HP Proliant servers before about G6 (G6 & G7 support GPGPUs like the Tesla) have a very long boot sequence (minutes) which often confuses third-party add-in cards–which usually expect a hardware reset signal either within one minute or so of boot-up or never. I have seen this before on SATA cards (and took to repurposeing a compatible SAS interface card for SATA). HP supports no video cards on their earlier servers. I just threw in a the nearest available video card (a PNY GeForce 7600GS 512 PCI-e-16) into my server; it has trash on the screen and never settles down, though it does try to display the BIOS boot-up and desktop, yet the results are unusable. I expect that any other mid-range Nvidia PCI-e card would do the same. The right fix for this is probably a modified card bios. An easier fix might be a memory poke or subroutine call to the right absolute hardware location or kernel interface during Linux boot-up to belatedly force a reset to the video card, e.g. from /etc/init.d before Linux goes to runlevel 5, or perhaps a Linux kernel command line boot cheat code. Does anyone know the driver interface to Nvidia mid-range cards to force a reset, e.g. from a ten line C program?
This sounds like a very unfortunate system BIOS bug, and updating your BIOS to one that works sounds like the best bet. You can try to reset the GPU by running “nvidia-smi -i 0 -r” but I’m not particularly hopeful that it will work.