GPU beta still running and re-sends are about.
Some are timing out on slower systems.
I have some on PCs with Intel GPUs.
Linux systems appear to be completing OK although some have driver i/o errors.

Paul.

Now in production but limited units until proven.
Powerful GPUs may need a dedicated core, mine don't.
I use app_config to allow more to run, example below.
(note Boinc Tasks edit breaks the GPU section, OK for CPU).

<app_config>
<app>
<name>arp1</name>
<max_concurrent>1</max_concurrent>
</app>
<app>
<name>mip1</name>
<max_concurrent>1</max_concurrent>
</app>
<app>
<name>opn1</name>
<max_concurrent>4</max_concurrent>
</app>
<app>
<name>opng</name>
<gpu_versions>
<gpu_usage>.25</gpu_usage>
<cpu_usage>.1</cpu_usage>
</gpu_versions>
</app>
<app>
<name>beta29</name>
<gpu_versions>
<gpu_usage>.25</gpu_usage>
<cpu_usage>.1</cpu_usage>
</gpu_versions>
</app>
</app_config>

Edit: remove mangled GPU bits from CPU tasks that break later BOINC versions. Mangled by edit using BoincTasks.

Paul.

Current stress test of WCG and Open Pandemics systems are paying handsome dividends for me. Only got one GPU new enough, but it's pretty decent and is powering through those units. I've used a config app file to run multiple GPU units concurrently so the GPU is maxed out since there is a lot of idle time if running only one.

I've been tweaking it up over the last couple of days, seems 14 GPU units running together keeps it going flat out.

I'm looking forward to the stats run tonight.

I didn't change any proportions, but I have seen some GPU NVIDIA jobs running for real over the last few days, and the points is more than double.

Beta just sat there, but these things seem to be working.
I am only seeing one GPU run at a time per computer, so that will do for awhile.
I noted the app config above, but the message wrapped up with a mangling factor.

I will skip GPU for an ancient Vista machine and a laptop.

A couple of my machines are Ryzen 7 game machines with upper end GPUs so they are good candidates. I bought them at a good price from a reputable local CPU repair/builder.

It kind of looks like I would just have to change the GPU to .1 to run 10 simultaneously.
Some graphics cards need a dedicated CPU thread. Apparently we need to give it some kind of chunk of the CPU. What if the CPU line were left out?
The app_config.xml above has some more references for each project app,open CPU, open GPU (which I extracted) plus GPU beta, but I have the four BOINC divisions set to run so many of each one. I have already cut back the counts a bit since they have more GPU activity. The GPU queue is a separate animal. It looks as if the tools is set to read the app_config file for each computer if desired to put in this change in the c:\ {BOINC} {DATA} directory for each one.

<app_config>
<app>
<name>opng</name>
<gpu_versions>
<gpu_usage>.1</gpu_usage>
<cpu_usage>.1</cpu_usage>
</gpu_versions>
</app>
</app_config>