Asking specific questions of senior CPU or GPU engineers is an interesting exercise. Other than being far more interesting sources of information than marketing folks, engineers sometimes have to think years in the past for an answer. That's because that engineer is probably already hard at work on the next-next generation product.
Right now, we have relatively little information on Nvidia's next-gen Blackwell gaming GPUs. But even if we could ask senior engineers about it, they'll have to think back to Blackwell. At this point in time, Blackwell gaming GPU development is certainly at the post tape-out and software development stage prior to its probable release later this year.
The generation following Blackwell is reportedly named Rubin. It's named after Vera Rubin, a pioneering astronomer. Her groundbreaking work includes the properties of galaxies and evidence of the existence of dark matter.
This information comes from Ming-Chi Kuo, a TF Securities analyst (via TechPowerUp). Kuo states that the R100 GPU will go into production in the fourth quarter of 2025, with full system solutions entering production in the first half of 2026. Nvidia's GPU nomenclature suggests R100 is a high end AI focused GPU and not a gaming variant.
It's reportedly to be built with TSMC's N3 node. It will incorporate HBM4 memory and is being designed with power efficiency in mind. Given that high end AI GPUs are set to cross the kilowatt range for just one card, any efforts to reduce power consumption and increase efficiency are welcome ones. Data centers packed with these systems are increasingly a challenge to manage.
Kuo's post mentions a 4x reticle design and Chip-On-Wafer-On-Substrate-L (CoWoS-L) technology. Depending on how many chiplets a future R100 GPU incorporates, the end result could be a very big chip. Nvidia could probably charge $50,000 for one and it wouldn't be surprising.
As for when we might see Rubin gaming GPUs, your guess is as good as mine. If we follow the Blackwell release cadence, a late 2026 debut is a good bet, with mainstream cards to follow in 2027. That's an eternity in computing terms.
A hypothetical GR102 chiplet-based RTX 6090 with 48GB of DDR7 would be welcome in my rig! Better start saving...
AMD, Intel, Microsoft, and Nvidia are all excited about cooperative vectors and what they mean for the future of 3D graphics, but it's going to be a good while before we really see their impact
AMD claims it has 45% gaming GPU market share in Japan but jokingly admits it 'isn't used to selling graphics cards'
