There is a new Nvidia project HERE.
It may be driver sensitive ... Sphynx had problems on some of his GPUs. Scole is having good results on 7xx, 9xx, and 10xx cards using 372.9. I'm running it on my 1080 using 375.7.
Printable View
There is a new Nvidia project HERE.
It may be driver sensitive ... Sphynx had problems on some of his GPUs. Scole is having good results on 7xx, 9xx, and 10xx cards using 372.9. I'm running it on my 1080 using 375.7.
Runs fine on 375.70 on any 500 series or 750ti.
They don't even have the forums set up yet. Would it be inappropriate to PM the admin and ask about getting badges =))
It is showing up in WuProp so you can get a star for it.
Mine error out immediately after starting. Ubuntu 16.04 LTS, 1060 6GB, driver 367.35
<core_client_version>7.6.31</core_client_version>
<![CDATA[
<message>
process exited with code 127 (0x7f, -129)
</message>
<stderr_txt>
../../projects/parlea.ru_andersonattack/anderson_attack_a5_02_x86_64-pc-linux-gnu__cuda_fermi: error while loading shared libraries: libcudart.so.7.5: cannot open shared object file: No such file or directory
</stderr_txt>
]]>
Why does error message say cuda_fermi for a pascal card?
Upgrade your driver to at least 375.70 or the latest one.Quote:
Originally Posted by nanoprobe
The application is called (Cuda_fermi) so I think it why that sows up. Both Scole and I aren't having any trouble with GTX 10xx cards. I'm running under Windows .... I don't know about Scole.
Synaptic package upgraded to 370.10 but they still error out. I wonder if 375 is out for Linux. They seem to lag behind windows.
Downloaded the 375.10 driver but as usual with me and linux there's a snag. Terminal says I have to exit x server to install driver but when I do that all I have is a black screen with a flashing cursor. *face palm*
There is 1 Linux host on the 1st page of "top computers" that is running Linux. The driver is 367.44 and the GPU is a 750TI. There are only a couple of Linux hosts on the following couple of pages and they aren't having any luck.Quote:
Originally Posted by nanoprobe
You need to log out of your GUI and use one of the character only Multi-Screens...Quote:
Originally Posted by nanoprobe
Logout.
Hit Ctrl-Alt-F1 to get to a character only login prompt.
Login and get to a root prompt. I normally go with "sudo bash" so all the rest of that session is in admin mode.
Not sure which gui you are using, but you need to stop it.
service lightdm stop
Will handle the default Unity interface
Then you can run the installer
When that's done, the best idea is to reboot the system.
You can restart the GUI by
service start lightdm
Or, again, whichever DesktopManager you run. (Like maybe gdm...)
From the character Multi-Screen, you can always hit the Ctrl-Alt-F7 to get back to your GUI if you haven't stopped it. On Linux, you always have a number of "terminals" available if you need a command prompt. F1-F12 are commonly used, with F1-F6 commonly having Character terminals started and F7 normally being the GUI.
Followed your directions but I get a new error that says nvidia-uvm kernel module is still running and the driver install fails. Can't figure out how to get nvidia-uvm to unload.Quote:
Originally Posted by Mumps
367.57 is the current Recommended Linux release. 370.23 is the current Beta linux release. Where did you find the 375.10 for Linux?
I tried installing 367.57 to see if that worked for my GTX 670 under Ubuntu 14.04.3. I got the same error you did. I eventually determined it was because BOINC keeps the driver loaded even when it's not running a GPU WU. So you'll need to add
service boinc-client stop
to the list of things you stop. As the next command after the gdm stop command should work fine. It then allowed me to update the drivers just fine.
