CryptoCurrency: Building a Rig Part 3
So, this has been quite a learning experience. Don’t mix between your distro’s nvidia drivers, and the nvidia drivers from the website. You end up in a weird in-between state where nothing works properly. More importantly it ends up not being reproducable if you do get it working. I also learned a little bit about risers, and how tempermental they can be.
Installing the NVIDIA Driver
You might be tempted to go grab the NVIDIA binary installer from nvidia.com, but don’t. When it comes to distros Debian is pretty bleeding edge. Ubuntu being based on Debian inherits this. They also struggle to keep things really user friendly. While I am more of a command line junkie, I went ahead and did this part through the UI to get back in business.
1. Open Software & Updates 2. Select the Additional Drivers tab 3. Set the NVIDIA Corporation...GTX 1080 to use the NVIDIA...387 driver. 4. Click Apply Changes, and wait. It may take a few minutes.
One of the benfits you get by doing it through the UI, is the UI will properly unload the Nouveau driver, and use the NVIDIA driver. Other tutorials tell you to blacklist it and reboot, etc. Which just ends up in X failing to work, and you struggling to prevent the conflicting driver from loading because something in the GUI tries to load it regardless of blacklists.
The Riser cards
Some advice given to me was order all of the ones you need at once. Don’t order them one by one, because you get a different batch with different problems. So I took this advice and ordered a set of 6 risers. As soon as I plugged the risers in and booted up, I logged in via SSH and noticed I couldnt really use the GPU. When I did, the process would hang and become unkillable. So I loaded up the console and took a look.
[ 5315.986588] pcieport 0000:00:1c.0: AER: Corrected error received: id=00e3 [ 5315.987249] pcieport 0000:00:1c.0: can't find device of ID00e0 [ 5315.995632] pcieport 0000:00:1c.0: AER: Corrected error received: id=00e3 [ 5315.995664] pcieport 0000:00:1c.0: PCIe Bus Error: severity=Corrected, type=Physical Layer, id=00e3(Receiver ID)
This message was scrolling so fast on the console, that I had to kind of unfocus my eyes to even try to read it. After some research, I decided to turn it off. Once I did, the errors disappeared. But after I started the miner, it bailed and threw an error about the bus going away. I noticed there is a little bit of wiggle room with how these PCIe x1 inserts fit in the motherboard. I’m not sure if its a normal thing to need, or if this was the reason. Either way I added irqpolling as well. To do this set your GRUB_CMDLINE_LINUX line to the following:
root@rig1:~# cat /etc/default/grub | grep CMDLINE GRUB_CMDLINE_LINUX_DEFAULT="quiet" GRUB_CMDLINE_LINUX="pci=noaer irqpoll" root@rig1:~#
Once you’ve done that, update your initrd with your new parameters by typing “update-grub”:
root@rig1:~# update-grub Generating grub configuration file ... Found linux image: /boot/vmlinuz-4.13.0-21-generic Found initrd image: /boot/initrd.img-4.13.0-21-generic Found linux image: /boot/vmlinuz-4.13.0-16-generic Found initrd image: /boot/initrd.img-4.13.0-16-generic Adding boot menu entry for EFI firmware configuration done root@rig1:~#
Now reboot, and you should be ready to start mining with your new risers.
It is truly like music to my eyes. No errors, just mining…
With 2x NVIDIA GTX 1080 ROG Strix 8G cards:
- bminer - 1000-1100 sol/s - EWBF's Zcash CUDA Miner 0.3.4b - 1000-1030 sol/s. - ccminer 2.2.5 (tpruvot fork) - 800-900 sol/s
I did manage to get ccminer to build, I will make a separate post on that. But the clear winner based on sol/s (and matches what the pool reports), is bminer.