Previously known good GPU crashing under load after migrating to a new system

aporia · September 24, 2019

Hi all.

I'm new here and didn't see any guidelines or faq answers relevant to my question, sorry in advance if I missed them.

I'm transitioning a known good RX 480 from an old 2500k-based system to a new Ryzen 3600 build. The card previously was working fine, but after installing it in the 3600 it now it gets unstable when drawing ~120 watts. At or above 130 watts it crashes to black screen under a stress test or a solid color under a gaming load (Determined by forcing a lower power limit in Afterburner). When it gets unstable the clock and power consumption sine-wave up and down resulting in lower performance than at 110 watts. This now happens in both systems. Both systems do fine with a venerable HD 5750 I had lying around.

I've been sure to DDU between card swaps and did a fresh install of Win10 on the 3600. The only thing I can think that might have caused it is I mistakenly powered on the system without the 8-pin PCI-E power connector. I powered it off before connecting when I realized my mistake. I google suggests that's no big deal, but maybe it is?

Any ideas on what went wrong or what I can do to fix it? I'm out of ideas except maybe flashing the BIOS, which I've never done before.

BTGbullseye · September 25, 2019

You did plug the power in after forgetting right? If yes, then updating the BIOS is a really good idea, especially if the motherboard is anything but a X570 chipset. (most will ship with a BIOS from before the 3rd Gen CPUs were available)

aporia · September 25, 2019

I did plug in the 8-pin after realizing my mistake (and unplugging the system). And I did update the motherboard BIOS, The BIOS I was referring to is the graphics card's BIOS. Since I last posted I tried a bunch of different old driver versions, so all I can think is that the card is physically damaged or the BIOS got messed up somehow. The card is acting like it either can't handle load current anymore, or it flips out when asked to clock up.

Come to think of it, is there some way to force the card to run at advertised stock voltage and frequency regardless of load? That could maybe test the latter idea.

EDIT: Turns out there is a way, and I found it. Unfortunately it didn't help. The card behaves fine with the GPU and memory clocks forced to stock as long as there is no load, but crashes in the same way under load. It seems the problem has only to do with the amount of current the card draws.

Edited September 25, 2019 by aporia