Hi all,

We are having a problem with correctly scheduling work to Linux clients because the BOINC client for that platform does not report the Nvidia driver version to us.

If you find that you are receiving the cuda60 application version and it is not working for you, the fixes are:

*) Upgrade to driver 334.21 or later

*) Try the patched (v7.3.15) client I have made available at:

http://secure.acellera.com/insecure/boinc/boinc (RedHat etc)
http://secure.acellera.com/insecure/boinc/boinc_ubuntu (Ubuntu)

(the patch itself is at http://secure.acellera.com/insecure/boinc/gpu_nvidia.cpp_patch )

This client reports the driver version and allows a correct scheduling decision to be made.

Matt

Did you send this path to boinc developers also? This is a very annoying thing in linux version.

Thanks
Zdenek

If Matt doesn't, I'd be happy to.

I imagine he's waiting for some positive feedback to confirm that the patched version works as intended, without any unexpected side effects.

Already sent upstream. Should be in the next release.

Matt

I'm probably having a senior moment but what am I supposed to do with that file as I can't get it to do anything?

Download it, do a "chmod +x boinc" to make it executable, then run it.

If you normally use the boinc mangler or boinccmd, put it in the same directory as those, replacing the "boinc" executable that is already there (take a backup first).

If your OS isn't a redhat clone, you might need to make symlinks for libssl.so.10 and libcrypto.so.10, to whatever libssl/crypto you have in /usr/lib

Matt

I'm running linux mint 16 but I'm getting errors libssl.so.10, libcrypto.so.10 not found when I run your boinc app.

ldd boinc (Matt's version)
libssl.so.10 => not found
libcrypto.so.10 => not found


ldd boinc (7.2.42 version)

libssl.so.1.0.0 => /lib/x86_64-linux-gnu/libssl.so.1.0.0 (0x00007f892f87f000)
libcrypto.so.1.0.0 => /lib/x86_64-linux-gnu/libcrypto.so.1.0.0 (0x00007f892f4a2000)


symlinks:

sudo ln -s /lib/x86_64-linux-gnu/libssl.so.1.0.0 /lib/x86_64-linux-gnu/libssl.so.10
sudo ln -s /lib/x86_64-linux-gnu/libcrypto.so.1.0.0 /lib/x86_64-linux-gnu/libcrypto.so.10


ldd boinc (Matt's version)
libssl.so.10 => /lib/x86_64-linux-gnu/libssl.so.10 (0x00007f1823727000)
libcrypto.so.10 => /lib/x86_64-linux-gnu/libcrypto.so.10 (0x00007f182334a000)

Everything looks ok, but when I run Matt's boinc app, I get an error "libssl.so.10, libcrypto.so.10 not found"

Did I miss anything?

Try this build:

http://secure.acellera.com/insecure/boinc/boinc_ubuntu

Try this build:

http://secure.acellera.com/insecure/boinc/boinc_ubuntu

ldd boinc_ubuntu

libssl.so.1.0.0 => /lib/x86_64-linux-gnu/libssl.so.1.0.0 (0x00007fdaaf900000)
libcrypto.so.1.0.0 => /lib/x86_64-linux-gnu/libcrypto.so.1.0.0 (0x00007fdaaf523000)

Thu 24 Apr 2014 04:49:30 AM EDT | | CUDA: NVIDIA GPU 0: GeForce GTX 780 Ti (driver version 337.12, CUDA version 6.0, compute capability 3.5, 3072MB, 2759MB available, 5345 GFLOPS peak)

It works! thanks!

Hi

Without any patch(es) , no problems found

Gentoo-10

1			24/04/2014 13.09.11	Starting BOINC client version 7.2.33 for x86_64-pc-linux-gnu	

Gentoo-10

6			24/04/2014 13.09.11	CUDA: NVIDIA GPU 0: GeForce GTX 780 (driver version unknown, CUDA version 6.0, compute capability 3.5, 3071MB, 2989MB available, 4818 GFLOPS peak)	
7			24/04/2014 13.09.11	CUDA: NVIDIA GPU 1: GeForce GTX 780 (driver version unknown, CUDA version 6.0, compute capability 3.5, 3072MB, 3002MB available, 4818 GFLOPS peak)	
8			24/04/2014 13.09.11	OpenCL: NVIDIA GPU 0: GeForce GTX 780 (driver version 334.21, device version OpenCL 1.1 CUDA, 3071MB, 2989MB available, 4818 GFLOPS peak)	
9			24/04/2014 13.09.11	OpenCL: NVIDIA GPU 1: GeForce GTX 780 (driver version 334.21, device version OpenCL 1.1 CUDA, 3072MB, 3002MB available, 4818 GFLOPS peak)	
10			24/04/2014 13.09.11	OpenCL CPU: Intel(R) Core(TM) i7-4770 CPU @ 3.40GHz (OpenCL driver vendor: Intel(R) Corporation, driver version 1.1, device version OpenCL 1.1 (Build 31360.31426))	

