Koptite7kimi gave some detailed parameters for the RTX 4080 and RTX 4070, at least from the video memory, much more generous than the RTX 3080.
Specifically, the RTX 4080 is based on the AD103 core, equipped with 16GB GDDR6X video memory, video memory speed of 21Gbps, full card power consumption of 500W, FE public version with a single 16pin external power supply.
The RTX 4070 is based on the AD104 core, with 12GB GDDR6 video memory, 300W power consumption of the whole card, and 16pin external power supply for FE public version.
From the power consumption alone, it is not difficult to guess that the performance of the RTX 40 series will be very impressive. The TGP of the previous generation RTX 3080 was 320W, and the RTX 3070 was only 220W.
Although the RTX 40 series is not so friendly from the perspective of energy conservation and environmental protection, in the context of the slowdown of Moore's Law, the performance and power consumption are also the reality that current semiconductor products have to accept, and the peak single-precision floating-point (FP32) performance of the Ada Lovelace big core is said to touch the terrifying 100T.
(7916967)