There is a much easier way without messing around in app_config.xml

When logged in at the WCG website, select "Settings" - "Device Manager" - click the profile of your device - set "Allow research to run on my CPU?" to "No"

Done.

That's a nice out-side-of-box thinking :-) How ever I'm afraid the maximum value you can use is 1.

I don't think there's a easy way to solve this. You either need a second client running on GPU only profile, and while the primary client would run CPU only tasks.

The easiest solution would be limiting number of concurrent OPN tasks from the device profile to 1, while allowing more tasks to other projects you are contributing.

Yea, I know about running multiple GPU tasks at the same time on the same GPU card. I just want to run only GPU tasks. I suspect I could easily run multiple GPU tasks at once but I'm a real neophyte with GPU stuff. Limiting the project to one concurrent tasks in the device profile seems to limit GPU tasks as well (makes sense). I figured an app_config file would be necessary. Here's the Boinc log from a startup:

w-archlaptop

1 7/7/2021 1:43:54 PM Starting BOINC client version 7.16.11 for windows_x86_64
2 7/7/2021 1:43:54 PM log flags: file_xfer, sched_ops, task
3 7/7/2021 1:43:54 PM Libraries: libcurl/7.47.1 OpenSSL/1.0.2s zlib/1.2.8
4 7/7/2021 1:43:54 PM Data directory: C:\ProgramData\BOINC
5 7/7/2021 1:43:54 PM Running under account ArchLaptop
6 7/7/2021 1:43:56 PM CUDA: NVIDIA GPU 0: GeForce GTX 1650 Ti with Max-Q Design (driver version 452.66, CUDA version 11.0, compute capability 7.5, 4096MB, 3324MB available, 2458 GFLOPS peak)
7 7/7/2021 1:43:56 PM OpenCL: NVIDIA GPU 0: GeForce GTX 1650 Ti with Max-Q Design (driver version 452.66, device version OpenCL 1.2 CUDA, 4096MB, 3324MB available, 2458 GFLOPS peak)
8 7/7/2021 1:43:56 PM OpenCL: Intel GPU 0: Intel(R) UHD Graphics (driver version 27.20.100.9316, device version OpenCL 3.0 NEO, 26102MB, 26102MB available, 240 GFLOPS peak)
9 7/7/2021 1:43:56 PM Windows processor group 0: 16 processors
10 7/7/2021 1:43:56 PM Host name: LAPTOP-4H1RR6MA
11 7/7/2021 1:43:56 PM Processor: 16 GenuineIntel Intel(R) Core(TM) i9-10885H CPU @ 2.40GHz [Family 6 Model 165 Stepping 2]
12 7/7/2021 1:43:56 PM Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss htt tm pni ssse3 fma cx16 sse4_1 sse4_2 movebe popcnt aes f16c rdrandsyscall nx lm avx avx2 vmx smx tm2 pbe fsgsbase bmi1 sme
13 7/7/2021 1:43:56 PM OS: Microsoft Windows 10: Professional x64 Edition, (10.00.19043.00)
14 7/7/2021 1:43:56 PM Memory: 63.73 GB physical, 73.23 GB virtual
15 7/7/2021 1:43:56 PM Disk: 1.86 TB total, 1.73 TB free
16 7/7/2021 1:43:56 PM Local time is UTC -4 hours

What about setting max_concurrent to 0.5?

<app_config>
   <app>
      <name>opn1</name>
      <max_concurrent>1</max_concurrent>
   </app>
   <app>
      <name>opng</name>
      <gpu_versions>
      	<gpu_usage>0.25</gpu_usage>
      	<cpu_usage>1</cpu_usage>
      </gpu_versions>
    </app>
</app_config>

Well, I tried 0.5 - or even 0.00001 - and it acted as if the value was the same as before. So, a fraction doesn't get rounded down.

Nanoprobe, the thread list says you updated this thread on the 10th but I can't see a post for you that day. Weird!

Anyway, not sure I'm making much progress but making some. This is what I have for my app_config.xml:

<app_config>
<app>
<name>opn1</name>
<max_concurrent>5</max_concurrent>
<gpu_usage>0.000000</gpu_usage>
<cpu_usage>0.000000</cpu_usage>
</app>
<app>
<name>opng</name>
<max_concurrent>25</max_concurrent>
<gpu_usage>0.250000</gpu_usage>
<cpu_usage>0.500000</cpu_usage>
</app>
</app_config>

I also added <use_all_gpus>1</use_all_gpus> in the cc_config.xml file on that machine. You don't want to set the project limit in the device profile to 1 as that applies to both CPU and GPU. I have that set to 10. I wasn't seeing any GPUs come in with it set to 1. However, with it set to 10, I currently have 3 running and 7 ready to start and two GPUs which is more than 10. More weirdness. The max_concurrent of 5 at least runs enough of the 10 allowed to keep any from missing the deadline. In the log, I am seeing:

Requesting new tasks for CPU and NVIDIA GPU and Intel GPU

I don't think it was mentioning Intel GPU before but I am not 100% sure. I'll try to look at the stdoutdae on that machine. I noticed that BOINCMGR has "Running (0.892 CPUs + 1 GPU) in the status column. That's why I put 0.5 in for cpu_usage for OPNG. I haven't seen the machine run more than one GPU at a time so maybe the NVIDIA GPU can only do one. If that's true, it would seem that there is no point to have gpu_usage and cpu_usage for opng in the app_config.

Edit: Checked stdoutdae on that machine and it had the same "Requesting new tasks for CPU and NVIDIA GPU and Intel GPU" message before I added use_all_gpus to cc_config.xml. Oh, and yes, I did restart BOINC.

Edit: I hate trying to think! The project limit in the device profile can only apply to CPU tasks. I have 10 CPU tasks on the machine now and it's got GPU tasks as well.