I haven't tried the .20 version yet but for my 2x on a 1070 and 970 each app absolutely needed a full CPU core and a little bit of spare. Any disturbance from the full CPU core and the GPU utilization went sporadic from the typical 99% in Win7. I could only run 3 CPU tasks on my 3770k. I don't know how much spare time is actually left in that wait loop/sleep cycle in NV drivers. It seems like the CPU is actually constantly feeding the GPUs on my system.
From the app_config Wiki, If you remove app_config, or one of its entries, you must reset the project in order to restore the proper values. Pretty drastic IMHO. I would like to see what your suggestion accomplishes. Seems a lot simpler and not so upsetting. Let us know if it works.
From the app_config Wiki, If you remove app_config, or one of its entries, you must reset the project in order to restore the proper values. Pretty drastic IMHO.
I agree. I can't remember ever having to reset a project - certainly not in the last few years. There always seem to be a less drastic alternative.
I use app_config.xml on pretty much all of my hosts and have made lots of changes to options followed by a 'reread config files' without ever having an issue or needing to do anything further.
Keith Myers wrote:
I would like to see what your suggestion accomplishes. Seems a lot simpler and not so upsetting. Let us know if it works.
I've never used the <cmdline>...</cmdline> option so I'm only guessing but I would suspect the client only does something like inserting or changing the option in the state file if the option is there in app_config.xml. If removed entirely, there is nothing to prompt the client into taking any action on the state file so what's already there just gets left untouched.
I imagine that a blank <cmdline> </cmdline> option would prompt the client into doing something (hopefully removing/modifying what's already there) even if it only issued some sort of warning/error message about an inappropriate/unable to be parsed option :-). If a blank option doesn't fix the problem, carefully editing it out of the state file would be a less drastic workaround than a complete project reset.
When I first tried the cmdline I had still several 1.18 tasks on the queue. I assumed that the unknown cmdline parameter would have been ignored but it turned out that the tasks were just erroring out. I removed the cmdline from the app_config.xml but the tasks kept still failing although I had rebooted the computer before continueing running. The end result was over 50 tasks lost.
Then the computer started to crunch 1.20 tasks and the cmdline was accepted but unfortunately with very bad results on the runtime. I have 2x GTX970 and the runtime went from 30 minutes to over three hours. CPU usage did drop to about 3 %. So I decided to remove the cmdline but with same results as above: several trashed tasks. I ended up editing the client_state file and didn't actually try the blank <cmdline> </cmdline>.
I had missed the text about resetting the project when removing app_config tags because I had not needed that before. Well it is live and learn .
We're still waiting for x86 version. Especially for low memory cards. I still have to crunch on CPUs only. It doesn't heat enough as it was with GPUs running.
Did the modifications that have been introduced with the 1.18/1.20 versions result in increased GPU load? Currently the GPU load seems rather high as it approaches almost 99 percent. This won't affect crunching only machines but when you think of daily use machines then other apps like a browers may add GPU load as well. That may result in GPU core clock throtteling. That utilisation cap may cost say 10 percent crunching speed as long as the non-BOINC-app is open.
/edit/ No problem, found a box in the settings, if necessary, just mark: don't use GPU while in use. Performance increase of 1,18 will compensate for that. Nevermind.
I've crunched on a Palit GTX 1050 Ti on my Windows 10 PC, in about 1600 s. Now I have installed a GTX 750 Ti on my older Linux box, a SUN WS dated 2008, and it takes 2800 s. What I noticed is the sound from the fans, much louder on the Windows PC than on the Gigabyte board of the Linux box with its Windforce fans.
I haven't tried the .20
)
I haven't tried the .20 version yet but for my 2x on a 1070 and 970 each app absolutely needed a full CPU core and a little bit of spare. Any disturbance from the full CPU core and the GPU utilization went sporadic from the typical 99% in Win7. I could only run 3 CPU tasks on my 3770k. I don't know how much spare time is actually left in that wait loop/sleep cycle in NV drivers. It seems like the CPU is actually constantly feeding the GPUs on my system.
From the app_config Wiki, If
)
Keith Myers wrote: From the
)
Cheers,
Gary.
When I first tried the
)
When I first tried the cmdline I had still several 1.18 tasks on the queue. I assumed that the unknown cmdline parameter would have been ignored but it turned out that the tasks were just erroring out. I removed the cmdline from the app_config.xml but the tasks kept still failing although I had rebooted the computer before continueing running. The end result was over 50 tasks lost.
Then the computer started to crunch 1.20 tasks and the cmdline was accepted but unfortunately with very bad results on the runtime. I have 2x GTX970 and the runtime went from 30 minutes to over three hours. CPU usage did drop to about 3 %. So I decided to remove the cmdline but with same results as above: several trashed tasks. I ended up editing the client_state file and didn't actually try the blank <cmdline> </cmdline>.
I had missed the text about resetting the project when removing app_config tags because I had not needed that before. Well it is live and learn
.
We're still waiting for x86
)
We're still waiting for x86 version. Especially for low memory cards. I still have to crunch on CPUs only. It doesn't heat enough as it was with GPUs running.
Did the modifications that
)
Did the modifications that have been introduced with the 1.18/1.20 versions result in increased GPU load? Currently the GPU load seems rather high as it approaches almost 99 percent. This won't affect crunching only machines but when you think of daily use machines then other apps like a browers may add GPU load as well. That may result in GPU core clock throtteling. That utilisation cap may cost say 10 percent crunching speed as long as the non-BOINC-app is open.
/edit/ No problem, found a box in the settings, if necessary, just mark: don't use GPU while in use. Performance increase of 1,18 will compensate for that. Nevermind.
Any plans or news for a CUDA
)
Any plans or news for a CUDA version of the application?
Best regards
Jürgen
Latest news about a CUDA
)
Latest news about a CUDA version was posted by Bernd in the Wish List section 3 days ago. It's ported and working but way to slow for release.
Link to original message.
Thanks for the link. So many
)
Thanks for the link. So many threads, so little time.
Best regards Jürgen
I've crunched on a Palit GTX
)
I've crunched on a Palit GTX 1050 Ti on my Windows 10 PC, in about 1600 s. Now I have installed a GTX 750 Ti on my older Linux box, a SUN WS dated 2008, and it takes 2800 s. What I noticed is the sound from the fans, much louder on the Windows PC than on the Gigabyte board of the Linux box with its Windforce fans.
Tullio