And according to Roman, he's seeing too much current going down just two wires. 

 

 

If the GPU doesnt care about how much current is getting to it through each wire, why is that not being balanced from the PSU end?  Surely if the power load was more evenly distributed this cable wouldnt be a problem?   I dont know enough about electricals and PSUs to say, but are there differences between different instances of PSU's where they handle the power draw differently?

2 hours ago, kevinfirth said:

 Surely if the power load was more evenly distributed this cable wouldnt be a problem? 

 

Exactly. So much current shouldn't be travelling down just two conductors. Roman was seeing 22 amps instead of 6-8 amps on one wire. 

I've seen it suggested that if all but two of the pins have excessive resistance then the current will be drawn down the two that haven't. Seems plausible. 

With the old PCIe cables there was a big margin, there isn't with the 12VHPWR cables. 

 

2 hours ago, kevinfirth said:

are there differences between different instances of PSU's where they handle the power draw differently?

 

I wouldn't think so, at least, not as I know of.

Roman seems to suggest that just one connector isn't enough, and the cards would be better with two 12VHPWR cables. 

I'm not sure why the decision was made to shrink the connectors and conductor size down in the first place, to be honest. Seems like its asking for trouble.  When you do that the quality of the cable and conductors need to be enhanced to match. 

 

Edited February 11, 20251 yr by martin-w

This will shed some light on the issue.

Basically we've gone from 3 shunt resistors with the 3090 to one with the 5090. 

 

 

Edited February 11, 20251 yr by martin-w

Makes me consider Asus when I eventually upgrade. All conductors monitored. 

 

 

They hit the record and never melted the power connector.

 

On 2/18/2025 at 4:24 AM, G-RFRY said:

They hit the record and never melted the power connector.

 

Yep, it's looking like it has to be a reasonably well used connector. Romans was well used, as was Tom's on OC3D. Tom had pin warnings on his Asus card, red lights on two pins. Asus cards monitor the amps on each pin.

I'm thinking, pins become lose, resistance increases, and the load therefore increases on the good pins. 

Some are looking at including thermistors, copper heatsinks and all manner of stuff to stop the connectors overheating if worn.

The cause is the issue, though, which seems to be not much margin, not over specced enough, so well used connectors can cause an issue. 

If I get a 5090 it will be a card with monitoring on each pin, like the Asus. 

Edited February 19, 20251 yr by martin-w

It's a real shame that we've got to the point where research is required simply to join a GPU with a PSU.  Especially considering the money involved 

27 minutes ago, micstatic said:

It's a real shame that we've got to the point where research is required simply to join a GPU with a PSU.  Especially considering the money involved 

 

Exactly. It's just baffling how they thought such a puny connector was a good idea, with very little margin, and nothing stopping the card from drawing all its power from one or two conductors. 

Edited February 19, 20251 yr by martin-w

https://www.guru3d.com/story/fsp-12v2x6-cable-has-temperature-sensor-that-addresses-prevents-overheating-in-rtx-50/

6 hours ago, G-RFRY said:

https://www.guru3d.com/story/fsp-12v2x6-cable-has-temperature-sensor-that-addresses-prevents-overheating-in-rtx-50/

 

At least they are trying. 

In the video where Jayztwocents addressed the issue it seems that some brands, even respectable ones like Corsair, use substandard cables where only some of the connectors make a proper connection.

To me it seems the result of an ever more common practice by companies to save a few bucks and hope they can get away with it. Both Nvidia and some PSU manufacturers are to blame, and the problem arises when on both ends substandard components or design choices are made.

PSU manufacturers could built in all kinds of expensive monitoring stuff, making us pay through the nose for them and yet the issue could be prevented by just sticking to high standards.

https://www.tomshardware.com/pc-components/gpus/asus-rog-astral-rtx-5090-breaks-four-world-records-pushed-beyond-3-45-ghz-with-35-gbps-vram