PC Graphics

[Magnum1903]

Hi everyone,

I have 4x AMD Instinct Mi100 GPUs and a Gigabyte G292-z20 server and the cards are not recognized. The PCIe slots show up as empty in the BMC.
The server is on latest firmware and latest BIOS.
I have above 4g decoding enabled and uefi set in the BIOS.

I’ve rebased the cards in different risers and measured that the GPU power plugs have the right voltages.

I've tested the PCIe slots with an Nvidia card, which is detected just fine.

Are these cards not supported in that server or am I missing some BIOS setting?

Any help and suggestions are much appreciated.

[fsadough]

Gigabyte G292-z20 is certified with NVIDIA GPUs. Please contact Gigabyte for support. 
G292-Z20 (rev. A00) | GPU Servers - GIGABYTE Germany

[cdrfvb]

Hi fsadough,

actually on the website you referenced here there is also a sticker "AMD Instinct", so I assume there should also be compatibility certification with AMD. And you reference exactly this server at least in the datasheet of the Instinct MI50. Has compatibility stopped after MI50? Because with MI50 it works for me with this server.

[fsadough]

Yes, MI-50 is a compatible with this server but not MI-100. The is also a fine difference between CERTIFIED and COMPATIBLE

[cdrfvb]

Ok, thank you for the info... That saves me from the mistake of buying MI100s as upgrade for my server. Does that also apply for all other newer cards? Is there anything that can be done to make MI100 or any other cards apart from the MI50 work? A bit sad, as the MI50 will drop out of ROCm support soon.

About certified or compatible, I guess you are right, but then, I assume the OP just wanted to know how to make it run or if it is at all possible to run. My language was not precise, I admit.