The power consumption of NVIDIA GeForce RTX 40 series graphics cards may exceed 450W
Nvidia’s GPU based on the Ada Lovelace architecture corresponds to the GeForce RTX 40 series, which is rumored to be released at the end of this year, which may be one of the most important releases in 2022. It is said that the Ada Lovelace architecture GPU will no longer be manufactured by the Samsung fab, and will be manufactured by TSMC’s 5nm process. The GeForce RTX 40 series competitor will be AMD’s Radeon RX 7000 series, which is also expected to launch in late 2022.
Recently, Twitter user @greymon55 revealed that the flagship AD102 will have three TGPs, namely 450W, 650W, and 850W, but it is not the final specification, and there may be some deviations in the end. Twitter user @kopite7kimi said that the power consumption of GeForce RTX 4080/4080 Ti/4090 reached 450W/600W/800W respectively, but the credibility is questionable, and they should be reserved.
This isn’t the first time news of the high power consumption of the GeForce RTX 40 series has been heard, but it may be higher than previously thought. Earlier reports pointed out that the TDP of flagship products will reach 450W or above, and even reach the level of 600W. This also explains why Nvidia is so eager to promote the new PCIe 5.0 external power supply interface.
It is rumored that the AD102 of the Ada Lovelace architecture is a large chip of 600mm², equipped with 18432 CUDA cores, with GDDR6X memory, and the bit width is 384 bits. The overall performance is doubled on the basis of GA102, and the core clock of 2.2 GHz can provide 81 TFLOPs of computing performance (FP32).
Unlike AMD, Nvidia’s Ada Lovelace GPU will retain the traditional design, which won’t come in an MCM multi-chip package like AMD’s RDNA 3 architecture-based Navi 31, which is AMD’s first consumer GPU to use the technology. NVIDIA temporarily only uses MCM multi-chip packaging on the Hopper architecture GPU, which is different from the Ada Lovelace architecture GPU, which is a product for data centers and AI computing.
