The problem with Nvidia 30x0 cards is that the memory temperature is the main reason for throttling, but there is no direct way of monitoring the memory temperature, only checking when the performance cap reason changes to "temp".
Two things: first, the Perf Cap Reason is "Thrm", not "Temp". Second it's now possible to monitor the memory temperature using the latest HWiNFO. It shows up as "GPU Memory Junction Temperature" and if it goes above ~110C the card will start throttling regardless of the GPU temp.There's a lot of discussion about what temperature is actually safe for long-term use with these cards but it's generally thought anything near 110C is not a good idea.