It’s time (a bit late maybe?) for a post with latest news (and leaks) about the upcoming GeForce RTX 40 Series based on the Ada Lovelace GPU architecture. This post will be regularly updated until official launch.
GeForce RTX 4090
– GPU: AD102-300, TSMC 4nm, boost clock: 2520 MHz
– CUDA cores: 16384 (128 SMs, 128 cores/SM)
– ROPs: 384
– memory: 24GB GDDR6X, 384-bit
– Power consumption: default TGP: 450W, max TGP: 660W
– Power connector: single 16-pin
The TGP is the Total Graphics Power, the power consumption of the whole graphics card.
In a world with cheap electricity prices, this card could have a sense, but today…
GeForce RTX 4080 16GB
– GPU: AD103-300, TSMC 4nm, boost clock: 2505 MHz
– CUDA cores: 9728
– memory: 16GB GDDR6X, 256-bit
– Power consumption: default TGP: 340W, max TGP: 516W
GeForce RTX 4080 12GB
– GPU: AD104-400, TSMC 4nm, boost clock: 2610 MHz
– CUDA cores: 7680
– memory: 12GB GDDR6X, 192-bit
– Power consumption: default TGP: 285W, max TGP: 366W
75 billion transistors
A full AD102 GPU (with 18432 CUDA cores or 144 SMs) would have more or less 75B transistors.
RTX 4090 Possible VGA cooler
Lot a power must be cooled correctly and here is a possible VGA cooler for a RTX 4090:
RTX 4090 with massive 4-slot VGA cooler
This custom RTX 4090 has been found in a Lenovo Legion Gaming PC.