Help with adding second GPU

cl_2077 · 2020-06-02 17:57:57

I'm planning on adding a second Quadro GPU to my system and am doing some prep. The GPUs will be identical. I'm also using the latest NVIDIA driver and am using the GNOME desktop. I'm currently using the GPUs to support my deep learning work.

I read the Arch Wiki entry at https://wiki.archlinux.org/index.php/NV … abling_SLI as well as https://wiki.archlinux.org/index.php/NVIDIA. If I'm to understand the workflow is as follows:

1. Install new gpu
2. Restart my machine and boot into runlevel 3
3. Use nvidia-settings to update xorg.conf
4. Reboot

My understanding is that this process is the same whether are not I use NVLINK, with the exception that if I use NVLINK I need to also:

1. Run lspci | grep VGA to get my first GPU's PCI Bus ID
2. nvidia-xconfig --busid=[first GPU bus ID] --sli=AA
3. Verify that SLI is enabled using, nvidia-settings -q all | grep SLIMode
4. Reboot

Is there anything else I need to consider? Also, and apologies if this is off topic for the forum, has anyone benefited from using NVLINK for deep learning? I get the faster GPU communication to speed up distributed training. But I'm still fuzzy on using Unified Memory to support larger batch sizes [for pytorch in my case]. Again, my apologies for going off-topic.

Lone_Wolf · 2020-06-04 15:30:53

nvidia-settings & nvidia-xconfig don't have a good reputation on archlinux (they tend to create problems instead of solving them) , you'd probably be better off making the necessary changes manually .

archlinux / systemd only have runlevels for legacy use, you probably want to boot to multi-user.target , see https://wiki.archlinux.org/index.php/Sy … _boot_into .

Troubleshooting nvidia setups can be very hard, and a sli setup with 2 quadro cards isn't exactly common .
How much experience with archlinux do you have ?

cl_2077 · 2020-06-04 22:24:32

Thank you for the link! I read through it just now and it shored up my misunderstanding. I also did searched the wiki on nvidia-settings & nvidia-xconfig problems. I see what you mean.

Regarding your question, I'd say that I'm still a relative novice. I've been using Arch since Christmas, installing and maintaining it on three of my machines. Before that I've used Fedora for about a year. I'm comfortable with Basic-Intermediate console commands and can write some simple bash scripts. I regularly go to the Arch wiki and am working my way through Robert Love's, "Linux Kernel Development" to get smarter. But honestly, I'm still a work in progress when it comes to linux skills.

What would you recommend?

Lone_Wolf · 2020-06-07 15:25:30

Simplify your current setup :

switch from display manager to manually starting X using startx / xinitrc , https://wiki.archlinux.org/index.php/Xinit
Replace the autogenerated xorg.conf with a custom one in xorg.conf.d that changes only what's needed.

It's ok to ask for help with that, but focus on understanding what happens.
If you're succesfull in making these changes your understanding of your system will be improved a lot and you'll have hands-on experience.

Since you'll be changing/adding hardware and may very well face a temporarily unbootable system, prepare .
Make sure you have another system ready that works reliably, look into backup methods, verify you have a recent archlinux installation iso that works etc.

Documenting what you do and what changes you make can be very helpful.

What your trying to do doesn't require genius level skills, but you're the only one that can determine if you're up to the task.

Arch Linux

#1 2020-06-02 17:57:57

Help with adding second GPU

#2 2020-06-04 15:30:53

Re: Help with adding second GPU

#3 2020-06-04 22:24:32

Re: Help with adding second GPU

#4 2020-06-07 15:25:30

Re: Help with adding second GPU

Board footer