H100 with SXM5 PCIE adapter issues

[theblink]

Good afternoon,

I am wondering if anyone has an h100 with the integrated voltage adapter pcie card to work with 12V power. Currently I am not able to get the card properly booted due to a blue 54V led blinking on my SXM5 adapter. Did anyone happen to come across this and know how to fix it? My bios sometimes sees the card as a H100 but it mostly says:

PCIe Slot 1: Enabled, Gen3, Full Length
Device: NVIDIA Corporation, 16 lanes, Health Status: Ok


The GPU is throwing out this error.

NVRM: Xid (PCI:0000:41:00): 143, Error status 0x65 while polling for FSP boot complete, 0x0, 0xbadf5108, 0xbadf5108, 0xbadf5108, 0xbadf5108
NVRM: GPU 0000:41:00.0: RmInitAdapter failed! (0x62:0x65:2790)
NVRM: GPU 0000:41:00.0: rm_init_adapter failed, device minor number 0

 

[theblink]

I was able to troubleshoot this this and resolve my issue. My VBIOS on my GPU was too old for the latest OS/Kernal. Big shoutout to @MildHotSauce for helping me figure this out.

 

[gsrcrxsi]

were you able to update the BIOS? if so, how?
or did you move to an older OS?
or give up?

 

[theblink]

I did not update the VBIOS due to it being a QS. I am not sure if the normal SXM5 H100 VBIOS is compatible and I do not want to brick the card.

 

[gsrcrxsi]

So what was your solution?

 

[theblink]

I had to use Ubuntu 20.04 LTS. I had to first turn on the GPU's external power supply wait 30 seconds then boot up my server. Once your in install CUDA 12.2 and NVIDIA Driver 535.183.01 (Tesla branch).

 

[gsrcrxsi]

i can understand needing to use the older driver. but i'm surprised you had to use the older OS also.

glad you got it working though.

 

[theblink]

The older driver is not supported in the newer OS. It failed to install 20.04 was the only version I got the driver to work in and detect the card.