PC (sometimes) shuts down a split second after being powered on

[PPR]

Good time of day everyone,

I've been seeing some strange behavior with my machine (configuration is detailed below) recently, and was hoping somebody could help. A couple of days ago, I tried to fit a 10Gbit NIC into the machine, and while the card worked -- it would sometimes (but not always) so happen that when starting the computer, it would power on for a split second and then go back off (the power LED that's on the case would remain on, for some reason, as well as the motherboard's RGB lighting). The only way then to start the machine would be to power-cycle the PSU. This would happen once in every ~5 times in which I powered on the machine.

Switching the newly-installed NIC into a different PCIe slot made no difference; and the same is true for replacing said NIC with an identical one. Taking the NIC out made the machine start correctly almost every single time -- apart from one time in which I was able to reproduce the issue without said NIC being installed (by power-cycling the machine a couple dozen times, in succession). This tells me that this is not a mere incompatibility between the NIC and some other component as the issue still exists, albeit it is much more rare.

The configuration is as follows:
AMD Ryzen 9 5950X
ASUS Crosshair VIII Dark Hero (BIOS version is 4602, the latest one)
ASUS ROG Thor 1200P PSU
128GB of RAM at 3200MHz (four sticks)
ASUS TUF 4090 OC (in a 600W configuration, using a CableMod cable)
Creative Sound Blaster AE-9
The system has two NVMe drives installed, as well as one SATA SSD.
4 case fans and two CPU fans (mounted on a Noctua NH-D15 chromax.black)

The system is indeed loaded with components, some of which may consume significant amounts of power -- with the sound card itself requiring its own PCIe 6-pin power connector (it consumes 75W through the PCIe slot and another 75W through the cable). Still, with all that hardware -- but without the added NIC -- it took a lot of tries to reproduce this. Until a couple of months ago, that same PC had a 3090 card installed and no discrete sound card, and I hadn't seen anything like this in two years of use.

What I suspect is that the initial spike (or rather, surge) that happens during power on, when all said components are installed, would sometimes trip the PSU's protection circuitry (presumably, having to power-cycle the PSU indicates that said protection circuitry did indeed kick in), resulting in the aforesaid -- and that this would happen considerably more often with the added initial load that is incurred by the NIC. Still, the NIC's peak power consumption is a mere 6W, and as such I cannot understand why it would make such a big difference in terms of recurrence. Moreover, it is unclear to me why the power LED remains on after a failed power on attempt.

It is also noteworthy that I've installed an identical NIC into my second machine (a 5900X with a 3090 card and an 850W Antec PSU) -- and on that machine I am not seeing anything like that.

Does anyone have any ideas? Could my suspicion be correct? Would enabling ErP S5 mode (counterintuitively so) be a good idea to try and work around this issue?

Thanks a lot!

 

[michalrz]

Interesting issue, I've never seen that one.

The only thing that comes to mind is - significant electrical potential difference between the PC in question and the rest of the network.
As is, we don't know how your network is setup, how far away the router/switch is, is it hooked up to the same power source at your home, how much wiring is there in total.

Ethernet should be and usually is immune to such issues. If there's a surge on start-up with the NIC connected, it's because there's a tiny bit of parasitic capacitance in the entire network. But - again - these usually don't manifest like that.
Unless it's actually the switch/router that is having power issues, and there is an unwanted, constant or momentary current flows, spikes, etc.

What I'd do is get a small test-network going near your PC. Some random switch/router, maybe a laptop or something connected to it, and see if the issue persists.

It's a freaky issue for sure, and my theory is just one idea.
Welcome to the forums btw!

Edit: oh, if the issue persists without any wires connected to the NIC - pls disregard

 

[PPR]

Interesting issue, I've never seen that one.

The only thing that comes to mind is - significant electrical potential difference between the PC in question and the rest of the network.
As is, we don't know how your network is setup, how far away the router/switch is, is it hooked up to the same power source at your home, how much wiring is there in total.

