It seems that a lot of GPU problems revolve around specific versions of drivers. Though NVidia has their own support structure, you can often learn from information reported by others who fold.
I have loaded the 346.46 drivers and the CUDA Devlopment package from the nvidia website and the example programs work so I assume that CUDA is working but FAH still reports that it isn't detecting CUDA:
*********************** Log Started 2015-05-01T22:35:40Z ***********************
22:35:40:************************* Folding@home Client *************************
22:35:40: Website: http://folding.stanford.edu/
22:35:40: Copyright: (c) 2009-2014 Stanford University
22:35:40: Author: Joseph Coffland <joseph@cauldrondevelopment.com>
22:35:40: Args: --child --lifeline 5522 /etc/fahclient/config.xml --run-as
22:35:40: fahclient --pid-file=/var/run/fahclient.pid --daemon
22:35:40: Config: /etc/fahclient/config.xml
22:35:40:******************************** Build ********************************
22:35:40: Version: 7.4.4
22:35:40: Date: Mar 4 2014
22:35:40: Time: 12:01:17
22:35:40: SVN Rev: 4130
22:35:40: Branch: fah/trunk/client
22:35:40: Compiler: GNU 4.1.2 20080704 (Red Hat 4.1.2-46)
22:35:40: Options: -std=gnu++98 -O3 -funroll-loops -mfpmath=sse -ffast-math
22:35:40: -fno-unsafe-math-optimizations -msse2
22:35:40: Platform: linux2 2.6.18-164.11.1.el5
22:35:40: Bits: 64
22:35:40: Mode: Release
22:35:40:******************************* System ********************************
22:35:40: CPU: Intel(R) Core(TM) i5-4590 CPU @ 3.30GHz
22:35:40: CPU ID: GenuineIntel Family 6 Model 60 Stepping 3
22:35:40: CPUs: 4
22:35:40: Memory: 7.74GiB
22:35:40:Free Memory: 5.24GiB
22:35:40: Threads: POSIX_THREADS
22:35:40: OS Version: 3.16
22:35:40:Has Battery: false
22:35:40: On Battery: false
22:35:40: UTC Offset: -7
22:35:40: PID: 5524
22:35:40: CWD: /var/lib/fahclient
22:35:40: OS: Linux 3.16.7-21-default x86_64
22:35:40: OS Arch: AMD64
22:35:40: GPUs: 1
22:35:40: GPU 0: NVIDIA:3 GK104 [GeForce GTX 760]
22:35:40: CUDA: Not detected
22:35:40:***********************************************************************
My question is, how does FAH try to detect CUDA? Since the CUDA examples are working I must have some environment variable or permission or location set incorrectly but without knowing what its detection algorithm is I am having a hard time debugging it.
I know from other posts that if CUDA is not present it will try to load an openCL WU but I've been trying that for days and all I ever get are BAD_WORK_UNIT errors until it gives up:
Hardware configuration: Intel i7-4770K @ 4.5 GHz, 16 GB DDR3-2133 Corsair Vengence (black/red), EVGA GTX 760 @ 1200 MHz, on an Asus Maximus VI Hero MB (black/red), in a blacked out Antec P280 Tower, with a Xigmatek Night Hawk (black) HSF, Seasonic 760w Platinum (black case, sleeves, wires), 4 SilenX 120mm Case fans with silicon fan gaskets and silicon mounts (all black), a 512GB Samsung SSD (black), and a 2TB Black Western Digital HD (silver/black).
The opencl work units are closely linked to options in the CUDA code by CUDA version, so that is why the client checks for a CUDA version, even though it doesn't use CUDA directly.
I don't know how the detection works. I just know that in all but one case, No CUDA means no folding. There are several opensuse threads on this forum. Please review those for a solution.
FAH Detects CUDA based on the kernel modules you have install for your hardware. Driver packages from 3rd parties (such as Linux ditros) might work ok for 3d games or desktop effects, but I have never gotten OpenCL or CUDA working with them. It is almost a requirement that Nvidia users download and install the drivers directly from Nvidia and according to their instructions.
Once you have done that you can run "sudo lsmod | grep nvidia" to see if all of the required modules are loaded. There should be two of them, nvidia and nvidia-uvm.
E
"In Theory there is no difference between Theory and Practice. In Practice there is."
This thread has been dormant for a while. I suspect it is the same problem I have on OpenSUSE. viewtopic.php?f=89&t=28751
I had installed the Nvidia proprietary drivers which include library for CUDA and OpenCL, but the user fahclient did not get access to the GPU. With this the folding did not work using the GPU.
My own user was added to the group video which I think is required to access the GPU. When I started the FAHClient with my own user it detected CUDA and was able to fold with the GPU.
bruce wrote:Is there an entry in the path or an environment variable that points to the library(s)?
It is not access to the library that is a problem, but access to the graphic device. Any program kan use any library on the system, but what use is the library without access to the device.