Im new to the mining and finaly got my 3 Rigs set up each with 13x Aorus 1080ti - after having troubles with riser usb cables who where not working everything is running so far. But i have problems with the temp - some cards are around 83C° the rest would be ok-ish around 70°C ( Room temp was around 20°C id guess ) but they are not running at maxium Power because of the temp :(. I have never used linux befor and cant find any programs that would allow me to tweak / undervolt or set my fans to 100%.
So any advice? How are you guys keeping your cards cool? Is there some sort of Afterburner for Linux?
Can’t see very well on this pic but, i think cards should have a little bit more space between them and make sure cards heat it’s not vented against each other. Also i think it’s normal cards in the middle to be with a little bit higher temp. I have rig with 6 cards and the 3 cards in the middle are always with higher temp compare to this that are 1st and last in the row.
So on the Ubuntu setup you’ll need to do the following:
If you only want to control power (undervolting or overvolting) then do:
sudo nvidia-smi -pl - where some number is what you want the voltage to be. I generally undervolt mine to 80% power but I run them in the 55-60 degree range only. If you want to see current info on the cards do sudo nvidia-smi and it will show baseline voltage settings. nvidia-smi -pm ENABLED -i X (where X is the instance number of the gpu) will enable persistence mode and make the changes stick across reboots. A quick google of nvidia-smi will help you.
If you want to control individual fan speeds, gpu and memory clocks you’ll have to do this:
Disable the gpumanager - I believe someone already posted you to a link on AskUbuntu. GPUManager is re-writing your xorg.conf at boot so that’s why you keep losing your changes for coolbits
The nvidia-settings utility is what allows you to control fan speeds per gpu, per gpu clocks and per gpu memory clocks. This utility will only control gpu’s that have been bound to an Xserver display. I believe someone above posted you a link to a guys script that will configure your xorg.conf for you that allows this. You’ll know your xorg.conf is right when you open nvidia-settings, look at the gpu’s and you have individual fan control. Be careful overclocking the gpu as the numbers are pluses to the baseline card performance so if it’s running at 2000 and you add +500 then it will crash the XServer (and the machine) because you overclocked it beyond the max. Linux won’t hold your hand and tell you it’s too much, it just gives up.
ok. Here is a sample xorg.conf that should bind 3 cards to 3 X server displays. Compare this to your xorg.conf file and see what differences you have. I also stay away from the nvidia-xconfig command as the xorg.conf it wrote for my rig never booted correctly.
Your BusID’s in the xorg.conf must match the actual PCI device numbers. You can get the bus id’s using the lspci command. You also need to have your device and screen sections to match (1 device section, 1 screen section) and then have an entry for them in ServerLayout.
Coolbits 28 allows all tuning options (fan control, gpu overclock and gpu memory overclock).
Before you do this copy your existing xorg.conf (which works) to xorg.conf_backup. Make changes to xorg.conf and reboot. If the system doesn’t come up, hit CTRL+ALT+F1 to get a console login. login, sudo to root and copy your backup file to xorg.conf, then reboot again. That should get the system back up in gui mode.
Extend your rig frame to allow a box fan to be mounted to it. Point the fan so that it updrafts. (Sucks the heat from the cards and pulls it upwards.) This lowered my card temps by about 10-15C. My issue was that the output of one card was being sucked into the inlet of the next card. The box fan fixed this for less than $20.