ldd /usr/bin/boinc
        linux-vdso.so.1 (0x00007fff317ff000)
        libcurl.so.4 => /usr/lib64/libcurl.so.4 (0x00007f9bb5bec000)
        libssl.so.1.0.0 => /usr/lib64/libssl.so.1.0.0 (0x00007f9bb5981000)
        libcrypto.so.1.0.0 => /usr/lib64/libcrypto.so.1.0.0 (0x00007f9bb55aa000)
        libdl.so.2 => /lib64/libdl.so.2 (0x00007f9bb53a6000)
        libz.so.1 => /lib64/libz.so.1 (0x00007f9bb5190000)
        libX11.so.6 => /usr/lib64/libX11.so.6 (0x00007f9bb4e53000)
        libXss.so.1 => /usr/lib64/libXss.so.1 (0x00007f9bb4c4f000)
        libXext.so.6 => /usr/lib64/libXext.so.6 (0x00007f9bb4a3c000)
        libpthread.so.0 => /lib64/libpthread.so.0 (0x00007f9bb481f000)
        libstdc++.so.6 => /usr/lib/gcc/x86_64-pc-linux-gnu/4.8.2/libstdc++.so.6 (0x00007f9bb4515000)
        libm.so.6 => /lib64/libm.so.6 (0x00007f9bb4220000)
        libgcc_s.so.1 => /usr/lib/gcc/x86_64-pc-linux-gnu/4.8.2/libgcc_s.so.1 (0x00007f9bb400a000)
        libc.so.6 => /lib64/libc.so.6 (0x00007f9bb3c5f000)
        libldap-2.4.so.2 => /usr/lib64/libldap-2.4.so.2 (0x00007f9bb3a15000)
        liblber-2.4.so.2 => /usr/lib64/liblber-2.4.so.2 (0x00007f9bb3805000)
        libresolv.so.2 => /lib64/libresolv.so.2 (0x00007f9bb35ee000)
        /lib64/ld-linux-x86-64.so.2 (0x00007f9bb5e4f000)
        libxcb.so.1 => /usr/lib64/libxcb.so.1 (0x00007f9bb33cd000)
        libXau.so.6 => /usr/lib64/libXau.so.6 (0x00007f9bb31c9000)
        libXdmcp.so.6 => /usr/lib64/libXdmcp.so.6 (0x00007f9bb2fc3000)

Hi huys :D

I have many WUs fault with 780Ti and cuda6



in gpugrid work directory have a problems with libcudart and libcufft

libraries exist :

$ ls -lh | grep lib
-rw-r--r-- 1 saba users 360K 16 mar 14.20 libcudart.so.4.2.9
-rwxr-xr-x 1 saba users 322K 12 apr 19.12 libcudart.so.6.0
-rw-r--r-- 1 saba users  30M 16 mar 14.20 libcufft.so.4.2.9
-rwxr-xr-x 1 saba users  32M 12 apr 19.41 libcufft.so.6.0

but excutable doesn't find them

$ ldd acemd.821-60.bin
        linux-vdso.so.1 (0x00007fffef10c000)
        libdl.so.2 => /lib64/libdl.so.2 (0x00007fd097934000)
        libpthread.so.0 => /lib64/libpthread.so.0 (0x00007fd097717000)
        libnvidia-ml.so.1 => /usr/lib64/libnvidia-ml.so.1 (0x00007fd097478000)
        libcuda.so.1 => /usr/lib64/libcuda.so.1 (0x00007fd096513000)
        libcudart.so.6.0 => not found
        libcufft.so.6.0 => not found
        libstdc++.so.6 => /usr/lib/gcc/x86_64-pc-linux-gnu/4.8.2/libstdc++.so.6 (0x00007fd09620a000)
        libm.so.6 => /lib64/libm.so.6 (0x00007fd095f15000)
        libgcc_s.so.1 => /usr/lib/gcc/x86_64-pc-linux-gnu/4.8.2/libgcc_s.so.1 (0x00007fd095cff000)
        libc.so.6 => /lib64/libc.so.6 (0x00007fd095954000)
        /lib64/ld-linux-x86-64.so.2 (0x00007fd097b38000)
        libz.so.1 => /lib64/libz.so.1 (0x00007fd09573e000)
        librt.so.1 => /lib64/librt.so.1 (0x00007fd095536000)

will try to reset the proiject

now I'm installing cuda-6.0.37 and I will create a symlink to gpugrid work directory

[edit] libraries missing for all excutables

$ acemd.800-42.bin
acemd.800-55.bin
acemd.821-60.bin

libcudart.so.4 => not found
        libcufft.so.4 => not found

I'm still getting errors on the CUDA6 short tasks on my GTX 560. Yes, I tried the boinc version in this thread. Doesn't make a difference. Almost 98% of all CUDA6 tasks error out either immediately or they crunch for some time and then error out. Error codes don't give a clue about what has gone wrong

GPUGRID is the buggiest project I've crunched for thus far

---

Team Belgium

I use the patched (v7.3.15) ubuntu client file above with nvidia-331.38 driver. GPUgrid doesn't get CUDA 6 jobs and run fine with lower version CUDA jobs.

