Using a server PSU with a desktop motherboard... which keeps killing it

[mediacomposerman]

After lots of hard work, we built a rackmounted Threadripper workstation with Thunderbolt connectivity (those were the 3 requirements we aimed to nail). In fact, I had hired a specialist company to help build it for us..... and we've still got problems, Houston.

The mobo is a Gigabyte TRX40 Designare, case is InWin PLG, which came with dual AcBel R1CA2122A PSU's (800W, CRPS with PDB). The integrator already told us to only use 1 PSU at a time (negating the benefit of dual-redundant PSUs) so we have, though I think he misinterpreted the problem.

It's not a "it won't post" problem... the mobo completely and utterly dies, upon plugging in the AC plug. The CMOS reset button light and onboard power button light turn on for about a quarter second, and after that the mobo is utterly dead. Until... you unplug power and let it sit for 15 minutes.

Sometimes we can swap to the other PSU module we have and get back quicker, usually not. And sometimes we have to disconnect all aux PCIe power connections to the cards (GPU, mobo, TB3), then let it sit for 15 minutes, then plug in the AC, then while in standby plug the PCIe's back, and then turn it on.

We've been through that (slow, tedious) troubleshooting cycle ~30 times now. It often dies... and always comes back to life -- and when it's on (or on standby), the computer works happily for days and days on end without a hitch.

I already tried to set the BIOS Power Supply Idle control from Auto to Typical, ErP, setting Power Loading to Enabled, as well as several BIOS resets... no change.
(I had hoped that Power Loading would be the panacea, since its description is: "Enables or disables dummy load. When the power supply is at low load, a self-protection will activate causing it to shutdown or fail. If this occurs, please set to Enabled.")


My best theory is that the PSU/PDB and this mobo are not compatible with each other... for some reason, and the initial current trips some sort of power self-protection or fuse. Happy to hear any insights, I don't really have experience with PDB's.
I'm running out of options... my next step could be to either try a different CRPS PSU (though that wouldn't change the PDB), or try to dismantle the whole PSU module and cram a "normal" SFX PSU in there like a Corsair SF750 (if I'm not 2mm short).

 

[ddaenen1]

Are you sure these PSU's are ATX standard? I just looked at the data sheet and there is no mentioning. Seems to me they are for a specific application only.

 

[RageBone]

if you are able, please measure if you are getting 5V standby from the power-supply in the bugged state.
If you don't measure 5V where you expect 5VSB to be, the PSU likely trips an OCP and stays off.

I had that happen on a X399 giglebyte gaming7*?
In this case, a Mem VRM Phase was dead and directly connected the 5VSB rail to the CPUs Vmem (1.2V for ddr4)
It behaved similar to what you describe, with the exception that there was no reviving it or the CPU.
Which speaks for your case not being such a severe fault.

The other common option is for Motherboards to cut off as many things needing power as possible in standby to for instance, be ERP ready.
That is often implemented through the option to disconnect 5VSB from most components on the board.
On Asus boards, that is tied into the CPU presence detection.
In my Experience, other electrical faults can trigger that disconnect as well.

So a reasonable step would be to make sure that the CPU and Motherboard are or at least can be made fully working reliably.
Test with a different known good PSU and check for rouge standoffs in the case and test outside the case on clean flat and soft surface like a cardboard box.

 

[thunderdolt]

Is 800W enough for that build? Depending on which TR you're running, it could be spiking to 300W or even 350W all by itself. That doesn't leave much headroom for something like a 350-400W 3090 on top of the rest of the system.

Assuming 800W is enough, then it does sound like an issue with the PDB and/or the mobo. Gigabyte isn't exactly known for quality on their consumer stuff. I've personally had one of their TRX40 boards be DOA and then had the replacement go up in smoke a couple of months later (and the 3-month turnaround on that was hilariously unacceptable for a production machine).