Driver cannot initialize GPU (RTX PRO 4000 Blackwell) behind bridge PLX PEX8747 (Supermicro System)

NVIDIA Open GPU Kernel Modules Version

595.84 also tried 595.71.05

Please confirm this issue does not happen with the proprietary driver (of the same version). This issue tracker is only for bugs specific to the open kernel driver.

• I confirm that this does not happen with the proprietary driver package.

Operating System and Version

Gentoo Linux (Rolling Release)

Kernel Release

Linux hostname 6.18.35-gentoo-r1 #2 SMP PREEMPT_DYNAMIC Wed Jul 1 13:09:43 CEST 2026 x86_64 Intel(R) Xeon(R) CPU E5-2630 v4 @ 2.20GHz GenuineIntel GNU/Linux

Please confirm you are running a stable release kernel (e.g. not a -rc). We do not accept bug reports for unreleased kernels.

• I am running on a stable kernel release.

Hardware: GPU

NVIDIA RTX PRO 4000 Blackwell

Describe the bug

It seems the driver cannot initialize GPUs behind a PCIe bridge PLX PEX8747 in a Supermicro SYS-2028GR-TR (yeah rather old system, but should be fine for the PRO 4000).

[    8.606516] NVRM: GPU1 Plx_Pex8747_GetBar0: Device has no BAR0!
[    9.889042] NVRM: GPU2 Plx_Pex8747_GetBar0: Device has no BAR0!
[   11.281680] NVRM: GPU at PCI:0000:07:00: GPU-2bead814-a1c4-14fa-56a5-499a8d4c17d4
[   11.281685] NVRM: Xid (PCI:0000:07:00): 79, GPU has fallen off the bus.
[   11.281689] NVRM: GPU 0000:07:00.0: GPU has fallen off the bus.
[   11.281696] NVRM: GPU2 krcRcAndNotifyAllChannels_IMPL: RC all channels for critical error 79.

Rest of the output is in the attached dmesg.txt
A GPU (Blackwell RTX PRO 4000) attached directly to the CPU just works fine.
Turing Cards (RTX 4000-6000) also work(ed) behind the bridge. (some removed for the newer ones)
For the Blackwell GPUs I had to upgrade the driver to the open kernel module but the RTX 6000 still works behind the PLX bridge today.

My guess is, that the driver doesn't know how to handle the memory BAR of a device behind a bridge and instead tries to access the bridge directly?

If you need more logs, I'm happy to provide.

dmesg.txt
lshw_bus.txt

To Reproduce

Just launch the server with the cards attached.

Bug Incidence

Always

nvidia-bug-report.log.gz

nvidia-bug-report.log.gz

More Info

It would be a pity to not use these cards (3 of 4) or to have to buy a server during these times
for 2-4x times the price of the cards itself.

So any help is very appreciated.
Big thanks in advance!

Start with the attached dmesg.txt, lshw_bus.txt, and nvidia-bug-report.log.gz, focusing on the PLX PEX8747, missing BAR0 messages, and Xid 79. Compare the initialization evidence for the Blackwell GPUs behind the bridge with the directly attached GPU and the working Turing cards. Done means the Blackwell GPUs initialize reliably behind the bridge without the BAR0 errors or GPU-fallen-off-the-bus failure.