In critical situations a safety factor of 1.5 is...just silly. Let's look at that from the human perspective of an elevator. Rated for 2 tons, or 4000 pounds. Your failure condition is then 6000 pounds. Assuming an average human mass of about 200 pounds per person, this is a rated load of 20 people and a failure of 30. That sound like enough...right? Well, it's move-in day. Somebody wheels a safe onto an elevator, and enough furniture on a dolly to be 500 pounds. They are also there, that's 700. You can now fill-in 3300 pounds...right? Well, not if the elevator is old. It was rated new, but all components wear, and all experience components generally experience cyclic loading...meaning the failure point is much lower.
Let's stretch this to an electrical connector. You have a pin, that has a surface area that contact and conducts. You have thermal expansion and contraction due to electrical loading. You have power draws which regularly cycle, because nobody really "gaming" runs these full bore constantly. You have a mechanical connection embedded in a dissimilar material (copper in plastic). You are now "ballparking" how current is drawn only by (total load in watts)/voltage/connectors = amperage load per pin...which is fantastic is nothing else matters. Now...what are you going to tell me is "thick enough" I can tell you "thick enough" makes sense in purely the power in each wire...but you should really take a look at how those wires connect.
For an example, let's look at a high power connector:
175 amp connector
Note how that connector is pretty beefy...and it's basically designed to be that beefy so no cross-sectional area drops below the main conductor. Note also that the 12 volt high power connectors have a pin going into a shaped metal sheet...meaning that they might not exactly make contact everywhere, unless very firmly set?
8 pin connector end picture
It's almost like the safety factor here has to be huge, especially when you load between 30ish and 900ish (575 regularly rated). Yeah...it doesn't take a lot to see a slightly bent connector, something not inserted fully, or even something fully inserted but having forced the mechanical connection back through the block that would rapidly cause issues with a 870 watt draw swing. And yes, 870 watts is reasonable even if you are only at 1 ms. Consider that that energy going through a resistor (which a connection can be modelled as) will dump an enormous amount of heat, into a functionally thermally insulated plastic block... It's not going to take a rocket scientist to see why we might see issues here. Of course, the "reasonable" response of 4 8 pin connectors is just too expensive of a solution...but that's what you get when your GPU can pull as much power as the 5090.
