NVIDIA “Rubin” GPUs will enter production in 2025

NVIDIA “Blackwell” GPUs have not even been officially launched on the market and analysts already estimate that the corporation is working hard on the next architecture. With the name “Rubin” – a tribute to Vera Rubin – several factors indicate that its production will begin in the last quarter of 2025.

The statement comes from TF International Securities analyst Mich-Chi Kuo, who also reveals that the first GPU they will produce will be the R100 (not to be confused with the video card of the same name developed by ATI in 2002).

See more:

According to the professional, the NVIDIA “Rubin” GPU will enter mass production in Q4 2025, presenting a huge leap in performance and focusing on its design that will supposedly significantly reduce energy consumption – which is important, since your plates are already approaching KW.

The analyst points out that market trends will continue to be seen before its launch: the manufacturer will officially reveal them to its commercial partners and, before its arrival, some customers will have prior access to the hardware – for evaluations and creation of compatible systems.

Well, we don’t even need to mention that the promise is that NVIDIA “Rubin” GPUs will further raise the level of use of artificial intelligence – the corporation’s current focus and which should not deviate anytime soon. If the “Ada” architecture already has this as a highlight and the “Blackwell” ones also follow this path, then it is unlikely that the next one will deviate from the proposal.

Estimates for the NVIDIA R100 “Rubin” GPU

Mich-Chi Kuo also speculates that the NVIDIA R100 “Rubin” GPUs will use TSMC’s 3nm process with EUV FinFET. Compared to the architecture that will be launched in 2024, which uses the TSMC-N4P (4 nm) with a 3.3x reticle design, the next one will have a 4x reticle design.

However, some procedures will be similar in their production – such as the use of TSMC’s CoWoS-L packaging – which is also being seen in the production of B100 GPUs. It is expected to be the first to use HBM4 stacked memory and will have eight layers.

The NVIDIA “Grace Rubin” GR200 (CPU+GPU) may feature a “Grace” Refresh CPU, also manufactured using a 3 nm process. However, it is worth noting that these are just market estimates and NVIDIA itself has not even revealed a “Rubin” architecture so far. With mass production in 2025, it is possible that they will only arrive in 2026.

Fonte: TechPowerUp