New RTX 60 series specs leak suggests huge RT performance and memory bandwidth improvement

NVIDIA is expected to maintain its 2-year launch cadence for GeForce graphics cards, as leaks about the next-generation RTX 60 series are starting to surface. The GeForce RTX 60-series will be powered by NVIDIA's "Rubin" architecture, which has already debuted on their AI and datacenter GPUs.

A new leak has surfaced from YouTuber RedGamingTech, combining a few interesting pieces of information about the upcoming NVIDIA GPUs. The first key point is that NVIDIA will use a variant of the current TSMC 3nm FinFET node to produce the RTX 60-series GPUs. Of course, this puts an end to the speculation of a new sub-2nm node debuting with the RTX 60-series.

Of course, NVIDIA will likely collaborate with TSMC to derive a custom node based on the 3nm FinFET that suits NVIDIA's needs, as they have done in the past. Furthermore, it has been leaked that the GPU chips will follow the "GR20X" numbering scheme. This is something that wouldn't come as a surprise if you have been paying attention to NVIDIA's GPU nomenclature over the past few years. Reportedly, the top-end GPU will be the GR202, followed by the GR203 and GR205 GPUs.

The GR202 GPU is likely to correspond with the RTX 6090 flagship graphics card. According to the leak, it will feature a 512-bit GDDR7 memory bus, 192 SMs, and 32GB of GDDR7 VRAM. The GPU will likely see a cut-down number of SMs in the RTX 60-series. The GR203 RTX 6080 is expected to feature a wider 320-bit bus with 20GB of VRAM, while the GR205 RTX 6070 will reportedly have a 256-bit bus and 16GB of VRAM.

The leak also states that the RTX 60 series will feature 6th Gen Tensor cores along with 5th Gen RT cores. NVIDIA will need to improve the performance of its Tensor cores massively if it wants DLSS 5 to have any chance of succeeding. The RTX 60 series is expected to offer at least 2x the path tracing performance as the RTX 50 series, according to the leak, which is a good step in that direction. Moreover, RedTechGaming also states that pure raster performance is around 30-35% higher than the RTX 50 series.

Of course, any performance figures will need to be verified after launch by third-party reviewers. Nevertheless, it seems like NVIDIA is once again banking on its Tensor and RT cores to provide the bulk of the performance improvement over the Blackwell generation cards.