Future NVIDIA RTX 4000 “Ada Lovelace” GPUs could cost you dearly… in electricity. According to several separate rumors, the next line of graphics cards
NVIDIA could peak at almost a kilowatt (kW) of TGP for high-end models. Just that.
Furious performances, and record consumption. This is what NVIDIA could offer us with its next generation of RTX 4000 “Ada Lovelace” desktop graphics cards, expected in the second half of 2022. This new range of 5 nm engraved chips could develop high-end performance, but at the price of a scary TGP, at least if we believe the rumors coming from two leakers on Twitter.
Towards a flagship over 800W?
In this case, the AD102 chip (the most high-end of the future line up from NVIDIA), could go up to a TGP of more than 800 W. The often well-informed leaker Greymon55 even estimates that the chip could go up to 850 W in its full version, intended for the future RTX 4090. Expected in September 2022, this GPU would be available in three versions to power the RTX 4080, 4080 Ti and 4090, with TGP ranging from 450 to 850 W depending on the model.
And if the two leakers at the origin of this (very) copious consumer rumor make it clear that their information refers to non-definitive figures, WCCFTech believes for its part that the TGPs mentioned could turn out to be right in the end.
Consistent rumors…
The specialized site indeed recalls that the next graphics cards from NVIDIA should use new PCIe Gen 5 power connectors developed to offer up to 600 W per connector. However, previous rumors suggested that the RTX 4090 could be equipped with two next-gen connectors. The latter would then have no trouble supplying it with 800 or 850 W.
With this new generation of graphics cards, NVIDIA would bet as a reminder on doubled performance compared to current models. What partly explains this possible rise in TGP (the reference RTX 3090 is limited to 350 W).
In terms of specifications, the NVIDIA AD102 chip could carry 18,432 CUDA cores for a clock speed of 2.2 GHz. We could thus benefit from a computing power of 81 TFLOPs in FP32. The existing RTX 3090 is “content” for its part with 36 TFLOPs of FP32 computing power.
Source: WCCFTech
37