is for upcoming generations.
That does depart the necessary query of who else is Rubin for. Will Nvidia preserve these chips solely within the enterprise area or will there be client variations for gaming and different normal use? It might additionally imply that Nvidia shall be alternating between client and enterprise GPU any more, which is sensible contemplating how a lot revenue the tech large beneficial properties from the latter.