According to new rumours, AMD's Navi 31 GPU will not launch with 120 Compute Units (CUs) and 15,360 Stream Processors. Instead, AMD's next flagship GPU is now rumoured to offer 96 CUs and 12,288 Stream Processors. There are changes for other Navi 3x GPUs too.
https://www.notebookcheck.net/New-AMD-Navi-31-Navi-32-and-Navi-33-GPU-details-emerge-with-up-to-12-288-Stream-Processors-for-rumoured-RX-7900-XT.616559.0.html
TFLOPs are useless when comparing different brands and architectures, including from the same vendor. You'd think people finally got this with Ampere, with stuff like RTX 3090 and it's "fake" 35,5TFLOPs that can only keep up with RDNA2 RX 6900 XT's ~23TFLOPs in gaming. Certainly no way near the ~55% faster as it's TFLOPs would imply if one would look by that.
And with RDNA3 and Lovelace, we have even bigger unknowns. So it gets even more confusing to the average reader when talking TFLOPs in tandem with gaming performance speculations.