8 sheets of 3090, using the nvidia-smi command during operation, only 4 sheets of 3090 are left

[346283191]

Information about my system

my question
i have 8 sheets of 3090, using the nvidia-smi command during operation, only 4 sheets of 3090 are left.

nvidia-bug-report.log.gz (981.2 KB)

How can I solve it, thank you very much

[346283191]

Can someone help me look at the problem

[generix]

Resource conflicts:

[    3.705241] pci 0000:56:00.0: BAR 1: no space for [mem size 0x10000000 64bit pref]
[    3.705242] pci 0000:56:00.0: BAR 1: trying firmware assignment [mem 0x7ffa0000000-0x7ffafffffff 64bit pref]
[    3.705243] pci 0000:56:00.0: BAR 1: [mem 0x7ffa0000000-0x7ffafffffff 64bit pref] conflicts with PCI Bus 0000:4a [mem 0x70000000000-0x7ffffffffff window]
[    3.705244] pci 0000:56:00.0: BAR 1: failed to assign [mem size 0x10000000 64bit pref]

Please check for a bios update and try either pci=realloc or pci=realloc=off kernel parameter.

[346283191]

Thank you very much. I’ll try it

[generix]

Additional info: all three non-working gpus are connected to the same pci bridge 0000:4a, so this is quite clearly a bios bug. If no update is available, you should in any case contact the vendor.

[346283191]

It was confirmed that we inserted the 10 Gigabit network card of Intel into the slot of PLX chip of sys-420gp-tnr of supermicro, resulting in a conflict. When we inserted the network card into the card through CPU, the problem was solved