Are these PSU readings ok? Also weird RGB light behavior when switched off at the wall

Associate
Joined
10 Dec 2007
Posts
1,677
I have a 5950x on a Dark Hero motherboard. I currently only have a 3070 but bought a Corsair RM1000x (2015 model) as it would allow easy upgrades to whatever Nvidia comes out with.

I know software readings are far from accurate but I believe they should give a decent understanding of any potential issues. While pure idle in the BIOS or Windows (hwinfo64) my 12v rail is showing 11.984v

However, any slight change in load such as moving the mouse or a program requesting data (I guess anyway) puts it to 11.928v.

During a game it shows it can dip to 11.872v. What I don't understand is this is when the 3070 was drawing over 200w and the CPU about 70w. During an all core cinebench run the rail goes to 11.816v
I know stock the 5950x can draw about 135w so how on earth is the 12v rail lower than when a 3070 is sipping more power?

Another weird behavior that initially led me to check this stuff. When the PC is shut down the keyboard rgb, the white start button led on the motherboard and the i/o cover rog rgb is still active.

The problem is this also happens when the switch at the wall socket is turned off... it only goes off if I switch the plug off at the back of the PSU. From then it doesn't matter if I have it on at the wall socket or not, no lights will shine. How on earth is the board getting power from a socket that's not on?
 
Associate
Joined
4 Jun 2021
Posts
456
Location
Yorkshire
The Dark Hero has voltage measurement points on the board, if you really want to know for sure what the voltages are.

The RGB and USB power staying on when the PC is shut down is normal - they are powered from the 5V standby from the PSU. If I remember correctly there are BIOS options to turn those off.

Don't know what's going on when you switch off at the wall, though. If they fade after a short time, then it's probably just the PSU caps discharging.

Any difference if you pull the plug out as well as switching the socket off?
 
Soldato
Joined
18 Nov 2011
Posts
4,213
Location
London
I have an older ROG STRIX X370-F, which has similar lighting oddities to you.

I think with the keyboard there is a BIOS setting for USB power after power off, which can be handy for charging bits but does keep the led on my mouse and keyboard on.

As for the motherboard lights, for me at least there is no BIOS option to disable it. I found that I am able to set this within the software Armoury Crate. It used to be called Aura sync, but I think they've all bundled their tat into one app now. Anyway, there is an option for RGB lights when powered off in there.

Although to be fair, even after setting it for me, it forgets it. A lot...
 
Associate
OP
Joined
10 Dec 2007
Posts
1,677
@VersionMonkey @SonicSW20 I thought it would be something like that so I went to sleep. Woke up and the lights were still on. Either 8 hours isn't enough to discharge the caps or it's getting power while off at the wall. What I'll do next time is pull the plug and see if the lights stay on.

@HecFam Yeah I found the BIOS option to do it (Stealth mode) but again this is just masking the problem rather than fully fixing it.
 
Associate
OP
Joined
10 Dec 2007
Posts
1,677
Okay my god... I figured it out. When switching off at the wall I noticed it didn't make the "click" noise more of a mushy feel. What was happening was even when the switch was visibly off I had to really press down at the bottom part to hear the click. This is explains why the lights stayed on all night... it actually was on lmao. All behaves fine now (lights go off after about 5 seconds) when actually switched off at the wall.

Only reason I figured it out was because I changed sockets and my surge protector light was still on after I switched it off.

So yeah only issue now is seeing if these 12v rail voltages are fine.
 
Associate
Joined
19 Jul 2015
Posts
483
...During a game it shows it can dip to 11.872v. What I don't understand is this is when the 3070 was drawing over 200w and the CPU about 70w. During an all core cinebench run the rail goes to 11.816v...
I don't know exactly where on the motherboard the 12V sensor is, but assuming it's connected to the ATX 12V connectors, then it will be affected by the voltage drop in the cable when the CPU is drawing power, but not affected by the voltage drop in the GPU power cables. This is of course one of the reasons why on-board voltage readings shouldn't be taken at face value.
 
Associate
OP
Joined
10 Dec 2007
Posts
1,677
@Mr Evil Makes sense.

I honestly think I'm just trying to find issues with the PC at this point and it's not healthy. You spend months planning this stuff and reading about people who have horror stories with their PC that when something just works for you, it's hard to take it at face value.

That's the aida64 stress test passed. Max 70c temps on this 16 core.

