The <max_concurrent>n</max_concurrent> is functional *only* for CPU exclusive tasks, and now wonder what happens if you set this tag to zero. Suppose you get due shortage of the selected projects, something else via the "if there is no work for...", but you explicitly don't want CEP2 [too heavy] or HPF2 [because they always fail anyhow due the :711 bug]. Will a section like the below for hpf2 cause them to just sit there being skipped until the user aborts, or the task expires and is eventually auto-aborted?

<app>
<name>hpf2</name>
<max_concurrent>0</max_concurrent>
</app>

Going to test this [not the intend of the function, but while we're at it finding the holes and gotchas and bonus side... still happily running maximum of 4 of each science on my octo system, never an idle core with the 5 way science mix, really 4 as CEP2 is limited to 1 core.].

Have you tried this file (I am guessing you would like to run HCC GPU only + the rest of HFCC CPU)?

<app_config>
<app>
<name>hcc1</name>
<gpu_versions>
<gpu_usage>.083</gpu_usage>
<cpu_usage>.5</cpu_usage>
</gpu_versions>
</app>
<app>
<name>hfcc</name>
<max_concurrent>6</max_concurrent>
</app>
</app_config>

Disclaimer: I am not using 7.0.42 yet and I have not tested it. Use of the a.m. code is on your sole responsibility

Looking at the time of me writing the post saying that I tried the 0.5 setting and comparing the results status page I can see that those that ended up in an errored state were both before and after trying the 0.5 setting, while the invalids seem to be only after that setting. And at some point during the night finally the PC just crashed with a black screen.

Perhaps trying to go back to 7.0.36 or 7.0.38 would be interesting to see if I can get it working there again....since I also updated from catalyst 12.10 to 12.11 beta 11 before I updated BOINC.

OK, setting value to zero is as tested equal to "no limit". Have asked for an altogether block value, so always/frequent failing sciences wont start at all... they can expire or be aborted should they slip through the "if there is no work for..."

(Yeah, yeah I know about what You want on the device profiles, but having the safety on the client would stop them rogues for sure ;P)

Well, some of the advantages are:
1: Adding app_config.xml won't lose you any work. Even if you types an error you won't lose all cached work like is very common for app_info.xml
2: Removing app_config.xml won't lose all work, like always happens if removes app_info.xml
3: You can continue running other applications for the same BOINC-project without needing to add them to your app_config.xml
4: The client will automatically download new application-version and use your changed settings. There's no delay with possible no work since runs with outdated app_info.xml, and there's no manual editing & downloading of new files, with the possibilities of making an error, like it's for app_info.xml

Advantages with app_info.xml over app_config.xml:
5: Can run better optimized applications with app_info.xml. Not an option for most projects and not for WCG.
6: Can run applications on platform/OS-combination the project doesn't have an applicaton themselves. Not an option for most projects and not for WCG.
7: For very security-oriented people, can make sure only applications verified themselves runs, and there's no problems with suddenly "new" application being downloaded.
8: Can specify to example only download HCC-work to GPU and not CPU, but still continue downloading work for other CPU-applications. Similarly, can specify to only run the 64-bit applications on available but also using the 32-bit application in case no 64-bit is available.
9: If have both Nvida & Amd in same computer, can specify different parameters for CPU & GPU-usage.

Both methods handle the "run N tasks simultaneously on a single GPU", and for most users the method to add this option is "to copy & paste a message from the forums, and save it at the correct location". For this the app_config.xml have major advantages, as indicated by #1, #3 and #4.

app_info.xml on the other hand has a big advantage with #5... but, since WCG doesn't supply the application-code where's no optimizations available, so #5 isn't a consideration for WCG-users.