Ethernet should be and usually is immune to such issues. If there's a surge on start-up with the NIC connected, it's because there's a tiny bit of parasitic capacitance in the entire network. But - again - these usually don't manifest like that.
Unless it's actually the switch/router that is having power issues, and there is an unwanted, constant or momentary current flows, spikes, etc.

What I'd do is get a small test-network going near your PC. Some random switch/router, maybe a laptop or something connected to it, and see if the issue persists.

It's a freaky issue for sure, and my theory is just one idea.
Welcome to the forums btw!

Edit: oh, if the issue persists without any wires connected to the NIC - pls disregard

Thanks for the reply and warm welcome!

The network is indeed somewhat complicated for a home (several PCs, TVs, game consoles and other devices, and 7 switches in total, at different points). The theory is interesting, however I was able to reproduce the issue *once* without the NIC installed (by power-cycling the machine a couple dozen times, in succession). Moreover, said computer has been connected to the same network for over two years (through a different, on-board NIC), and there was nothing like this. Something is telling me that the initial spike trips the PSUs protection circuit for some reason -- which happens much more often when the NIC is installed.

Could it be that my combination of hardware (power hungry GPU, sound card and a multitude of other components) is a little bit too much for a 1200W 80Plus Platinum-rated PSU to bear, in terms of the initial surge?

 

[michalrz]

Could it be that my combination of hardware (power hungry GPU, sound card and a multitude of other components) is a little bit too much for a 1200W 80Plus Platinum-rated PSU to bear, in terms of the initial surge?

That is always a possibility, even with a beefy power supply.
It might be that the new NIC is simply the straw that is breaking the poor camel's back.

It's actually good that you have a way to get it to happen more often. Maybe - put the NIC back in, but try removing some other component from the PC. For example the soundcard, maybe an USB device or two.
If by doing so (lessening the load on the PSU) you'll be able to keep the NIC running properly, there's your answer.

 

[PPR]

That is always a possibility, even with a beefy power supply.
It might be that the new NIC is simply the straw that is breaking the poor camel's back.

It's actually good that you have a way to get it to happen more often. Maybe - put the NIC back in, but try removing some other component from the PC. For example the soundcard, maybe an USB device or two.
If by doing so (lessening the load on the PSU) you'll be able to keep the NIC running properly, there's your answer.

I will do that, thanks!

The PC is also connected to a surge protector, and not straight to the wall. Still, the device itself does kick in, so I doubt that it has anything to do with this.
This is so strange, as even during heavy load the PC never exceeds ~800W in power consumption. Does it make sense for the initial spike to be so significant?


Do you think that by experimenting with this I might risk frying some components? How safe would doing so be?

 

[michalrz]

Does it make sense for the initial spike to be so significant?

Yes, most definitely. A PC is a resistive load, capacitive load (capacitors suck in MASSIVE amounts of power when discharged and suddenly powered on), and even some inductive loads for good measure (fans).

Hard to put a number on that, but the inrush can be way, WAY beyond 800W for a split second. Your PSU is supposed to handle that. Now, if it was a circuit breaker that'd pop, it'd be a different issue altogether.

Do you think that by experimenting with this I might risk frying some components? How safe would doing so be?

I think that a quick test involving the idea of trying to re-balance the load on the power supply to rule it out is quite safe to do, as long as you don't try to add more loads. That is the theory, anyway.

If it is the power supply that isn't cutting it, the longer you have it, the higher the risk of a more severe failure.

If during the re-balancing you'll notice that it's actually a particular component (a completely different one) that is causing the issue, you'd want to know that as well - the longer something bad power-related is happening, the higher the risk.

 

[michalrz]

A good way to reduce risk is also to wait a bit until someone else chimes in in case I'm way off with my ideas

 

[PPR]

Thanks for all the input! Let’s see if anyone else has any experience with something like this.

