The rumor mill is on overdrive once again thanks to what is expected to be the reveal of the Nvidia RTX 40 Series in less than a week. There have been plenty of corroborating reports about the RTX 4090 launching first, followed by the RTX 4080 roughly a month later. It’s typical that more info spills out in the weeks leading up to official announcements, and there’s now talk of Nvidia offering the RTX 4080 in 12GB and 16GB configurations.
VideoCardz aggregated the info from Chiphell and Twitter leaker Kopite7kimi. According to these sources, Nvidia will take a similar approach to the RTX 3080 by offering two different configurations. The 16GB RTX 4080 will obviously be the more powerful of the two. It will feature higher clock speeds, a better binned GPU die with more available CUDA cores and the like, faster memory at 23Gbps, and a TDP starting at 340W.
The 12GB RTX 4080 will feature lower clock speeds, a cut down GPU die with fewer available specs, slower memory at 21Gbps, and a TDP starting at 285W.
Interestingly enough, the leak also mentions the RTX 4090 featuring 21Gbps memory, which would be slower than the RTX 4080 16GB. At the surface level, it doesn’t seem right that this would be accurate, but there could be design limitations preventing Nvidia from using faster memory on the RTX 4090. There’s also the possibility it could relate to controlling production costs.
12GB and 16GB Nvidia RTX 4080 models would make sense
The reason these rumors have an additional credence is due to precedents set by Nvidia, along with global circumstances. Aside from the RTX 3080, there are plenty of cases where Nvidia releases graphics cards that use the same numerical branding, but feature different specs. Nvidia does this for several reasons.
Creating chips is an imperfect process. Minor defects make some GPUs superior to the next. If you’re familiar with silicon wafers, the general rule is that the chips cut from the center typically perform best and have fewer defects. Manufacturers solve this with a process called “binning” where the bad bits get fused off. This results in worse performance, but the chip is still usable. Hence, we have lower-tier models using the same chip, but they aren’t able to function completely or clock as high. This saves a lot of money across millions of units.
A lot of this also likely relates to power consumption. Energy prices are through the roof, so Nvidia would not be popular if it sold power-hungry graphics cards already factory-overclocked for the sake of competition. AMD is expected to launch potent Radeon 7000 Series graphics cards using multiple GPUs, netting power efficiency and performance advantages.
Nvidia has a special broadcast planned at GTC on September 20 at 11 AM ET. That’s when we should hear about an RTX 40 Series announcement. If we’re lucky, it will provide specs on the RTX 4090, and the possible RTX 4080 16GB and 12GB variants. That would be an attractive, practical high-end lineup.