I can't wait for next week to roll around and for us to finally get some concrete numbers about Nvidia's new RTX 40-series graphics cards(opens in new tab), because I cannot cope with all the 'holy heck, if this is true…' rumours popping up every single day.
The latest is that the AD102 expected to be powering the RTX 4090(opens in new tab) is going to have 165% more transistors than the freakishly beefy GA102 monster that ran the RTX 3090 Ti(opens in new tab). Yes, yet another tweaker has come out with some GPU number porn: here's @kopite7kimi(opens in new tab) claiming that Nvidia's full-fat Lovelace chip is going to be sporting more than 75 billion transistors.
For a full frame of reference, the RTX 3090 Ti's GPU houses 28.3 billion transistors inside its massive 628mm² die, and the Navi 21 chip inside AMD's RX 6950 XT(opens in new tab) comes in at 26.8 billion in a 520mm² die.
I get that the Lovelace GPU of the RTX 40-series is going to operate with a smaller TSMC N5 (nominally 5nm) production process, as opposed to the Samsung 8nm and TSMC N7 (7nm) lithography of our reference chips above, but still, that's going to be an absolutely enormous graphics processors.
Surely right at the reticle limits; as big a GPU as it is physically possible to manufacture with today's technology.
Given that the RTX 4090 is expected to offer something in the region of twice the performance of the RTX 3090(opens in new tab), you'd expect it to be bigger. But this suggests Nvidia has needed to throw a whole lot more logic at the problem to be able to top its previous biggest GPU to this level.
The transistor number has come from an update to an old tweet of the serial Twitter leaker. Back in April they set up a Twitter poll asking users
Read more on pcgamer.com