I was considering getting a native ATX 3.0 PSU anyway, so this might as well be a good opportunity to do so.

 

[Furious_Styles]

Could easily be a mobo issue you're having. It sounds like the problem occurs when you're putting stuff into the PCI-e slots. With no problems encountered prior to the soundcard and NIC installs I wouldn't really suspect the PSU, both of those do not take lots of power.

 

[stormy1]

pci-e allocation issue? check the docs for shared lanes.

 

[StoleMyOwnCar]

Does it happen the exact same way every time? Personally, I think since you have this issue even WITHOUT the card hooked (though more rare) up, we can sort of rule out the card being the primary issue. It would have to never happen without the card hooked up. If it happens even once without it hooked up, you have some instability somewhere that's causing this. I also kind of find it hard to believe that a 1200W PSU is struggling with that setup at all, but that's just me. One quick note, I think if you use the slot right below the GPU for your network card or other devices, I think it's going to put the GPU into X8 mode (?). PCIE4 X8 is basically PCIE3 X16. That's kind of okay, but for a 4090 it's pushing up to the limit of 3.0 X16.

Have you tried running memtest, too? You have an awful lot of RAM on an AM4 platform. I have run 64 GB (4 sticks) on an X570S Carbon (same CPU) at 3200 without issue for well over a year, but I don't know about faster speeds with twice the capacity.

Edit: Rereading this... I think my money would be on the PSU, followed by sound card. Maybe motherboard, but unlikely?
Let's just point out what happened between each of your swaps:
1. 3090 (no sound card): no issues.
2. 3090 + SC (sound card): ??? No idea
3. 4090 + SC: Issues, but rare.
-- Differences between 1 and 2:
GPU draws slightly more power, and I'm assuming it uses an 4x8pin -> 12vhpwr connector. Sound card is suddenly completely saturating the maximum possible output of a PCIE slot and also using an extra 6 pin 12v connector, although it uses a different amount from the GPU...
4. 4090 + SC + NIC: more frequent issues.
--- Differences:
You hooked up an extra slot worth of PCIE powered device.

If we look up your PSU on the PSU cultists list, I'm assuming it's this model:

• ASUS | ROG THOR 1200W [non-II] [1][8][11]

There are some afternote items attached to it:
[1] Units experiencing tripping issues with high transient power draw GPUs like AMD Vega, 6900 XT and Nvidia RTX3080/3090. Generally fixed in newer batches.
[8] Seasonic PRIME based units experience shutdowns with RTX3080/3090 (and possibly RX6900 XT) GPUs. The cause is not the OCP tripping but a PSU design flaw as evident by the PSU not latching off on shutdown and 1000W+ models being affected too. Doesn’t manifest in 100% cases as it’s also dependent on motherboard model and GPU OC. Seasonic provides a replacement 24-pin ATX cable to fix this via support. Appears to be fixed post 2021 although there’s no official confirmation from Seasonic.

Your 4090 is basically a very overclocked 3090, wattage wise (?), and already causes rare instability. Then, you introduced more instability by adding a PCIE device to your motherboard. You said you've been running this build for at least 2 years, so this PSU is that old. So with some reasoning, maybe this is the issue.

 

[GoldenTiger]

Your 4090 is basically a very overclocked 3090, wattage wise (?)

4xxx cards largely eliminated the huge transient spikes of the 3xxx series. Old thread though, so whatever .

 

[StoleMyOwnCar]

4xxx cards largely eliminated the huge transient spikes of the 3xxx series. Old thread though, so whatever .

The transient spike was point #1 on the PSU cultist list, not point #8, which doesn't seem to be related. It's a design flaw in the PSU. Since their PSU is old, it might just be faulty, and in fact I think it's pretty likely since it's related to more cards using the power from the PCIE slot.

That said, yeah you're right, this thread is old and I didn't notice because it had a recent poster.