Is "GPU Rail Voltage" an indicator of 12V reliability on a PSU?

[magellan]

How does anyone know their DMM is accurate and to what decimal place of accuracy?

 

[EarthDog]

How does anyone know their DMM is accurate and to what decimal place of accuracy?

I don't know. Perhaps by reading the specs? Don't they normally tell you the accuracy of the device in the specs somewhere?

I see it in the Fluke 87's manual, for example... they also tell you how to check and calibrate.

 

[magellan]

I only have a cheap DMM so it has nothing in the manual on how to check or calibrate it. I'd expect a Fluke to calibrate itself.

 

[yoadknux]

@yoadknux
I'm using the octopus adapter cable for my 4090 -- 3 8-pin PCIe cables feed the octopus adapter, but one 8-pin PCIe connector on the octopus adapter is using an adapter to 3 separate molex cables (8-pin PCIe cables only have 3 +12V lines), because my seasonic 850gm PSU only has 3 8-pin PCIe cables. I was planning on adapting the 8-pin EPS CPU power cable to a 8-pin PCIe (since I don't use this seasonic PSU for anything 'cept powering the 4090). I bought the seasonic Gm850 5 years ago, I never figured any videocard would need more than 3 8-pin PCIe power connectors back then.

I actually did an experiment and bought a CableMod corsair compatiable 12VHPWR -> 4x PSU PCI-E (not the octopus, direct cable). My RM850x doesn't have enough outputs for it so I put the HX1200 back.
The voltage drop is identical, ~11.8V under max load. So this debunks the "more connectors = less current = less drop" that I wrote earlier.
But the 12V input + GPU input are ~0.018V apart so the connection is fine. And the motherboard voltage drops from 12.168V -> 12.096V so on the PSU side of things it should be fine.

 

[Flamethrower1972]

I actually did an experiment and bought a CableMod corsair compatiable 12VHPWR -> 4x PSU PCI-E (not the octopus, direct cable). My RM850x doesn't have enough outputs for it so I put the HX1200 back.
The voltage drop is identical, ~11.8V under max load. So this debunks the "more connectors = less current = less drop" that I wrote earlier.
But the 12V input + GPU input are ~0.018V apart so the connection is fine. And the motherboard voltage drops from 12.168V -> 12.096V so on the PSU side of things it should be fine.

That makes a lot of sense to me because just adding another cable will not have any affect on the output.

 

[EarthDog]

The voltage drop is identical, ~11.8V under max load. So this debunks the "more connectors = less current = less drop" that I wrote earlier.

...or it just means you have quality connectors. More/shitty ones can cause drop/problems.

EDIT: Here you go... it's about cars, but electric is electric and it CAN cause a significant drop (contrary to what the peanut gallery says). It SHOULDN"T drop a lot, but again, if they are shitty connectors or malfunctioning, they absolutely can cause voltage drops.

Connections that Affect Voltage - vestest.com

Wire used in vehicles these days are made from materials that pass current easily. Get in touch with vestest.com for more details.

vestest.com

Manufacturers know how much voltage each of their connections, switch contacts, or lengths of wire will drop. Connectors are listed as having a voltage drop across them based on each amp that flows through them. Small connectors can cause a voltage drop of 4mVolts per amp, whereas, large connectors can cause a voltage drop of 10mVolt per amp. Each connector is rated to handle a certain amount of amperage based on the physical size of their blades or pins. Vehicle connector blade sizes used today range from .8mm (.032 Inches) to 8.0mm (.315 inches). Their amperage carrying ratings range from 0 to 15A, 25A, and 33A for the most common connectors in use.

A small connector rated to drop 4mVolts per amp would drop .060Volts (60mV) with 15amps flowing (.004V X 15A = .060V, E=IR). A large connector rated to drop 10mVolt per amp would drop .300V (300mV) with 30 amps flowing (.010V X 30A = .300V, E=IR). Voltage drop is always proportional to current flow.

 

[Flamethrower1972]

...or it just means you have quality connectors. Mkre/shifty ones can cause drop/problems.

You do have a good point, but it would not be that much of a difference.

 

[yoadknux]

Just a reminder, I had the Corsair RM850x and Corsair HX1200 tested, and both had voltage drops.
Today I tested the Corsair HX1000i (2024 model). It also uses Type-4 cables like the two other PSUs, but it is ATX 3.1 / PCI-E 5.1 certified.
The difference is slim to none. Voltage drop is still in that ~0.3V range in 100% stress tests with 600W power limit, and ~0.15-0.2V in normal gaming load.
By the way, the HX1000i has a digital console, so I can monitor the 12V rail. So now I have 3 12V readings, one from the PSU, one from the motherboard, and one from the GPU, and all 3 are completely different LOL but I guess the PSU one is the more reliable. The PSU one drops about 0.1V under full load, so I guess the GPU drops another 0.1V-0.2V across the cable and connector.
As for overall power consumption (also reported by the PSU), it seems to be around 700-750W (with 1.1V + 600W PL) so of course a 850W PSU is enough for a 14900k/4090 system, especially if its running stock.

 

[Woomack]

Most modern PSUs run between 11.90 and 12.20V, depending on the load, and that's normal. I would start to worry if I saw 11.7V or less and 12.5V or more. Actually, my HX1200, which eventually died, had 11.6x-11.70V under ~50% load.
All FSP, Deepcool, and Corsair PSUs that I tested in the last two years had ~11.5-12.2V range in tests up to 80% load (it's hard to 100% load a 1200-1350W PSU with standard components).
Modern PSUs shut down when voltages are too low or too high. All popular brands have OCP/OVP.

GPU rail voltage is another thing, and I'm not sure why it's often much higher than it's supposed to be. On RTX4070 Super and RTX4080, I get up to 50W or even more than the card's TDP.

ATX3.1 gives you pretty much only one thing - the PSU is supposed to run with spikes up to 100% above the specified values. Everything else is not important.

Since most new PSUs are single-rail, you can measure +12V with a simple multimeter anywhere on the PSU connectors and it will give info about the PSU. Everything else already passes through PC components, so readings are different. You can also check what the BIOS or the motherboard software shows. Hwinfo64 is usually accurate, but not always. Some chips, like those on ASUS mobo, sometimes give weird readings.