BOINC 7.0.40 update heads up

[nanoprobe]

I just updated to 7.0.42, after trying for days to time my cleaning of tasks to when I was available to update.
Added the app_config for hcc and some other cpu projects.
2 things learned.
A) it doesn't seem neccesary to add each cpu app like with app_info, without them it still downloads and runs them, so the app_config really only needs to be set up for multi gpu wu's, at least for me thus far.
b) if you want to run cpu hcc units along side the gpu ones, add gpu+cpu threads you want to run from hcc1 in conccurent threads.

Yeroon

If you want to run a specific number of cores on a CPU project then you'll need to add that to the config file. If not it will use all the available cores that are not used to run GPU tasks.

[Yeroon]

I'm aiming for full use of the cpu without hindering the gpu wu's, and its a lot easier this way then with the app_info. I never did get DSFL to work with the app file, and it works flawless now.
DSFL wu's use another 20-30w of power over 2 cores, and give me more heat (good thing, its winter) compared to hfcc/hcc cpu units. I think I'll add the rest of the projects to the list now that I don't have to make each one work in the app file.

Thanks for the thread guys!
Yeroon

[OldChap]

Stitch: When you get to here this is for running 2 cards on one rig. I selected 10 per card in this instance running on a 4 core 8 thread processor

<app_config>
<app>
<name>hcc1</name>
<max_concurrent>20</max_concurrent> This is the total number of threads you want to run on BOTH cards added together
<gpu_versions>
<gpu_usage>.1</gpu_usage> This is 1 divided by the number of threads you want to run on ONE card
<cpu_usage>0.4</cpu_usage> This is the number of cores(threads) you have on your CPU divided by the max_concurrent above
</gpu_versions>
</app>
</app_config>

If you have read through this thread you will already know that you need to finish up all the work running under app_info before you make the change to app_config

[bluestang]

Thanks nano for the heads-up on the inevitable!

Just switched over my work pc to the app_config, will do my home pc tonight.

[StitchExperimen]

<!-- Processor i7 3930 @4.1Ghz with Hyper Threading on -->
<!-- memory is 1866XML 4 channel -->
<!-- Dual video cards Gigabyte 7950 core @1000 mem @950 -->
<!-- GPU temp 54c fan 36% -->
<!-- COMMENT: I have no more processing power to give I can only do 28 wu concurrently, 16 wu and 14 wu throw errors. -->
<!-- my actual times to completion are a low ~25 sec. to a high of ~33 sec. -->
<!-- -->
<!-- The formula below needs explanation -->
<!-- the first is concurent wu -->
<!-- the second is how many wu the GPU0 card strips away -->
<!-- the third is the cpu cores in fraction you want to allow usage -->
<!-- (in my case between 25 and 36) 12 threads times 3 or .333 repeating -->
<!-- and the left over wu are sent to GPU1 -->
<!-- -->
<app_config>
<app>
<name>hcc1</name>
<max_concurrent>28</max_concurrent>
<gpu_versions>
<gpu_usage>.071400</gpu_usage>
<cpu_usage>.333300</cpu_usage>
</gpu_versions>
</app>
</app_config>


<!-- Application configuration -->
<!-- -->
<!-- This mechanism allows you to specify scheduling parameters associated with specific applications. It is available with 7.0.40+ client versions. -->
<!-- -->
<!-- To do this, create a file app_config.xml in the projects' directory, with the following format: -->
<!-- -->
<!-- <app_config> -->
<!-- <app> -->
<!-- <name>uppercase</name> -->
<!-- <max_concurrent>1</max_concurrent> -->
<!-- <gpu_versions> -->
<!-- <gpu_usage>.5</gpu_usage> -->
<!-- <cpu_usage>.2</cpu_usage> -->
<!-- </gpu_versions> -->
<!-- </app> -->
<!-- </app_config> -->
<!-- -->
<!-- Include one <app> element for each application you want to configure. -->
<!-- -->
<!-- name -->
<!-- short name of the application. -->
<!-- max_concurrent -->
<!-- the maximum number of tasks of this application to run at a given time. -->
<!-- gpu_usage -->
<!-- the number of GPU instances (possibly fractional) used by GPU versions of this app -->
<!-- cpu_usage -->
<!-- the number of CPU instances (possible fractional) used by GPU versions of this app -->

[OldChap]

<max_concurrent>28</max_concurrent> This one is showing that you want to run a total of 28 threads on your rig.

