Rack Servers

[ezmaass]

So, I first struggled to get the K80 powered.  I found the proper cable for the NVIDIA as it appears the pinning may be specific.  Prior, the system would fail to boot and report an over-current situation.  With the proper power cable for the card, the system now boots.

My problem NOW is that Windows Server 2016 just doesn't appear to see the card at all.  It doesn't show up in Device Manager, and when attempting to install the NVIDIA drivers, the installer immediately fails and says it can't find compatible hardware.  There's also no other hardware in Device Manager that would otherwise be the card just improperly classified, etc.  It's simply not being seen.

I've searched in the BIOS but I didn't see anything there that looked suspicious.  I've also enabled hardware inventory updates, but I don't think that has any impact here, either.  Any ideas?  Am I missing something that needs to be enabled?

I've installed the card on Riser 3 if that's meaningful.

Thanks!

[DELL-Stefan R]

Hey ezmaass,

saw your PM, nice to see that you found the correct cable.

So what about the PSUs are there 2 1100W PSUs in this system? Like it's mentioned here (https://dell.to/3fDd7Ci).

NOTE: When using the system with the Nvidia K80 GPU card, ensure that you install both PSUs with a minimum of 1100 W each and set the PSU configuration to non-redundant mode.

Best regards,
Stefan

[ezmaass]

Hi Stefan, Yes, the system has two 1100w PSUs and they are, indeed, configured in non-redundant mode. I suspect that the system has the proper power since it was previously reporting an over-current (error 36) situation when the cable wasn't correct. Now, it boots fine, but the card doesn't appear to be visible to the OS - leading me to believe something is still wrong at the hardware level. Are there any diagnostics that I might look at to at least confirm or deny whether the system can see something plugged into the PCIe slot? Any issues with using Riser 3 instead of Riser 2? I selected Riser 3 as it looked better for airflow, but I understand both should work. Any BIOS settings that I need to look at? I've read the R730 documentation on installing a GPU, and nothing is mentioned beyond just the elementary basics of and aforementioned power supply requirements. I suppose, at this point, I'm trying to rule out whether it's something at the hardware level (where the machine literally isn't talking to the card or something), or whether it's an OS issue (Win 2016 Server) where the OS needs to be adjusted to see the card. I've been leaning towards something with hardware/BIOS simply because I'd expect the OS would still discover the device in Device Manager, even if incorrectly classified, no drivers, etc. Thanks for the help and any insights!

[ezmaass · Accepted solution]

I was just able to solve this.

The solution was silly, but it worked - I simply reset the card in the PCIe slot and also flipped around the power cable.  I expect the cable had nothing to do with it, but I'm including it here anyway since it technically was a variable that changed.  The card looked and felt firmly seated, but nonetheless perhaps it wasn't fully?

In any case, when the machine booted, I noticed that it booted differently - fans were at full speed for a good 10 or 15 seconds during the boot sequence which never happens.  Once into Windows, the Device Manager immediately showed two "3D Video Device" (the K80 has two GPUs on it).  The NVIDIA installer then worked perfectly.

So, net result - it was likely that the card was just not fully seated.  Thanks for the help nonetheless.

[DELL-Stefan R]

Hi ezmaass,

I'm so happy to read that everything is working fine now

[bw8783]

Exact same issue happened to 2 of my gpu servers. swapped the cable around the other way and reseated cards and it booted fine, and perfect since