Nvidia's next-generation flagship graphics card is expected to require 450W of power for the GPU alone, and the board as a whole (TGP) will approach 600W. These projections come from Tom's Hardware Germany Alumnus Igor Wallossek, who insists that the same industry insider info and assumptions helped him predict the RTX 3090's power draw "almost exactly right on the watt" ahead of its launch.
Wallossek says Nvidia's GeForce RTX 3090 Ti provides insight into the company's plans for the RTX 4090. Add-in-Board partners (AIBs) have confirmed to Wallossek that the GeForce RTX 3090 Ti PCBs will be similar to the next-gen cards. Still, he warns those starting rumors that they will be "pin-compatible" are indulging in purely wishful thinking, according to his knowledge. Rather, the insight that the RTX 3090 Ti board design provides is mostly about power delivery/consumption.
In his report, the Igor's Lab editor outlines his past GeForce RTX power consumption predictions and applies a similar model to the RTX 4090. We have only reproduced the latter here, but please check out the source if you want to look back at history.
Wallossek says the TGP figure of 600W, which seems to be derived from his insider contacts and the clues found in the RTX 3090 Ti, is like "Ada's Playground."
From the TGP figure of 600W, Wallossek deducts several reasonable subcomponent consumption figures. With Ada Lovelace architecture GPUs like the RTX 4090, the newer GDDR6X from Micron is expected to eat through ~20W less power than the previous gen. Even though they are faster VRAM chips, the quantity required may be cut in half due to their increased density. Other significant deductions made to come to the 450W max GPU consumption come from MOSFETs, fans, PCB losses, and a lot of wasted energy which will be converted to heat.
Though Igor's Lab has a good track record on these topics, please add a sprinkling of salt just in case.
Previous leaks have provided indications that the biggest Ada GPU, probably AD102 and used in a GTX 4090, will offer as many as 18,432 CUDA cores (75% more than the RTX 3090). Both these figures are likely to be somewhat reduced in practice for shipping consumer products due to yields. Ada GPUs may also feature in the first PCIe 5.0 compliant graphics cards.
Unfortunately, the expected GeForce GTX 40-series isn't expected to launch until September. If this timescale is correct, leaks should be pretty intense and get more accurate during the summer. We might also hope for some official Nvidia teasers around Computex time.
For some, waiting for the GeForce RTX 40-series is a bit too far away, and with GPU prices plummeting there may even be some bargains to be had in the coming months. If you are in the market for a graphics card, you can always start by picking through our GPU benchmarks hierarchy, and checking prices and availability in your region.