<gpu_usage>.071400</gpu_usage> This number is showing that you have set up each card to run 14 wu at once

<cpu_usage>.333300</cpu_usage> This number shows me that you want to give each wu 1/36th of a processor to work with??? Make this number of cpu core/threads DIV max_concurrent

Trying for 16 on each card on this rig should give you

<max_concurrent>32</max_concurrent>
<gpu_versions>
<gpu_usage>.0625</gpu_usage>
<cpu_usage>.375</cpu_usage>

[Johnmark]

I tried updating my BOINC client to version 7.0.40 and completed tasks would not upload. Message tab said anonymous platform not supported. Different message that you get now that is ignored so if you're using an app info file do not update. Switched back to 7.0.36 and all is working again.

EDIT: Here's what the error message looks like.
World Community Grid | Message from server: This project doesn't support computers of type anonymous

Just want to check to ensure that the anonymous message no longer appears in the log.

Last night I tried to change a machine over to 7.0.42 and app_config, but was not successful. Noted that error in the log with Red lettering at one point. I ended up going back to the stable release, but then found out the WCG server had run out of WU's None of the machines where getting any new wu's for a while last night.
Plan to give it another try today.

Thanks nanoprobe, worked fine this time with no error what so ever in the log!

[nanoprobe]

If you didn't you need to uninstall the older version of boinc first. Then install the new version. All of your old settings will be saved by default. After the install you must go into the project folder and delete (or remove) the app_info file first. After doing that replace it with the app_config file and then restart boinc. You should see World Community Grid | Found app_config.xml in the event viewer. If you don't see that or you get the red error message something else is wrong. You can't have the app_info and app_config files in the project folder at the same time.

[StitchExperimen]

Old Chap at 16 using your numbers it throws "Computational Errors"
The most I can do is 28 concurrently. 14 in one card and 14 in the other GPU.
My processor isn't strong enough. It pegs it at start up at 100% and take awhile before it eases off.
I try 16 and 14 and that is to many and starts the "Computational Errors".
Thanks Old chap for trying.

[OldChap]

Ok Stitch, it was worth a try.

I feel you need to be testing various different numbers as the usual may not apply here.

What say you run for 2 days at 16 then 8 then 15 then 9 etc. You should get a feel fairly quickly about what runs best for you.

[bluestang]

Can the GFLOPS parameter be put in the app_config file?

I had this setup perfectly matched for my system with app_info. Now I'm seem to be getting a horrible points differential with that same system. Maybe it will settle down over the next day or two, but there shouldn't be any difference since everything is the same.

[StitchExperimen]

Old Chap
I tested and here's the figures I came up with.
4 wu first iteration 50 seconds.
8 wu first iteration 39 seconds, second iteration 34.32 seconds
16 wu first iteration 32.4 seconds, second iteration 31.5 seconds
28 wu first iteration 27.6 seconds, second iteration 27 seconds.

[Yeroon]

Can the GFLOPS parameter be put in the app_config file?

I had this setup perfectly matched for my system with app_info. Now I'm seem to be getting a horrible points differential with that same system. Maybe it will settle down over the next day or two, but there shouldn't be any difference since everything is the same.

With only the gpu multi-treading in the app_config, boinc is now actively adjusting the time to complete the upcoming units to the point where its right on the money. Now I don't know if it counts the total time or divides the estimated time by threads to get an accurate queue based on preferences but it seems to do the cpu tasks fairly accurate as well.

Very happy with 7.0.42, and its new app_config.

[BulldogPO]

Does this work also with nvidia?

[Snow Crash]

Does this work also with nvidia?

Yes ... currently being used on a rig with a 480 and another with both a 660Ti and 670.

[STEvil]

If I never had app_info to start with do I still have to wait for the work to cycle through?

I set up app_config.xml with this but its only starting 1 wu on the 7970 still...

<app_config>
<app>
<name>hcc1</name>
<max_concurrent>3</max_concurrent>
<gpu_versions>
<gpu_usage>.33</gpu_usage>
<cpu_usage>2.67</cpu_usage>
</gpu_versions>
</app>
</app_config>

I'm running a 2600K with HT on. BOINC is set to use 90% of the available CPU (leaves 1 thread open).

[Yeroon]

If I never had app_info to start with do I still have to wait for the work to cycle through?

I set up app_config.xml with this but its only starting 1 wu on the 7970 still...

<app_config>
<app>
<name>hcc1</name>
<max_concurrent>3</max_concurrent>
<gpu_versions>
<gpu_usage>.33</gpu_usage>
<cpu_usage>2.67</cpu_usage>
</gpu_versions>
</app>
</app_config>

I'm running a 2600K with HT on. BOINC is set to use 90% of the available CPU (leaves 1 thread open).

What OS? any chance you have file extensions hidden and its actually app_config.xml.txt?
Did you restart boinc?

[bluestang]

I had to restart BOINC again after applying my app_config on 1 rig, the other took right away.

[STEvil]

Win7-64, yep I restarted BOINC. File extensions are unhidden.