Hi huys :D


now I'm installing cuda-6.0.37 and I will create a symlink to gpugrid work directory

[edit] libraries missing for all excutables

$ acemd.800-42.bin
acemd.800-55.bin
acemd.821-60.bin

libcudart.so.4 => not found
        libcufft.so.4 => not found

As Matt said in the original post, you'll need nvidia driver 334.21 or later to run cuda 6.0 tasks. If you have installed the right drivers, try the following.

I would uninstall cuda-6.0.37. The project provides the necessary cuda libraries in the GPUgrid project directory.

Try adding the path to the GPUGrid folder to a new file in /etc/ld.so.conf.d directory.

sudo nano /etc/ld.so.conf.d/gpugrid.conf

on a single line, type in the complete path to your GPUgrid folder and then save the file.

next:

sudo ldconfig

Then check to see if the app finds the libraries.

You don't need to install CUDA yourself. We ship all the necessary runtime libraries along with the application.

Matt

microchip,

Looking at the results from your GTX560 machine (#167410), I would say that your GPU is bad.

Matt

microchip,

Looking at the results from your GTX560 machine (#167410), I would say that your GPU is bad.

Matt

I doubt that. I've done stress tests on both its memory and the GPU itself. It passes every time. It crunched without issues in the past the CUDA5 apps. Every since CUDA6 tasks came, they started to fail on it

---

Team Belgium

I have a Gigabyte (GV-N66TOC-2GC) GTX660Ti which failed about 1/4 tasks under Linux. Tried different generations of Linux, system updates and a range of different drivers (from 304 to 337) and it kept failing work. It's a dual fan model and ran fairly cool and quiet, so temp probably wasn't an issue.

Since I swapped it into a Win7 system 3days ago it hasn't failed a single WU...
The GTX670 that went from the W7 to the Ubuntu 14.04 system hasn't failed either!

I've had a similar experience with a GTX660 and a GTX260 in the past. Such problems could be down to some oddity between the driver, OS and card, or be an OS, motherboard chipset issue or even a card firmware revision/driver problem.

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

I would uninstall cuda-6.0.37. The project provides the necessary cuda libraries in the GPUgrid project directory.

Try adding the path to the GPUGrid folder to a new file in /etc/ld.so.conf.d directory.

sudo nano /etc/ld.so.conf.d/gpugrid.conf

on a single line, type in the complete path to your GPUgrid folder and then save the file.

next:

sudo ldconfig

Then check to see if the app finds the libraries.

tnx for the trick
cuda is installed in /opt/ (outside boinc directory)

now i'm crunching 1 WU with cuda4.2

so ..

$ ldd acemd.800-42.bin
        linux-vdso.so.1 (0x00007fffeffff000)
        libdl.so.2 => /lib64/libdl.so.2 (0x00007f050160e000)
        libpthread.so.0 => /lib64/libpthread.so.0 (0x00007f05013f1000)
        libnvidia-ml.so.1 => /usr/lib64/libnvidia-ml.so.1 (0x00007f0501152000)
        libcuda.so.1 => /usr/lib64/libcuda.so.1 (0x00007f05001ed000)
        libcudart.so.4 => /home/saba/boinc/projects/www.gpugrid.net/libcudart.so.4 (0x00007f04fff93000)
        libcufft.so.4 => /home/saba/boinc/projects/www.gpugrid.net/libcufft.so.4 (0x00007f04fdf6b000)
        libstdc++.so.6 => /usr/lib/gcc/x86_64-pc-linux-gnu/4.8.2/libstdc++.so.6 (0x00007f04fdc62000)
        libm.so.6 => /lib64/libm.so.6 (0x00007f04fd96d000)
        libgcc_s.so.1 => /usr/lib/gcc/x86_64-pc-linux-gnu/4.8.2/libgcc_s.so.1 (0x00007f04fd757000)
        libc.so.6 => /lib64/libc.so.6 (0x00007f04fd3ac000)
        /lib64/ld-linux-x86-64.so.2 (0x00007f0501812000)
        libz.so.1 => /lib64/libz.so.1 (0x00007f04fd196000)
        librt.so.1 => /lib64/librt.so.1 (0x00007f04fcf8e000)

waiting to finish

Hi, I think that says something does not work.

Right now I have installed the Nvidia driver 331.67 and BOINC 7.3.15 (the GPUGRID recommended) reporting the cuda version installed, but I get only tasks for CUDA 4.2 but step BOINC 7.2.42 (recommended version) I have tasks CUDA 6 . Greetings.