Tesla V100 16G PCIe on Dell Precision 7920T

[Donalaren]

Hello, I am in urgent need of your assistance. I am attempting to use a Tesla V100 16G PCIe graphics card on a Dell Precision 7920T.

System
Motherboard: Dell Precision 7920T
Card: Tesla V100 16G PCIe
System: Ubuntu 22.04
RAM: 64G

• Memory Map IO above 4GB is enabled.
• There is no CSM in BIOS of Dell 7920T.
• BIOS version is updated (2.50.0).

Description
I first attempted this on Ubuntu 22.04 using @xCuri0 's ReBarState. Followed this method, after building and running RebarState, set the value to 32. But this approach proved ineffective. After rebooting and reopening the RebarState, I can see the Current ReBarState 32/Unlimited. nvidia-smi still can't see my Tesla V100 ( lspci | grep -i nvidia correctly identified the GPU model 0000:73:00.0 3D controller: NVIDIA Corporation GV100GL [Tesla V100 PCIe 16GB] (rev a1)).
Through dmesg I can see:

[   45.605811] nvidia 0000:73:00.0: enabling device (0140 -> 0142)
[   45.606250] NVRM: This PCI I/O region assigned to your NVIDIA device is invalid:
               NVRM: BAR1 is 0M @ 0x0 (PCI:0000:73:00.0)
[   45.606542] nvidia: probe of 0000:73:00.0 failed with error -1

This is consistent with the situation without ReBarState. I have only followed this method to build and run ReBarState on the Ubuntu system without making any modifications to the BIOS. My actions were as follows:

git clone https://github.com/xCuri0/ReBarUEFI.git
cd ReBarUEFI/ReBarState
mkdir build
cd build
cmake ..
make
sudo ./ReBarState

So I suspect ReBarState may not be functioning. How to verify if the RebarState is working in this scenario?
Note that I'm not suspecting the software itself; I believe my execution may be incorrect. Any suggestions would be most gratefully received.

[xCuri0]

@Donalaren Tesla V100 and all NVIDIA datacenter GPUs don't use Resizable BAR they just has a large BAR by default, the issue here is likely with your BIOS configuration.

Dell 7920T should be recent enough to not have the weird issues on older motherboards requiring UEFIPatches.

All ReBarState does is read/write the ReBarState EFI variable which you could also read yourself but it's going to give the same result as what you're seeing.

EDIT: Seems like you aren't running ReBarUEFI modded BIOS even, in that case ReBarState does absolutely nothing.

[Donalaren]

Thank you very much for your reply; it has given me renewed hope.

Do you believe RebarState cannot resolve this compatibility issue with the Tesla V100? Yet I've observed on other websites that they've addressed Tesla GPU problems (including Tesla P100 and Tesla P40). Moreover, they all mention switching the GPU mode to graphics mode via NVFlash. However, NVFlash currently fails to detect my Tesla GPU, I suspect this stems from the absence of Rebar, as per this webpage. Regardless, Rebar is essential, so I intend to give it a try.

Thus, we return to the RebarState software. I acknowledge it is an extremely useful tool—thank you for sharing it—but I seem to have yet to grasp how to utilise it properly. Clearly, my understanding of this statement was flawed. I mistakenly believed that , I only needed to run ReBarState without modifying BIOS on the Ubuntu system.
Right, I understand now. Even if I wish to implement Resizable BAR on a Linux system, I still need to flash the modified BIOS using hardware BIOS flasher such as CH341A. However, I encountered difficulties when carrying out the first step Adding FFS module. Dell's BIOS file is in .exe format, which UEFITool cannot recognise. I've observed that some Dell workstations have successfully implemented Rebar in List of tested motherboards. How might one resolve this issue about the EXE format?

[xCuri0]

Tesla V100 isn't a rebar GPU it's going to make no difference.