The three RTX 4070 models—4070 Ti and 4080—will be upgraded.
Nvidia left significant gaps in its offering when it introduced their 40-series GPUs in late 2022 and early 2023. It’s a smart move because it enables Nvidia to fend off any shocks from AMD. Now, the business is prepared to close those gaps with Super cards in order to compete with AMD’s 7000-series GPUs and give some of its SKUs a boost that they may have never received from the gaming market.
The RTX 4080, RTX 4070, and the RTX 4070 Ti are three GPUs that Nvidia is preparing Super versions of, according to Twitter user Hongxing2020, who Videocardz claims has a trustworthy track record. With regard to the memory subsystem of its GPUs, the business was notoriously frugal for its 40-series GPUs; this would right that wrong by providing more VRAM and broader memory buses. The change would silence its detractors who complained that it was charging too much for cards with insufficient VRAM and constrained memory buses while also making it more competitive with AMD’s offerings.
10.18 update
— hongxing2020 (@hongxing2020) October 18, 2023
【super】I'm back
4080 super
4070ti super
4070 super
date TBD
…
With its Turing design, the corporation notably switched from selling Ti cards to Super versions while ostensibly dropping the branding for Ampere and returning to Ti. By delivering a mid-cycle refresh of some GPUs, it may now just bring it back, like it did for Turing. Given that the 4090, 4080, and 4070 all utilise different dies, it will probably just repurpose some defective dies and lower them down the stack. A few of these are AD102, AD103, and AD104.
It appears likely that the RTX 4080 Super will receive an AD102 die, similar to way Ampere did. This might enable Nvidia to increase the RTX 3080’s memory bus from 256 bits to 320 bits. In the process, it might upgrade from 16GB to 20GB of memory. Nvidia has lots of room because there is a huge difference in CUDA core counts between the two GPUs.
The RTX 4070 Ti Super has a crazy moniker that makes it sound like Nonsense. However, because that GPU now occupies the entire AD104 die, it will probably also go up to the AD103 die. This might make it possible to upgrade its memory bus from 192 bits to 256 bits and memory from 12GB to 16GB. There is also some room here because it uses 60 streaming multiprocessors (SM) units as opposed to the 4080’s 76.
The RTX 4070 employs a scaled-down version of the AD104 die because the entire chip is already present in the Ti version, hence it is unable to use it. The 4070 has 46 SMs, compared to the Ti card’s 60, leaving a significant difference in SMs. According to Techpowerup, this would enable Nvidia to put the standard 4070 in a better position to compete with the RX 7800/7700 XT cards by positioning it nearer to its Ti sister.