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

[yoadknux]

Hi guys.

I recently did some tests with my 4090 with Corsair HX1200 & Corsair RM850x. The nice thing about those PSUs is that they use the same cable kit, so going back and forth between those PSUs is very easy without disconnecting anything from the actual components.

One difference I noticed between the two PSUs is by how much the GPU Rail Voltage varies. Running the same stress test loop, the two PSUs dropped to slightly different minimal values under load.

The Corsair RM850x varied from a range of 11.693V to 12.104, while the Corsair HX1200 varied from a range of 11.617V to 12.149V (See GPU Rail Voltages on the two screenshots below)

I know that the standard on those is +-5% which is something in the range of 11.4V - 12.6V, so both PSUs are fine and within range, but I'm wondering whether this value indicates which PSU has more reliable 12V rails. Doesn't the RM850x beat the HX1200 in this category, since it has lower variance?
If not, is there another category that I can test that will indicate which PSU has a more reliable 12V rail?

Thanks!

 

[freeagent]

Those PSU's are getting their assets kicked.

To be fair you are doing 500w though. Is that a 450w connector?

 

[EarthDog]

That tells you nothing you want to know, sorry.

1. Spec is spec and the results are quite close anyway.
2. You'd probably see different voltages running it again
3. You're measuring it through software. Use a DMM for accuracy...

....but regardless if you know that, it won't tell you a thing about 'which is more reliable'.

EDIT: I ran FS:E (I don't have a 4K and 4090) and the voltage range was from 12.102 to 12.210. This is a Tier 1A Seasonic 850W Prime Titanium unit. Not sure how long you let that stress test run...

 

[yoadknux]

That tells you nothing you want to know, sorry.

1. Spec is spec and the results are quite close anyway.
2. You'd probably see different voltages running it again
3. You're measuring it through software. Use a DMM for accuracy...

....but regardless if you know that, it won't tell you a thing about 'which is more reliable'.

EDIT: I ran FS:E (I don't have a 4K and 4090) and the voltage range was from 12.102 to 12.210. This is a Tier 1A Seasonic 850W Prime Titanium unit. Not sure how long you let that stress test run...

I did the stress test, it's 20 loops. The results are consistent through various runs.
I think that the power connector has gone through some punishment (I had it in and out at least 5 times... with all the playing around w/ thermal pastes). Could explain the drop.

Post magically merged: Apr 28, 2024

Those PSU's are getting their assets kicked.

To be fair you are doing 500w though. Is that a 450w connector?

Yes, that's a direct 12vhpwr connected to the GPU on one side and to two output ports of the PSU on the other side.
In gaming it typically hovers around 11.8V.

 

[EarthDog]

Well, I can't run the Ultra version of that stress test, but I doubt my results would be any different.

The range we're seeing, I suspect, is the difference between 'average' PSUs and one of the best out there. But again, this doesn't show you anything about reliability...

I wouldn't sweat this for another second.

 

[yoadknux]

Well, I can't run the Ultra version of that stress test, but I doubt my results would be any different.

The range we're seeing, I suspect, is the difference between 'average' PSUs and one of the best out there. But again, this doesn't show you anything about reliability...

I wouldn't sweat this for another second.

Hmm, I guess I'm just trying to decide which PSU to keep in my system, since they both work fine, the efficiency difference between them is like 3% at best, and the only "indicator" I have is the rail voltages which you say don't really matter. Any other test I should look at then?

 

[freeagent]

I am not pulling big power like you are with my little 4070Ti

 

[yoadknux]

I am not pulling big power like you are with my little 4070Ti

View attachment 366541

I wonder if there is a calibration difference between the cards, since your drop is about ~0.3V from maximum to minimum, mine is also around that number, just different absolute values.

 

[EarthDog]

You're not comparing apples to apples (4070ti and 4090)... different psus too... it's also not a function of the card's programming. It's a PSU thing.

The card is programmed to get a specific voltage (at a given clock...youve seen VF curves). Any droop is going to be a function of the psu and card attached to it (as in how much power it draws)

I'd use the 850. 1.2Kw is overkill and since you're worried about efficiency, you'd be in the sweetspot of the 850 more often... but especially with just desktop. Where you desktop load is around 125-150W which is ~10% of that 1.2kw psu. In others words, well out of the sweetspot (80 plus gold starts at 20%, titanium sets a bar at 10%).

But if you start having stability issues, id look there first.

Are these older psus? Both gold?

...and again, it's software, so it could be off in the first place.

 

[yoadknux]

You're not comparing apples to apples (4070ti and 4090)... different psus too... it's also not a function of the card's programming. It's a PSU thing.

The card is programmed to get a specific voltage (at a given clock...youve seen VF curves). Any droop is going to be a function of the psu and card attached to it (as in how much power it draws)

I'd use the 850. 1.2Kw is overkill and since you're worried about efficiency, you'd be in the sweetspot of the 850 more often... but especially with just desktop. Where you desktop load is around 125-150W which is ~10% of that 1.2kw psu. In others words, well out of the sweetspot (80 plus gold starts at 20%, titanium sets a bar at 10%).

But if you start having stability issues, id look there first.

Are these older psus? Both gold?

...and again, it's software, so it could be off in the first place.

Well as you've seen in the thermals topic, I'm ever the scientist, tweaking things that were already working to begin with

Both PSUs are relatively new. The RM850x Gold I bought two years ago for my 3080Ti+9900k gaming rig (which is now 4090+14900k). The HX1200 Platinum I bought three years ago, used one year to power a mining rig (2x 3070Ti + desktop PC), then left unused in a closet.

