Many thanks for your detailed response and the other responses too really appreciate it. Trying to get my head around it. So with hitting 115% on the normalised is no cause for concern ?
many thanks again
On non modded cards, no it's not, because every single card in that lineup with that same bios will hit that in the specific game or test you are running, if they are all at the same clock/overclock settings.
Depending on the game you're running and what you're rendering, it can create different loads on different scenes, stressing different areas of your card.
(note: this is getting sort of wordy, as it explains how a too high TDP normalized can sometimes help you find a problem with a mod).
On cards with shunt mods, it's a LOT more complicated, because you have to shunt mod EVERY "2512" shunt on the card and you need to know the power values (under any specific stress test or scene) BEFORE you modded the card, and then compare the power values AFTER you modded the card. This can be done in HWinfo's GPU sensors section. If all the values decrease by close to the same amount on the input rails, it means you did a good job on the mod. the point of a shunt mod is to make every power rail report a certain % lower power than it's supposed to report, and all the power rails should report the same decrease (they need to be balanced).
TDP Normalized% is Nvidia's power "balancing" reporting in action.
Some people have bad solder joints between the "stacked" shunts they are modding, then one power rail gets badly out of balance and reports more power to the controller chip than it's "supposed" to report.
And that rail will often cause a very high TDP Normalized power limit and a throttle.
Some people were trying to shunt mod 3090's and not modding the PCIE Slot shunt resistor. Then they were throttling at a "raw" (reported to the card bios) 200W TDP, which was a real value of 400W (you double the power reported value when you mod a 5 mOhm shunt resistor when stacking a second 5 mOhm shunt on top of it because it causes the card to report HALF the power on that particular power rail the shunt is on!), when they should only be throttling when the card is reporting an absolute maximum of 400W (this would be 800W real, you'd never reach this without LN2 cooling anyway).
The failure to mod the PCIE Slot shunt resistor caused the "memory" (MVDDC) power rail to report 225W of power draw (Note: it wasn't drawing anywhere near this, was probably about 110W), when its maximum limit is 120W, and PCIE Slot was also reporting 79W when its maximum limit was 75W. (these are not 'real' readings. The current sensing chip was auto correcting the imbalanced power due to the missing shunt mod making PCIE Slot report too high).
Because 225W over 115W is a higher PERCENTAGE overdraw, compared to 79W over 75W, TDP Normalized was reporting about 190% Normalized power, and thus a massive throttle.
(PCIE Slot was also causing a throttle event also but was overshadowed by MVDDC.
