Meet the next next-generation: At Computex 2024 this morning, Nvidia announced the Rubin GPU architecture, the successor to Blackwell. Rubin GPUs are set to splashdown in 2026, bringing support for 8-Hi HBM4 stacks, and will be followed up with Rubin Ultra GPUs, supporting 12-Hi HBM4 stacks. A powerful new CPU called Vera has also been announced, as well as a Vera Rubin board combining the GPU and CPU into a superchip. The upcoming state-of-the-art processors are named after Vera Florence Cooper Rubin, a pioneering astronomer who established the presence of dark matter.
Waiting on Blackwell
Let’s not get ahead of ourselves. We are still waiting for the next-gen Blackwell GPUs, so it is understandable that the details on Rubin remain thin on the ground. Blackwell architecture parts are expected to debut as B100 and B200 GPUs starting later this year.
Once those first Blackwell GPUs have rolled out, Nvidia has lined up Blackwell Ultra. During today’s keynote, it was highlighted that Blackwell Ultra will push performance forward by adding 12-Hi HBM3E compatibility.
Vera Rubin – CPU, GPU, and accelerator board
The Rubin architecture will debut with support for 8-Hi HBM4 stacks. We have already reported on leaks indicative of R100 CPUs, so that is how we would expect the first Rubin GPUs to land. The latest information we have says they could enter mass production in late 2025. This dovetails with Nvidia’s official announcement(s) today, penciling in Rubin for an (early) 2026 debut.
Rubin Ultra GPUs are set to follow up in 2027, in line with Nvidia's stated "one year rhythm" for data center releases. The architecture will earn its Ultra spurs due to its compatibility with 12-Hi HBM4 stacks.
The Nvidia slide we saw also mentioned the Vera CPU. This processor will feature on a new Vera Rubin accelerator board, alongside one of the aforementioned Rubin GPUs, and is destined to replace current offerings like the Grace Hopper superchip. At the time of writing, we don’t have details about what is new with the Vera CPU. One might suppose that the new Vera CPU will deliver various optimizations to the Arm CPU cores and/or a bolstered overall configuration.
Other highlighted qualities of the Rubin platform include the speedy NVLink 6 Switch performance of up to 3,600 GB/s, and a touted CX9 SuperNIC component offering up to 1,600 GB/s.
Stay On the Cutting Edge: Get the Tom's Hardware Newsletter
Get Tom's Hardware's best news and in-depth reviews, straight to your inbox.
Our first reports on next-gen Rubin architecture GPUs were published way-back in November 2023, when hardware leaker @kopite7kimi first flagged up the codename and two GPUs: the R100 and the GR200.
Mark Tyson
Freelance News Writer
Mark Tyson is a Freelance News Writer at Tom's Hardware US. He enjoys covering the full breadth of PC tech; from business and semiconductor design to products approaching the edge of reason.
More about gpus
Latest
2 CommentsComment from the forums
bit_user My eyes just glaze over, at this point. The horsepower and specs of this hardware is so far beyond anything I have access to or could even use, I barely see any point in even trying to keep track of it.
TBH, I'm having more fun tweaking with my N97 mini PC, lately. I'm doing some perf/W scaling analysis on Gracemont, right now.
Reply
Metal Messiah. Well, since TSMC recently laid out plans for up to 5.5x reticle size chips by 2026 (100x100mm substrate), I'm curious to know how the industry reacts, and whether RUBIN arch will still use a more economical 4x reticle design by that time.
There is at least some surety these R100 GPUs are expected to be manufactured using TSMC CoWoS-L packaging technology on the N3 process node.
A greater than 8x reticle size in a 120x120mm package configuration is also on TSMC's pipeline though, integrating four different SoICs.
Reply
