According to recent reports on the RTX 40 SUPER cards, it appears that some of the new models won’t require additional power. Kopite7kimi mentions that the power consumption values for the RTX 4080 SUPER and RTX 4070 Ti SUPER will remain the same (although there is some debate about the latter model’s name). When we say “power not changing,” we’re comparing it to the original non-SUPER versions. In the case of the RTX 4080 SUPER, this means it will maintain a 320W TDP, while the RTX 4070 Ti SUPER is expected to have a 285W TDP. The only exception is the RTX 4070 SUPER, which will see a 20W increase in power.
These statistics are particularly significant in the context of NVIDIA Founders Edition cards since most custom cards from board partners frequently incorporate alterations to these power configurations. A notable case is the RTX 4090, where numerous custom variants surpass a 450W TDP.
In reference to previous rumors, the RTX 4080 SUPER is anticipated to feature a complete AD103 GPU, boasting 10,240 CUDA cores and 16GB of GDDR6X memory. Details about memory speed are presently unavailable, but they can be reasonably speculated. The RTX 4070 Ti SUPER might utilize either AD102 or AD103 GPUs, equipped with 8,448 CUDA cores and an expected 16GB GDDR6X memory setup, marking an improvement over the 12GB configuration of the RTX 4070 Ti. Lastly, the RTX 4070 SUPER, which is potentially the most intriguing card in this lineup, holds promise for everyday gamers due to its expected pricing. This card is said to house 7,168 CUDA cores, a significant upgrade from the original model’s 5,888 core specification.
As previously stated, the board partners don’t foresee any major announcements for this year. Consequently, it’s logical to assume that the RTX 40 SUPER series won’t make its debut until next year. However, NVIDIA has a history of making adjustments to specifications even after sharing them with board partners. Nevertheless, board partners are primarily focused on power requirements and whether the GPU is compatible with their existing designs, rather than being concerned with CUDA core specifications or clock frequencies.
Source: @kopite7kimi