Nvidia has revealed the name and planned launch date of its GPU architecture two generations hence. Give it up for Feynman, named after physicist Richard Feynman, and due out in 2028. Well, it's due out in AI GPU format in 2028, more on which in a moment.
Richard Feynman, of course, won the Nobel prize for his work in quantum electrodynamics in 1965. He also worked on the Manhattan Project, was a member of the Rogers Commission that investigated the Space Shuttle Challenger disaster and was arguably as well known as a charismatic science communicator as a scientist per se.
The revelation came as a fleeting call out by Nvidia CEO Jensen Huang during his recent GTC keynote, which also debuted Nvidia's new custom Arm CPU, Vera. Huang gave no details about the new Feynman graphics architecture, other than to mention it by name as the successor to Nvidia's next-gen Rubin technology.
Currently, of course, Nvidia is offering its new Blackwell GPU architecture, both in its AI chips and the latest RTX 50-series gaming GPUs. Following Blackwell, Rubin arrives as an AI architecture in 2026, but given Blackwell gaming GPUs didn't launch until just into this year, we're probably not expecting Rubin-based gaming GPUs until early 2027.
But what of Feynman? Well, it's not absolutely guaranteed that it will transition to gaming GPUs. By way of example, Nvidia's Volta architecture of 2017 was never seen in gaming format. However, it's the only exception to Nvidia's recent rule of architecture sharing across data centre and gaming, so it remains likely that Feynman in some form will become a gaming GPU.
What else do we know? Based on current Nvidia timelines, if the Feynman architecture for the data centre and AI acceleration is a 2028 chip, that probably puts Feynman as a gaming GPU into early 2029.
Now, we know that Nvidia's current Blackwell GPUs are made on TSMC's N4 node. Rubin, meanwhile, is widely expected to be manufactured on one of TSMC's N3 nodes. But what about Feynman?
By 2028, TSMC's next-gen N2 node should be pretty mature. But Nvidia also says it is currently evaluating Intel's foundry services for future manufacturing, which are all about Intel's 18A node. Intel also has a 14A node in its foundry roadmap.
When 14A was added to Intel's roadmap about a year ago, it was pencilled in for a late 2026 or early 2027 ramp up. There's little further indication of whether 14A is on track, other than the fact that Intel's 18A node isn't expected to scale to volume production until early in 2026, so you wouldn't really expect 14A to come out in the same year.
Even so, 14A could be a candidate process for Feynman, as could Intel's 18A process. Beyond that, it's all speculation. If Blackwell was a bit of a disappointment, in part thanks to using a similar manufacturing process to Lovelace RTX 40-series GPUs, we're expecting much more from Rubin.
Where that leaves Feynman is unclear, other than the fact that you can expect to hear that name wheeled out increasingly often from here on.
You must confirm your public display name before commenting
Please logout and then login again, you will then be prompted to enter your display name.