Once the drivers were updated, I just started the boinc-client service again. Too bad that still didn't get successful work out of the project. :) I'm trying the Beta drivers next.Code:09-Nov-2016 21:14:08 [---] CUDA: NVIDIA GPU 0: GeForce GTX 670 (driver version 367.57, CUDA version 8.0, compute capability 3.0, 1997MB, 1964MB available, 2915 GFLOPS peak)
09-Nov-2016 21:14:08 [---] OpenCL: NVIDIA GPU 0: GeForce GTX 670 (driver version 367.57, device version OpenCL 1.2 CUDA, 1997MB, 1964MB available, 2915 GFLOPS peak)
Looks like the missing piece is probably not the drivers, but the Cuda RunTime Library... Go here to download the 7.5 version of the Cuda Toolkit which is what it looks like the app is expecting to have installed...
https://developer.nvidia.com/cuda-75-downloads-archive
Have patience, because that huge package thinks for quite a while without showing any progress bars along the way. Read the installation prompts because that humongous package comes with the 352.39 drivers, which you don't want to install over your newer ones. I also opted to not install the samples.
You may be able to get away with installing the current 8.0 Cuda Toolkit if it creates links for backwards compatibility, but I haven't tested that.
Next hurdle I ran into is it looks like the app is also expecting a compiler is available. At the least a modern libstdc++6 needs to have been installed. That's as far as I've gotten now. My 14.04.3 is busy trying to upgrade to 14.04.5 for my next attempt. :)
Looks like a no-go on 14.04 for me. But let me know how far you get on 16.04.
Yo gentlemen, and I use that term loosely, check out the 2nd post on THIS thread ... the admin tells what to do to get it going under Ubuntu :D
Thanks Mumps. You can find the 375.10 driver here. And from the FWIW dept upgrading to the 370.23 driver makes my 1060 run hotter and pull 40 more watts of power for no appreciable increase in crunching speed that I could see @ GPUGrid. Will be interesting to see what the 375.10 does. Might be downgrading the driver and skipping the new project if I get the same results.
Thanks for the link Bryan.
See the project forums for my end result. I actually got it working on Ubuntu 14.04.5 with the 370.23 drivers on my GTX670. Not easy, not clean, but definitely satisfying. :))
Now that I ran that one WU, I've detached that host to run elsewhere. :)
OK. So for Ubuntu 14.04.5 64bit, here's the order I did things in that seems to work best. This is probably pretty close to what would work for MiNT, and you can probably skip the dpkg install on Ubuntu 16.04.
Download the three parts (I downloaded mine to the default "~/Downloads" folder.)
libstdc++6_5.4.0-6ubuntu1~16.04.4_amd64.deb
cuda_7.5.18_linux.run
NVIDIA-Linux-x86_64-367.57.run
Then run the following series of commands from a "sudo bash" command prompt. Choose not to restart the system after the update-manager step is done so the rest of the steps can get run.
WU's downloaded from the project worked immediately for me. You can reboot whenever it's convenient. I actually ran that code as a script on the final host I set up this morning and it worked immediately. There may have been some extra packages installed on that particular host prior to today though, so you may also need to have installed packages like libstdc++6 or gcc, but you can investigate those items if it fails to work right away for you as shown here.Code:service lightdm stop
service boinc-client stop
apt-get update
update-manager
cd ~/Downloads
bash ./cuda_7.5.18_linux.run
cp /tmp/cuda_install_*.log ~/
bash ./NVIDIA-Linux-x86_64-367.57.run
dpkg -i libstdc++6_5.4.0-6ubuntu1~16.04.4_amd64.deb
echo /usr/local/cuda/lib64 >>/etc/ld.so.conf.d/cuda.conf
ldconfig
service boinc-client start
service lightdm start
disregard this post - wrong project.
And ATI??? Dang, now I gotta go update drivers on those hosts too??? Sheesh! :)
I don't see an ATI app at andersonattack and there is still no work available on cuda either?
My bad...it was XANSONS for COD
Save your time...sorry!Quote:
Originally Posted by Mumps
Right on both counts...sorry!Quote:
Originally Posted by Bok