Hi, I am trying to run OpenCL code on a new "mini-cluster" machine (4 x AMD Opteron(tm) Processor 6380 (16 Cores)) and I obtain the following error when the
clGetPlatformId() function is reached:
Setting of real/effective user Id to 0/0 failed
FATAL: Module fglrx not found.
Error! Fail to load fglrx kernel module! Maybe you can switch to root user to load kernel module directly
Segmentation fault (core dumped)
I have tried both AMD APP SDK v2.8 and 2.9 without success, the machine has not AMD GPU (only a Matrox one), and I would like to use meanwhile
the AMD CPUs before buying a proper GPU for OpenCL development. Thanks for your help.
HAve you installed the new catalyst which is supporting 2.9 or 2.8 properly in your system?
Please provide us the system information like CPU, GPU, BIOS, OS etc
I have not installed catalyst because I do not have an AMD GPU, although I will surely buy one in the next weeks.
CPU: 4 x AMD Opteron(tm) Processor 6380 × 15
GPU: VGA compatible controller: Matrox Electronics Systems Ltd. MGA G200eW WPCM450 (rev 0a)
OS: Linux Ubuntu 12.04 LTS
regards
Then probably fglrx is not installed properly. Please check for its installation and here are some steps for it
Ctrl+Alt+T
to open up the terminal. sudo apt-get update
, type your password, then hit enter. sudo apt-get upgrade
sudo apt-get install PACKAGENAME
sudo reboot
And you're done! fglrx should work as intended.
I tried your steps without success... but what I cannot understand is why OpenCL complains about the fglrx module, if I do not have an AMD GPU card installed ?
I am sure that if I try to install the fglrx module with a Matrox GPU card I will kill my system...
no clue with this problem?
yo dude,
i got your back. What are you trying to do with opencl?
Dude. openCL its meant for GPUs.
Are you gonna calculate something?
Matrix cards are only for display in 2013 at least.
Let me know what your doing and i can help. even tho i suck at linux. obviously not as much as OP tho.
reg,
PoTski
I just wanted to use OpenCL on the cluster in the meanwhile... of course I will buy a GPU
You could try to build packages using fglrx for your distribution (i.e. using the official binary driver installer), and then only install the fglrx-dev package, which contains the OpenCL runtime.
Let me revive this thread by saying I have the very same issue with Ubuntu 14.04 and 2 AMD R9 280X (Pitcairn) GPUs. It doesn't matter if I use the .sh installer or download the pre-built .deb packages. The result is the same. /etc/OpenCL/vendors is populated with correct .icd files containing the names of the implementations that are part of ld.so.conf files, so they should be able to lead at runtime. All OpenCL samples and clinfo as well segfault.
I did a clean of the system and the only libOpenCL.so is that provided by the driver itself (Crimson) and that installs under plain /usr/lib (not the nicest place to install under Ubuntu)
ldd /usr/bin/clinfo says that it is linking against libOpenCL.so, which is reported to be provided by fglrx-core (dpkg -S /usr/lib/libOpenCL.so.1). /dev/ati/card0 and card1 exist and seem to have proper access control (rw-rw-rw). There is no X installed on the machine, hence I can't run the .sh installer plainly, just generate Ubuntu/trusty packages. If I run it with --install, it complains that it cannot determine the version of X.
Any help?
Check if you have /etc/ati/amdpcsdb.default installed https://community.amd.com/message/2685179#2685179
I do, and it is a fairly long file.
Hi,
Latest Crimson driver, 15.302, and every single driver since Catalyst 15.101, have a nasty memory corruption in libamdocl.so that can crash any program linking against it at any time. It's a fairly random segfault. I have the valgrind report in another thread Memory corruption in latest crimson driver 15.302?
Nikos