The alleged specs of some of the first Nvidia GeForce RTX 40-series graphics cards have appeared online. The source, MEGASizeGPU, today provided some details about the RTX 4090 reference design. However, the biggest surprise was his assertion that two RTX 4080 reference designs were reportedly brewing in Jensen Huang’s green cauldron. Before going on, please remember that this rumor may or may not be accurate, so add a dash of salt to the broth.
Some reckon that Nvidia’s Ada Lovelace architecture GPUs for consumers could start to trickle into public view (officially) as soon as this month. But, first, it may begin with high-end launches and releases, just like it did with the Ampere generation. So it is with little surprise that the first confident leaks of tech specs emerging via the Twitterverse, other social media, and tech forums concern purported products like the GeForce RTX 4090 and RTX 4080.
MEGAsizeGPU’s assertions regarding the RTX 4090 are pretty straightforward. He says the RTX 4090 allegedly comes with 24GB of GDDR6 and a reference design PCB constructed from 14 layers. The Nvidia RTX 3090 Ti FE board featured a 12-layer design to put the PCB layers spec into perspective. Still, top-end models for the most strenuous demands (like the K|NGP|N edition) came with 14-layer PCBs to facilitate greater power concentration.
The Ada Lovelace specs Tweet gets edgy where the RTX 4080 is concerned. MEGAsizeGPU claims there are already two “completely different” designs in the pipeline; a 12GB model with ten-layer PCB, and a 16GB model with 12-layer PCB. Both use GDDR6X like the RTX 4090. Nvidia has put out SKUs which vary only (or primarily) by memory quota in previous generations, but it might be more logical if this lower-spec GPU is an RTX 4070 (Ti). MEGAsizeGPU later added that the 12GB sample had a 192-bit memory interface. If we follow through with the theory that the weakest GPU highlighted is the RTX 4070, as it is too different to market as an RTX 4080, that also lines up better with our previous coverage.
If you are worried that one of the first high-end RTX 40 cards to be launched might come packing a 192-bit memory interface – like previous gen 060 tier cards – please check our frequently updated Nvidia Ada Lovelace everything we know feature. In short, about the memory bus, Nvidia is starting to use a large L2 cache with RTX 40-series cards, which will help make up for raw memory bandwidth in their targeted gaming resolutions.