This year marks the tenth anniversary of the GeForce GTX 10-series. Since then, Nvidia's gaming GPUs have undergone fundamental changes to bring ray tracing and AI technology to the PC gaming masses. As we look toward the future, predicting the specifications for the upcoming RTX 60-series requires looking at historical data rather than mere guesswork.

With a Super refresh of Blackwell chips looking increasingly unlikely due to VRAM supply pressures, it is time to evaluate what comes next for Team Green. By analyzing ten years of GeForce cards across four primary tiers—60-class, 70-class, 80-class, and the top-tier (formerly Titan, now 90-class)—we can find patterns in Nvidia's hardware roadmap.

Manufacturing Evolution and the RTX 60-series

Nvidia designs its GPUs but relies on TSMC (Taiwan Semiconductor Manufacturing Company) for production. Because TSMC is the industry leader in cutting-edge processor technology, Nvidia utilizes their most advanced nodes to power high-end silicon.

The transition between "process nodes" is a critical factor in predicting future hardware. Here is a look at how Nvidia's manufacturing has shifted recently:

  • GTX 10 and RTX 20-series: Utilized a custom TSMC N16 node (and N28 for prior generations).
  • RTX 30-series: Switched to Samsung's 8LPH process.
  • RTX 40 and 50-series: Built on a custom version of the TSMC N5 node, known as 4N.
  • Rubin AI chips: Utilizing the advanced TSMC N3 node.

For the RTX 60-series, I suspect Nvidia will stick with the TSMC N3 node for cost efficiency rather than moving to the more expensive N2. This decision will directly impact the die density, or the number of transistors per square millimeter.

While current Blackwell and Ada Lovelace chips feature a density of approximately 120 million transistors/mm², TSMC's N3 is reported to reach around 200 million. If this trend holds, we could see a massive increase in density for future generations.

Transistor Density and the Future of CUDA Cores

While a higher density sounds like it would lead to a proportional increase in all components, there are limitations. The density increase applies primarily to logic, such as shader cores. For other essential GPU parts like cache and PCIe/VRAM circuitry, the density increase is much more modest—around 5% at best.

This means that while Nvidia can pack significantly more CUDA cores into its next-gen architecture, they are limited in how much they can expand cache and analogue systems. Historically, Nvidia has favored using smaller dies for most gaming products to improve wafer yields and profit margins.

The only major exception is the top-end 90-class chips. The RTX 5090 chip size is nearing the maximum limit of what TSMC's equipment can produce, a move driven by the demands of the prosumer AI market rather than pure gaming performance.

Ultimately, the data suggests that RTX 60-series GPUs may feature roughly 60-70% more transistors than Blackwell chips while maintaining a similar physical die size. The real question remains how effectively Nvidia will spend that transistor budget.