[nanoprobe]

If I never had app_info to start with do I still have to wait for the work to cycle through?

I set up app_config.xml with this but its only starting 1 wu on the 7970 still...

<app_config>
<app>
<name>hcc1</name>
<max_concurrent>3</max_concurrent>
<gpu_versions>
<gpu_usage>.33</gpu_usage>
<cpu_usage>2.67</cpu_usage>
</gpu_versions>
</app>
</app_config>

I'm running a 2600K with HT on. BOINC is set to use 90% of the available CPU (leaves 1 thread open).

I'm only guessing here but from what I read you have the app_config set to run 3 GPU tasks using 2 2/3rds of 90% of you CPU. That may not be a readable configuration and defaults to 1. Try setting the cpu_usage to 3 and set boinc to to use 87.5% of the CPUs.
I'm curious as to why you want to leave 1 CPU thread open since you're only using 3 threads to run GPU tasks?

[Snow Crash]

If you only have HCC selected in your WCG profile and also set "use my cpu" to No then there is no need to allocate more than 1 cpu per gpu WU.
When I put the app_config in place I also updated to 7.0.42 and had to re-start 7.0.42 2 times before it saw the app_config

[STEvil]

I was running it at 90% because leaving 1 thread open seemed to boost GPU output, but that was with my GTX670's.. i've put it back to 100% but it makes no difference, still only 1 thread with the 7970. The thinking behind leaving a thread open was to allow more CPU time to the GPU thread(s) without requiring waiting on other threads to become free from CPU tasks.

cpu_usage calculates based on number of threads of the CPU (# threads/max_concurrent) so doesnt take the 90%/100% setting into account, though it probably affects the number of cores reserved for use by the GPU task(s) which would make my method of using "90%" for CPU tasks as not needed anymore maybe. If I get the GPU tasks running I might have to try adjusting that number..

edit

Can I leave "<name>hcc1</name>" out of the file? I dont want to limit my GPU to only hcc1 tasks, though that seems to be all there is at the moment..

[Rob_B]

I installed 7.0.42 on one of my crunchers today. The one with a 2600k and a 7770.

I have this in the app_config file:

<app_config>
<app>
<name>hcc1</name>
<max_concurrent>4</max_concurrent>
<gpu_versions>
<gpu_usage>.25</gpu_usage>
<cpu_usage>.125</cpu_usage>
</gpu_versions>
</app>
</app_config>

And allow CPU tasks in the device profile for this cruncher.

I thought it would run 4 GPU WU's and 7 CPU WU's. But it is doing 4 GPU and 8 CPU WU's. Is this normal? Just trying to figure some things out before I install it on my other crunchers.

EDIT: I see what I was doing wrong. It should have been <cpu_usage>.25</cpu_usage>.

[nanoprobe]

I was running it at 90% because leaving 1 thread open seemed to boost GPU output, but that was with my GTX670's.. i've put it back to 100% but it makes no difference, still only 1 thread with the 7970. The thinking behind leaving a thread open was to allow more CPU time to the GPU thread(s) without requiring waiting on other threads to become free from CPU tasks.

cpu_usage calculates based on number of threads of the CPU (# threads/max_concurrent) so doesnt take the 90%/100% setting into account, though it probably affects the number of cores reserved for use by the GPU task(s) which would make my method of using "90%" for CPU tasks as not needed anymore maybe. If I get the GPU tasks running I might have to try adjusting that number..

edit

Can I leave "<name>hcc1</name>" out of the file? I dont want to limit my GPU to only hcc1 tasks, though that seems to be all there is at the moment..

If you want to run HCC GPU and other CPU projects then you'll need to add some code to the app_config file for that particular CPU project. Also, if you leave the max_concurrent to 3 then only 3 of your CPU threads will be used not matter what the other settings may be.. Setting max_concurrent to 8, leaving gpu_usage at .33 and setting cpu_usage to 1 should get you 3 GPU tasks running and leave the other 5 threads for whatever else you want to run. Like Snow Crash said you may have to shut down and restart boinc twice for the app_config to start. I had that happen to me a couple of times. First time it did I thought something was wrong with the config file.

[Rob_B]

Do you really need to add info for crunching CPU Wu?s?
I haven?t added any info for other projects and get CPU WU?s on this rig. Thought you only needed to add some thing if you want more control over what is being crunched.

At the moment I have this in an app_config file and I have 4 GPU WU?s and 4 CPU WU?s (from different projects)

<app_config>
<app>
<name>hcc1</name>
<max_concurrent>4</max_concurrent>
<gpu_versions>
<gpu_usage>.25</gpu_usage>
<cpu_usage>1</cpu_usage>
</gpu_versions>
</app>
</app_config>