App Config multiple GPU Tasks
24 Oct 2019 19:29:12 UTC
Topic 219827
(moderation:
hi,
I'm hoping for a quick troubleshoot after spending a lot of time online and in my .xml file trying to fix this.
Some system specs first:
Windows 10
SSD(270 GB) and HDD (1TB)
Asus maximus hero VIII Mainboard
Intel i7 6700K @4 GHz
16 GB DDR3 RAM
Gigabyte R9 390 (8GB RAM)
550 W PSU fully modular from Corsair
BOINC version: 7.14.2
Here is my app_config.xml file, saved in ..\BOINC\projects\einstein.phys.uwm.edu
<app_config>
<app>
<name>hsgamma_FGRPB1G</name>
<max_concurrent>N</max_concurrent>
<gpu_versions>
<gpu_usage>.5</gpu_usage>
<cpu_usage>.87</cpu_usage>
</gpu_versions>
</app>
</app_config>
I basically want it to only run Einstein (shouldn't be a problem because I have no other projects added) on 87% of the cpu cores and I want it to run 2 GPU tasks simultaneously.
Have a nice day.
Timo
Timo wrote:hi, I'm hoping
)
The cpu part doesn't refer to all cpu only a single cpu thread. So change it to 1 and you will only use 2 threads of a CPU for 2 work units. I can't comment on ATI cards since I don't use them.
Max concurrent is only used when want to limit the total amount of that specific type of work on the card. If that is the only card and you are only going to run 2 work units then you can remove the max concurrent section.
See corrected version above
Your goals seem easily met
)
Your goals seem easily met using the preferences pages on your account at Einstein. Is there a reason you prefer to use the config file directly?
To add to this the
)
To add to this the app_config.xml only tells Boinc what to run (schedule) it will not control what the particular app actually use/get in real time, that's up to the operating system of your computer/device.
If you set it to 1.0 CPU and 1.0 GPU then Boinc will start 1 task for each GPU that is available assuming there is enough CPUs available, the app will then use whatever it can and the OS will allow.
If you set it to 1.0 CPU and 0.5 GPU then Boinc will start 2 tasks for each GPU (if there is enough CPUs available) and then the app will still use whatever it can and the OS will allow.
When/if you "run out" of available resources then Boinc won't start any more tasks.
Boinc can not micro manage how much of the available CPU and/or GPU resources a particular app gets scheduled by the operating system, for this you need to use some other external program/tool.
Timo wrote:I basically want
)
You are misunderstanding how the app_config.xml mechanism works. The cpu_usage value doesn't set the number of CPU cores that are used to run CPU tasks. It sets the fraction of a single CPU thread that is 'budgeted' for GPU support duties for each GPU task that is running. If you have that set to 0.87 and you are running 2 GPU tasks, then BOINC will budget for and subtract 2x0.87=1.64 CPU threads from whatever is the full count of available CPU threads in your machine. 'Available' simply means the full number you have hardware-wise, minus whatever number of threads that might already be subtracted if you have set the preference for how many cores BOINC is able to use (a separate website setting) to something less than 100%. To better understand the app_config.xml mechanism, try reading the documentation.
If your basic aim is to run 2 concurrent GPU tasks with just one CPU thread to support both those tasks with all other CPU threads then able to run CPU tasks, you could leave the % of cores that BOINC is allowed to use set at 100% and just use the following app_config.xml file to do the whole job. Note the cpu_usage of 0.5. That translates to one full thread being 'budgeted' for supporting the two GPU tasks, thus leaving all other threads (7 out of 8) free to run CPU tasks. This will work but the load on your machine will be quite high and response might be quite sluggish if you need to do other things.
If I'm misunderstanding your requirement, please advise.
Cheers,
Gary.
thanks for all the anwers.
)
thanks for all the anwers. all were very helpful and gained me a better understanding of the topic.
Thank you for this better version of it. Now the problem is: I have copied this exactly into my app_config.xml and deleted the client_override.
When I go to BOINC Manager´-> Options -> Read in configuration file nothing happens. It's still just running one gpu tasks.
I don't know what you mean by
)
I don't know what you mean by "deleted the client_override".
Is the file in the Einstein project directory?
Is its name exactly app_config.xml and not something like app_config.xml.txt?
Did you create the file with a plain text editor?
When you click in BOINC Manager to 'read config files' do you then check what response you get in the event log to indicate that the file was read and that there were no syntax or other error messages of any sort? You should just see a message that says, "Found app_config.xml".
Cheers,
Gary.
File addresses in Mojave. I
)
File addresses in Mojave.
I am running TexShop 4.27,recently installed on a new iMac running Mojave.
I do not succede in finding the appropriate typography of the address .... to be inserted in the LaTeX command
\bibliography{....}. Except for this, TexShop 4.27 typesets without problems.
This problem did not appear previously, with an iMac8.1 running OSX 10.6.8 and the corresponding version of TexShop.
Guidelines on how to solve the problem, or some good reference, would be welcome.
Thanks for help, Jean Jeener
Timo wrote:thanks for all the
)
Reboot your computer and restart boinc
thanks again for all the
)
thanks again for all the help
@Gary Roberts thanks, I run your troubleshooting and it worked. now I think I was just a noob.
for the results: time per gpu task was 8-9:xx min (with my r9 390), what would make it 16-19 mins for 2 tasks, now I'm running 2 tasks simultaneously its 16 mins under load, so I expect more like 14 mins.
just the cpu tasks still take more like 9 h istead of the 5:13 h avarage.
I'm glad to hear that you
)
I'm glad to hear that you have it working.
Please realise that running 2 concurrent GPU tasks doesn't make a huge difference in output for the FGRPB1G search. You should expect may a 5-10% improvement. It can be affected by CPU load so you should experiment with reducing that load (run fewer CPU tasks) to see what affect it has on both CPU task times and GPU task times. There is no 'one size fits all' recipe so you need to experiment.
Cheers,
Gary.