Roadmap of upcoming GPU generations Nvidia will launch Ruby GPUs from 2026

From Manuel Christa | Translated by AI 2 min Reading Time

Related Vendor

Nvidia announced several GPU generations at its in-house technology conference GTC 2025: Starting in the second half of 2026, the Ruby architecture will be launched. In 2027, an even more powerful version called Ruby Ultra will follow. The subsequent architecture has also been scheduled.

New graphics processors announced: Nvidia to launch Ruby architecture in 2026. (Imahe: Nvidia)
New graphics processors announced: Nvidia to launch Ruby architecture in 2026.
(Imahe: Nvidia)

The Rubin architecture is named after the astronomer Vera Rubin and is expected to see a significant increase in computing power. Nvidia plans to use HBM4 memory, which allows for a memory bandwidth of up to 13 TB/s. Alongside the new GPU architecture, Nvidia also introduced a CPU called Vera. This CPU uses 88 specially designed ARM cores with a total of 176 threads. An NVLink interface with a bandwidth of 1.8 TB/s ensures data exchange between the Vera CPU and Rubin GPUs.

Nvidia Ruby: Significantly higher computing power

Compared to the current top models based on Blackwell, Nvidia significantly increases the performance of the new Ruby GPUs. The upcoming Ruby NVL144 GPU will be equipped with 144 individual GPU dies and will achieve 3.6 exaflops in FP4 calculations. This makes it more than three times faster than the current Blackwell B300 NVL72 model.

Gallery

Similarly, the FP8 training performance is also set to increase significantly, from the current 0.36 to 1.2 exaflops. Although the memory capacity per GPU remains at 288 GB, Nvidia significantly increases the data rate within and between individual GPUs with the new technology.

In addition to these improvements, the speed of the NVLink interface will be doubled, reaching a total of 260 TB per second. Nvidia is also enhancing the connection between individual racks: The CX9 connection doubles its throughput to 28.8 TB per second.

Subsequent generation Feynman

In addition to Ruby and Ruby Ultra, Nvidia has already announced the subsequent GPU generation: This new architecture is planned under the name Feynman for the year 2028. Nvidia thus emphasizes its strategy to meet the growing demand for powerful GPUs for applications in artificial intelligence and data analysis in the long term.

The upcoming Ruby and Ruby Ultra GPUs are thus central components in Nvidia's long-term strategy to remain a leader in computing performance in the coming years. (mc)

Subscribe to the newsletter now

Don't Miss out on Our Best Content

By clicking on „Subscribe to Newsletter“ I agree to the processing and use of my data according to the consent form (please expand for details) and accept the Terms of Use. For more information, please see our Privacy Policy. The consent declaration relates, among other things, to the sending of editorial newsletters by email and to data matching for marketing purposes with selected advertising partners (e.g., LinkedIn, Google, Meta)

Unfold for details of your consent