As far as I can see then everything is installed as it should (fahclient_7.4.16_amd64,fahcontrol_7.4.16-1_all, fahviewer_7.4.16_amd64) and drives are installed and some extra libraries are install for cuda support as well.
But it appears that folding@home cannot detect my gtx 1070. I am running debian 9.1
What drivers are are you using? You need the actual Nvidia drivers for folding and you say you have extra libraries for Cuda which would be included in the Nvidia drivers.
Have you checked to see if the client has downloaded a new version of GPUs.txt? 7.4.16 was released before nVidia started selling the 1070. Your log ends before the section which would show a download of a newer GPUS.txt after a new install.
Rel25917: I am using the closed source nvidia driva that packaget for debian and install via its repo. That is the way the debian maintainer recommend installing the drive instead of for the homepage. The cuda libraries are split into different packages instead of being included in the nvidia driver
Joe_H: I could not see any trace of the a download of the GPU.txx downloads in the log. What is the path to both the log and the GPU-txt (Just to make sure we are looking at the same)
The issue here is whether you have installed the OpenCL library, not necessarily the CUDA library. You get both if you use the drivers directly from NVidia. The active GPU cores all require OpenCL support.
From the FAH perspective, the FAHCore (analysis code) can either be written in OpenCL (for any modern GPU, including both NVidia and AMD) or in CUDA (limiting it to NVidia only). Obviously FAH will develop an OpenCL version first and most of the time a secondary CUDA version is never added; you have to have the OpenCL libraries for your GPU.
FAH doesn't test its FAHCore with whatever Debian is recommending so if you do find an option to add OpenCL, you'll be operating at your own risk. We strongly recommend the latest non-Beta version of drivers directly from the supplier of your GPU.
#bruce: I am a bit confused.
The stuff I did paste do say:
15:19:14: GPUs: 0
15:19:14: CUDA Device 0: Platform:0 Device:0 Bus:1 Slot:0 Compute:6.1 Driver:8.0
15:19:14:OpenCL Device 0: Platform:0 Device:0 Bus:1 Slot:0 Compute:1.2 Driver:375.66
As far as I read it then both cuda and openCl libraries is alreay installed and detected. But folding@home cannot find any GPU's. And I have installed nvidia-opencl-common from the repo that should be the nvidia's opencl library as far as I know.
I am not an expert in this area so I wonder if there is a way to run the detection of the gpu and library by hand to see where it goes bad?
Open FAHControl, go to the configuration tab and then the expert tab, you will see an argument; gpu that has false next to it, click on it and then click remove. Click Add and then put gpu in the argument box and true in the value box. The client should now download gpu.txt. Reboot and you should be good to go.
beer wrote:Nathan_P: that seems to do the trick. Thanks
No worries, I've had lots of practise this last week as I've had to re build all my rigs and I took the opportunity to update the OS. I might even throw together a guide for getting Linux installed and F@H up and running.
as far as I remember then I did have a few issues that you might want to add:
1: Fah requres python-support which most modem (even Debian stable) do not have. Your recommended ways would be greate
2: As far as I know then Cuda support is not included in the standard nvidia-driver in some distroes (inc Debian and *buntu).
3: libopenCL.so was missing. This also neeeded to be setup before far would work
3: And then the last part with the option (as you noted)
If I could influence people then I would recomend that the fah-packages to be included in the most popular distroes for easy installation
Oh I had to solve all of them. Fortunately, for once I was in a patient mood so did some googling with each step and noted everything down. Going to try one more run through when I get back from my hols and if it works i'll put it up on my team's forum for all to see.
h a note - you can install just the opencl packages, however if you install the cuda toolkit you get cuda and opencl - more useful if core 22 comes with a cuda only version
I hope that when Debian Butch/10 comes out then someone has included those 3 package for easier installation so new users dont have to go trough those problems. If I did have the skills for that then I would have done it already
Its not difficult, IF you know where to look and have kept up on all the forum threds on these issues, - watch for the guide in a couple of weeks, it will be distro specific but its better than nothing
Nathan_P wrote:Its not difficult, IF you know where to look and have kept up on all the forum threds on these issues, - watch for the guide in a couple of weeks, it will be distro specific but its better than nothing
Hi Nathan_P,
Were you, by any chance, able to create the guide?
Needing help with installing OpenCL in Ubuntu, perhaps the procedures for Debian/Mint are the same.
It was almost complete when I found out that somoeone over at the evga forum had already created one for the exact version of Mint that i was writing for so work ceased. I may still have the notes but the one at EVGA is excellent.
Not sure if Debian is the same as mint but I use Mint 18.2. Let me know where you are getting stuck and i will see if i can help.
