According to a DigiTimes report, NVIDIA has already reached an agreement with TSMC regarding the fabrication of its next-gen Ada (Lovelace) GPUs. Considering the fairly long time left till the production and eventual launch of these GPUs, it’s better to take this report with a grain of sale as similar rumors emerged before the Ampere launch as well. It’s also possible that we’re looking at a deal regarding the manufacture of the Tesla-class accelerators, rather than the mainstream GPU lineup. Regardless, it’s too early to say for sure.
If this report turns out to be correct, then this will be NVIDIA’s first GPU lineup in quite a while to be fabbed on an advanced TSMC node. While Ampere is being manufactured by TSMC, Turing leveraged the older 14nm FinFet process, while AMD’s 1st Gen Navi lineup was produced on the twice as efficient 7nm process around the same time.
The most likely candidate is TSMC’s 5nm EUV, though there is a good chance that NVIDIA might opt for the relatively cheaper 6nm process. Jensen has repeatedly claimed that the process node isn’t as important to the company as the GPU microarchitecture itself. The NVIDIA CEO went as far as to say that the GeForce graphics cards are, “easily twice as energy efficiency compared to the second-best architecture (AMD) on the market, regardless of the number in the transistor side”.
Although Jensen’s claim is debatable, it does fall in line with NVIDIA’s policy when it comes to choosing the process node for its consumer GPUs. As such, I wouldn’t be surprised if Lovelace (RTX 40 series) is fabbed on TSMC’s 6nm node rather than the more advanced 5nm process which is what AMD will opt for its 3rd Gen Navi graphics cards.