The system works fine with either PSUs. Passes stress tests at max power limits, doesn't crash in games, no strange behavior. And I benched it A LOT when I went through 3 different thermal pastes and PTM7950. I'm really happy! The HX1200 on paper is the more efficient of the two, but the voltage drop (according to software) is larger.

EDIT: Huh... that's funny. I changed the 16HVPWR cable from my white sleeved one to an ugly rigid one and now the voltages look more reasonable (that's on the RM850x):


The ugly rigid cable:

 

[EarthDog]

Yeah, cables make a difference too! Glad you found the culprit.

I wouldnt worry about the voltage drop... but I still wouldn't use the 1.2Kw psu, especially if you're mostly running a desktop.

 

[magellan]

A better 16HVPWR cable w/lower resistance would definitely reduce the voltage drop seen at the videocard voltage regulator input pins. My 4090 is hooked up to a dedicated Seasonic 850W Focus GM850 and while playing Metro Exodus Enhanced Edition I've seen the GPU 16-pin HVPWR voltage drop to 11.8V, while the GPU PCIe +12V input voltage dropped to 11.865V. If I plug a DMM into the +12V molex leads on the seasonic GM850 PSU though, it indicates nearly +12V, if I measured the lump resistance of the cable I could figure out just how much current is being drawn. The GPU power usage noted during the above voltage drop was:
aftrbrner: 582Watts, HWiNFO64 GPU 16-pin HVPWR power max.: 550 Watts, GPU PCIe +12V Input Power max.: 16.4W.

 

[yoadknux]

A better 16HVPWR cable w/lower resistance would definitely reduce the voltage drop seen at the videocard voltage regulator input pins. My 4090 is hooked up to a dedicated Seasonic 850W Focus GM850 and while playing Metro Exodus Enhanced Edition I've seen the GPU 16-pin HVPWR voltage drop to 11.8V, while the GPU PCIe +12V input voltage dropped to 11.865V. If I plug a DMM into the +12V molex leads on the seasonic GM850 PSU though, it indicates nearly +12V, if I measured the lump resistance of the cable I could figure out just how much current is being drawn. The GPU power usage noted during the above voltage drop was:
aftrbrner: 582Watts, HWiNFO64 GPU 16-pin HVPWR power max.: 550 Watts, GPU PCIe +12V Input Power max.: 16.4W.

Thanks for providing your insight. After doing some research on this, I learned that:

a) The standard is +-5%, meaning anything in the range of 11.4V - 12.6V is acceptable
b) cables make a difference

I think that what happens (in my case): The cable is directly inserted into two slots that are typically reserved for a PCI-E power cable, rated at 150W each, yet the card consumes 450W. The increased current increases the resistance of the cables, causing a voltage drop. My theory is that if I use the Medusa adapter or a direct 16VHPWR -> 3 or 4 PCI-E then the voltage drop would be decrease.

 

[magellan]

@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.

 

[Flamethrower1972]

Everyone should have a good power supply tester.

Digital LCD Display PC Computer Power Supply Tester Power Checker Power ATX Measuring Diagnostic Tester Electronic Repair Tool

Huge amount of items at rock bottom prices

warehousesoverstock.com

 

[yoadknux]

Everyone should have a good power supply tester.

Digital LCD Display PC Computer Power Supply Tester Power Checker Power ATX Measuring Diagnostic Tester Electronic Repair Tool

Huge amount of items at rock bottom prices

warehousesoverstock.com

I don't know about "everyone"...
I'm as much as an over-thinker and the thread starter. And still, I think people exaggerate way too much when it comes down to power supplies, especially when discussing capacitance. A PSU has one role and that is to give the system power. Then it's either enough or its not enough, kinda like RAM, except when a PSU is overloaded or overheats it lets you know by shutting down and then you know its time to switch.

 

[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 see. It could be that the adapter itself has some resistance. Although like you wrote, the software measurement can be off compared to the real measurement.

 

[magellan]

These days it might be better to use 16AWG or maybe even 14 AWG for power cables. There's a guy on etsy, who's in the PRC I believe, who will custom fabricate power cables. All the custom adapter cables I have are 16AWG instead of the more common 18AWG. The octopus adapter that came with my Gigabyte 4090 Gaming OC is 16AWG I believe.

 

[yoadknux]

I just want to give an update about what I recently learned about GPU rail voltage.
- It is NOT a direct indicator of the 12V rail stability. If BOTH the GPU rail voltage & Motherboard 12V drop below 11.4V, then something is wrong. If the motherboard 12V remains high then it's fine.
- It can indicate a voltage drop on the cable itself. This is normal because the cable heats up from drawing high current and feeling the hot air coming out of the GPU heatsink (Might explain why EarthDog got such stable voltage).
- Based on 2 findings that I've seen online, It can indicate a problem with the connector - But not the absolute value. The delta between "GPU PCIe +12V Input Voltage" and "GPU 16-pin HVPWR Voltage" increases when there is a burnt connector. This was reported in:

https://www.reddit.com/r/nvidia/comments/169mep0/find_the_12vhpwr_connector_is_failing

https://www.reddit.com/r/nvidia/comments/146pqgk/rtx_4090_hwinfo_voltages

When those two values are >0.3V apart, then it's worth checking the connector for damage.

 

[EarthDog]

Great summary of what we've found/been saying!

EDIT: You could have determined it was the cable by gathering HARDWARE readings at another PCIe plug. But yeah, we've touched on these things here and there.

1. Not a barometer for reliable PSU
2. Spec for PSUs (5%)
3. Get hardware readings instead of software