AMD Instinct Mi50 Beeping constantly

[Sidiox]

I have a server with a pair of AMD Instinct Mi50s I use for local LLMs. However recently one of the GPUs started beeping and I can find absolutely no reference to this behaviour online (only gaming cards that have it via software).
As soon as the server POSTs one of the GPUs endlessly beeps. It keeps beeping during the OS load and during use. I can still see the GPU in rocm-smi and it seems to run models just fine, the beeping is very annoying.
It happens as soon as I plug it in, no matter the PCIe slot. I’ve reseated it and swapped slots and swapped power cables.

There is no error as far as I can tell either at dmesg or in the BIOS/IPMI.

Both cards run the exact same vbios 113-D1631400-X11 and thermally they are fine as well, well below 40C when idle.

I’m contemplating just desoldering the beeper at this point.

Additionally, the card has a GPIO injection switch a little hardware switch on the side for which I can’t find much information either

[Nuno]

Hi, @Sidiox !
I also have 3 of those GPUs running on my server and, on the first installation, they also beeped like hell.
It’s related to insufficient power supplied to it.
Check your PSU with a multimeter.

[Sidiox]

I don’t think it is the PSUs, I’ve been running the two GPUs in the server for a few months and it never beeped. Additionally, only one of the two GPU beeps, even if its the only one in the system.
Since it works fine temps wise I’ve opted to just remove the beeper from the GPU. They are second hand already so I think its fine. The beeper fell apart as soon as I heated its pads a bit so maybe it was already bad itself? Unsure.

[darkmann]

Well, there’s no putting it back on, now