The RAC is climbing again, and I am currently running 2 tasks per GPU, and at least when it was running CUDA, .5 core per GPU, but not sure if SOG defaults to 1-1 or if it has to be configed like CUDA. Regardless, it appears to be running ok right now, so I am going to leave it alone for a while to stabilize, and will look into tinkering with it once its been up and running for a while with none of those catistrophic GPU errors again. Crossing my fingers that that was a one time occurance...

---

Depends on whether you had the CUDA tasks running two up via a <count> entry in app_info directly or via an app_config. When you run the Lunatics installer it defaults to <count>1</count> entry in the MB section of the app_info. If you simply did a Replace of the r3557 app with the r3584 app, it would have left that as default. If you are running two up now with the SoG app, either you modified the count or have a MB app section defined to run two up in a app_config file.

I would let it run for a while so you can observe the trend. It should increase your RAC and that can be increased a great deal more with tuning parameters that the SoG app provides. There is quite a bit more tuning available compared to the CUDA50 app.

The SoG app will take whatever CPU core resource that it needs. All the .5 CPU resource does is give BOINC an idea of how many CPU tasks it should start. If you check with Task Manager or any of the other tools available to monitor BOINC, you will see that each GPU task is allocated almost a full CPU core. You can set that up explicitly in an app_config.

This is my app_config for example:

<app_config>
  <app_version>
    <app_name>setiathome_v8</app_name>
    <plan_class>opencl_nvidia_SoG</plan_class>
    <avg_ncpus>1</avg_ncpus>
    <ngpus>0.5</ngpus>
    <cmdline>-sbs 2048 -period_iterations_num 2 -tt 1500 -high_perf -spike_fft_thresh 4096 -tune 1 64 1 4 -oclfft_tune_gr 256 -oclfft_tune_lr 16 -oclfft_tune_wg 256 -oclfft_tune_ls 512 -oclfft_tune_bn 64 -oclfft_tune_cw 64 -high_prec_timer</cmdline>
  </app_version>
  <app_version>
    <app_name>astropulse_v7</app_name>
    <plan_class>opencl_nvidia_100</plan_class>
    <avg_ncpus>1</avg_ncpus>
    <ngpus>0.5</ngpus>
    <cmdline>-unroll 24 -oclFFT_plan 256 16 256 -ffa_block 16384 -ffa_block_fetch 8192 -tune 1 64 8 1 -tune 2 64 8 1</cmdline>
  </app_version>
<project_max_concurrent>7</project_max_concurrent>
</app_config>

You can see my callout for 1 CPU core in both the MB and AP sections.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

EDIT- I usually exit and restart BOINC to make sure it gets read- not sure if Options, Read Config files works with that file.

Look just above "Logging flags" here.