RM850X shuts down from 3090 FE power spikes

So I have a 3090 FE powered with 2 separate "pig tail" PCI-E cables from a RM850X (rest of build specs in profile).

 

I occasionally do some machine learning that creates power spikes. Specifically, the GPU can load back-and-forth from ~20W to ~340W. Unfortunately, these spikes lead to an immediate computer shut down.

 

While it is a hungry system, the build is otherwise stable even under heavy steady load (at most I get to 600W total) - however any large GPU spikes as mentioned crashes the entire system, even if total load is way below 850W.

 

No shutdowns happen when I use a 1KW PSU, which means the RM850X is the culprit here. Possibly OCP is being triggered.

 

Is an RM850X supposed to handle those ~300W spikes and therefore I have a defective unit? Or will all RM850X have OCP triggered because of those spikes?

An RM850x should most certainly be able to support that card.

 

What are some of your other components? CPU? Overclock? RGB fans? Etc.

 

The PSU supports the entire build under "stable" constant load, even at max (~600W total) - it crashes when there's a large power spike within that amount.

 

CPU is 10900K, RGB comes from 4 Phanteks Halos Lux but they're relatively limited in power. Haven't overclocked anything yet until I figure out this problem.

 

Even under 100% load, I'm getting under 600W which the PSU handles fine - the problem I have is that it crashes when the GPU "jumps" 300W in usage (even if it's all still well below 850W for the total system).

 

Sounds like you are triggering the OCP of the PSU due to those power spikes.

 

That makes sense.

 

But will OCP happen on all RM850x with these 300W spikes? Or is it just this unit that is a bit too sensitive?

 

In other words - should I RMA or change PSU model?

Machine learning programs that trigger a GPU spike when booting up the models on the card.

 

You can see the large GPU power spikes that happen (blue line):

 

 

The RM850x crashes as the large 300W GPU spikes happen. (I was able to take this screenshot when borrowing a 1KW PSU.)

 

And I've used a hardware watt meter + most major software with monitoring tool - NVIDIA System Management, GPU-Z, HWiNFO, Afterburner, OCCT, Precision 1, etc. Same results everywhere.

 

It seems OCP does get triggered, which is why at this point I'm trying to know if the RM850x model should be able to handle ~300W spikes and not get OCP (i.e. I should RMA a defective unit) - or if I should move to a different model because any RM850x will trigger OCP for these spikes.

If the gpu gives voltage spikes then its not the PSU at fault. seems 3090 has internal power failure. test the gpu in another pc or RMA it

if the GPU fails the PC doesn't just turn off. you would get a crash and a black screen.

Looks like the PSU can't deal with those sharp transients.. but the strange thing is it should work.

 

just out of curiosity, how old is this power supply?

Looking at that power chart suggest even my RM650 (2019) would handle it. I tested it and it make it just over 700W before it tripped up. Mind you it was very hot and the fan was maxed.

 

Makes me speculate that there is something wrong with the PSU so I wonder if the 3 single connector power cables might work better.