Overall, the GeForce RTX 4090 is better powered and better cooled than the GeForce RTX 3090/Ti.
NVIDIA has detailed the PCB of its GeForce RTX 4090 graphics card. Externally, the Founders Edition version of this card resembles that of the RTX 3090 / Ti with a dual fan cooling system; the fans are a bit larger though and are said to provide 20% more airflow.
Below, the cutout of the PCB resumes a triangular shape at the back.
The PCB uses a single 12+4 pin PCIe Gen5 power connector, the famous 12VHPWR capable of delivering 600 W. The default TGP of this graphics card is nevertheless set at 450 W. While the GeForce RTX 3090 uses a in 18+3 phases (18 phases for the GPU, 3 phases for the memory chips), its successor moves to a design with 20+3 phases (20 phases for the GPU, 3 phases for the memory). Our colleagues from TechPowerUp also mention the presence of a PID corrector (proportional, integral, derivative) guaranteeing better current stability.
Regarding the memory, all the 16 Gb chips are positioned on the same side, which facilitates their cooling; in addition, possibly thanks to a finer burning process, Micron’s memory modules would consume a little less than before, without quantified value. Still, these adjustments would lower the memory temperature by 10°C compared to the GeForce RTX 3090 / Ti.
NVIDIA also released more information about the AD102 GPU which powers the RTX 4090. This GPU features 76.3 billion transistors and up to 18,432 CUDA cores, 576 Tensor cores and 144 RT cores in a full configuration; remember that the GeForce RTX 4090 “only” uses 16,384 CUDA cores, 512 Tensor cores and 128 RT cores. So there is room for a GeForce RTX 4090 Ti in the future.