I stumbled upon a forum post by the hwinfo64 author in relation to another high end Asus board: https://www.hwinfo.com/forum/threads/voltage-rail-readings.5684/

0.096v accuracy is indeed interesting. If I take my normal 11.928 reading and add 0.096 I coincidentally get 12.024v which kind of makes sense and would match most reviews (tpu review has 12.030v).

If I add that value to the worst case value under aida64 stress test/cinebench of 11.816 that gives me 11.912v
 
Associate
Joined
19 Jul 2015
Posts
483
In the interests of Science™, I performed an experiment. I have the same motherboard, and a Seasonic Prime 1kW PSU. I used HWiNFO64 to measure 12V as reported by the motherboard, and a multimeter (uncalibrated but should be accurate) to measure the output directly at the PSU where the GPU and CPU cables are plugged in. I used FurMark and Prime95 to place a load on the PSU, and this is what I got:

Code:
     Meter   Mobo    AC W
Idle 12.292  12.208  82
GPU  12.330  12.208  380
CPU  12.301  12.152  200
Both 12.328  12.152  450


The PSU actually has a slightly negative output impedance, producing a higher voltage under load. This will be to compensate for the resistance of the cables.

The motherboard reading is unaffected by GPU load, as expected. It drops only a little when the CPU is loaded, going from 84mV lower than what the multimeter read to 149mV lower. Assuming 90% PSU efficiency, there would be 8.85A going to the CPU under load, implying only 7.3mΩ of resistance between the PSU and the 12V sensor on the motherboard, which is pretty good.

Also notice that the motherboard readings appear to be quantised. There's a 56mV difference between them (and I never saw any readings closer together than that), so maybe that's the smallest difference it can measure, in which case the above calculation could be off by quite a bit.
 
Associate
OP
Joined
10 Dec 2007
Posts
1,677
Damn that's some great testing, couldn't ask for better especially in my scenario.

I'm a noob when it comes to power stuff but safe to say the motherboard sensor is always lower than what is actually being fed? 12.301 to 12.152 seems like a pretty stark difference but you know I hope it's accurate. If I put a similar load on my 5950x it goes to 11.812v sometimes. If I added the difference you experienced then the rail would actually be 11.961v which would be more than reasonable.
 
Associate
Joined
19 Jul 2015
Posts
483
...I'm a noob when it comes to power stuff but safe to say the motherboard sensor is always lower than what is actually being fed?..
It's likely that all the 12V readings from a particular motherboard will be off by a roughly constant percentage.

...12.301 to 12.152 seems like a pretty stark difference..
It seems like a lot because we're only interested in the difference between the reading and exactly 12.000V, but it's only 1.2% difference between the two readings. The accuracy is affected by the tolerance of parts such as resistors, which are 1% as standard. Other components can have much wider tolerance. All the errors multiply together, so 1.2% overall is decent.
 
Associate
OP
Joined
10 Dec 2007
Posts
1,677
@Mr Evil Two things.

1: Your sig link still takes you to .eu so it doesn't work till you manually change to .com
2: My brother has an RM850x that I know is good. He's had it for about a year and never had a problem so I persuaded him to give it to me. My own science experiment. :p

I never did extensive testing like before but I'm basically getting the exact same voltage numbers I'm accustomed to seeing in hwinfo64 with the RM1000x. The only slight improvement was the 5v rail. Instead of 4.960v it actually showed 5v sometimes but mostly the standard 4.960v. The motherboard sensors definitely are likely being heavily quantised as you said before. It's almost inconceivable I'd get 2 different power supplies that literally give me the exact same sensor values. I bet it's just something like "if you detect a voltage around 12.0xx display 11.928". If voltage drops by this certain percentage then display "11.928". Repeat for 11.872 etc.

Anyway I think I'm happy now. Learned a decent amount about power supplies (well what to look out for anyway).

Edit: Now I think about it I've literally never seen a value on the 5v rail besides 4.960v and 5.000v. Again good evidence of heavy quantisation happening?
 
Associate
Joined
19 Jul 2015
Posts
483
1: Your sig link still takes you to .eu so it doesn't work till you manually change to .com
Thanks. It's been a massive pain changing everything over after I lost the .eu domain.

To work out if the quantisation we've seen is what it should be, I used HWiNFO64 again to see that our motherboard uses a Nuvoton NCT6798D chip to monitor voltages (among other things). I then looked up the datasheet for that chip, which says it has an 8bit ADC for measuring voltages and temperatures etc. An 8bit ADC means only 256 voltages can be measured, so a 56mV resolution is about right. A higher resolution would be nice, but 16bit ADCs cost more.
 
Back
Top Bottom