An Einstein Schizoid Embolism?

I reset the project on both computers and they are both performing as expected for GPU tasks, but the CPU tasks went away and don't appear to be coming back.  I will post the contents from the 3080ti machine as they are essentially the same.

<app_config>
   <max_concurrent>12</max_concurrent>
   <gpu_usage>.5</gpu_usage>
   <avg_ncpus>8</avg_ncpus>
   <cpu_usage>.5</cpu_usage>
</app_config>

<cc_config>
    <log_flags>
        <file_xfer>1</file_xfer>
        <sched_ops>1</sched_ops>
        <task>1</task>
        <app_msg_receive>0</app_msg_receive>
        <app_msg_send>0</app_msg_send>
        <async_file_debug>0</async_file_debug>
        <benchmark_debug>0</benchmark_debug>
        <checkpoint_debug>0</checkpoint_debug>
        <coproc_debug>0</coproc_debug>
        <cpu_sched>0</cpu_sched>
        <cpu_sched_debug>0</cpu_sched_debug>
        <cpu_sched_status>0</cpu_sched_status>
        <dcf_debug>0</dcf_debug>
        <disk_usage_debug>0</disk_usage_debug>
        <file_xfer_debug>0</file_xfer_debug>
        <gui_rpc_debug>0</gui_rpc_debug>
        <heartbeat_debug>0</heartbeat_debug>
        <http_debug>0</http_debug>
        <http_xfer_debug>0</http_xfer_debug>
        <idle_detection_debug>0</idle_detection_debug>
        <mem_usage_debug>0</mem_usage_debug>
        <network_status_debug>0</network_status_debug>
        <notice_debug>0</notice_debug>
        <poll_debug>0</poll_debug>
        <priority_debug>0</priority_debug>
        <proxy_debug>0</proxy_debug>
        <rr_simulation>0</rr_simulation>
        <rrsim_detail>0</rrsim_detail>
        <sched_op_debug>0</sched_op_debug>
        <scrsave_debug>0</scrsave_debug>
        <slot_debug>0</slot_debug>
        <state_debug>0</state_debug>
        <statefile_debug>0</statefile_debug>
        <suspend_debug>0</suspend_debug>
        <task_debug>0</task_debug>
        <time_debug>0</time_debug>
        <trickle_debug>0</trickle_debug>
        <unparsed_xml>0</unparsed_xml>
        <work_fetch_debug>0</work_fetch_debug>
    </log_flags>
    <options>
        <abort_jobs_on_exit>0</abort_jobs_on_exit>
        <allow_multiple_clients>0</allow_multiple_clients>
        <allow_remote_gui_rpc>0</allow_remote_gui_rpc>
        <disallow_attach>0</disallow_attach>
        <dont_check_file_sizes>0</dont_check_file_sizes>
        <dont_contact_ref_site>0</dont_contact_ref_site>
        <lower_client_priority>0</lower_client_priority>
        <dont_suspend_nci>0</dont_suspend_nci>
        <dont_use_vbox>0</dont_use_vbox>
        <dont_use_wsl>0</dont_use_wsl>
        <exclusive_gpu_app>Recode.exe</exclusive_gpu_app>
        <exit_after_finish>0</exit_after_finish>
        <exit_before_start>0</exit_before_start>
        <exit_when_idle>0</exit_when_idle>
        <fetch_minimal_work>0</fetch_minimal_work>
        <fetch_on_update>0</fetch_on_update>
        <force_auth>default</force_auth>
        <http_1_0>0</http_1_0>
        <http_transfer_timeout>300</http_transfer_timeout>
        <http_transfer_timeout_bps>10</http_transfer_timeout_bps>
        <max_event_log_lines>2000</max_event_log_lines>
        <max_file_xfers>8</max_file_xfers>
        <max_file_xfers_per_project>2</max_file_xfers_per_project>
        <max_stderr_file_size>0</max_stderr_file_size>
        <max_stdout_file_size>0</max_stdout_file_size>
        <max_tasks_reported>0</max_tasks_reported>
        <ncpus>0</ncpus>
        <no_alt_platform>0</no_alt_platform>
        <no_gpus>0</no_gpus>
        <no_info_fetch>0</no_info_fetch>
        <no_opencl>0</no_opencl>
        <no_priority_change>0</no_priority_change>
        <os_random_only>0</os_random_only>
        <process_priority>0</process_priority>
        <process_priority_special>0</process_priority_special>
<proxy_info>
    <socks_server_name></socks_server_name>
    <socks_server_port>80</socks_server_port>
    <http_server_name></http_server_name>
    <http_server_port>80</http_server_port>
    <socks5_user_name></socks5_user_name>
    <socks5_user_passwd></socks5_user_passwd>
    <socks5_remote_dns>0</socks5_remote_dns>
    <http_user_name></http_user_name>
    <http_user_passwd></http_user_passwd>
    <no_proxy></no_proxy>
    <no_autodetect>0</no_autodetect>
</proxy_info>
        <rec_half_life_days>10.000000</rec_half_life_days>
        <report_results_immediately>0</report_results_immediately>
        <run_apps_manually>0</run_apps_manually>
        <save_stats_days>30</save_stats_days>
        <skip_cpu_benchmarks>0</skip_cpu_benchmarks>
        <simple_gui_only>0</simple_gui_only>
        <start_delay>0.000000</start_delay>
        <stderr_head>0</stderr_head>
        <suppress_net_info>0</suppress_net_info>
        <unsigned_apps_ok>0</unsigned_apps_ok>
        <use_all_gpus>0</use_all_gpus>
        <use_certs>0</use_certs>
        <use_certs_only>0</use_certs_only>
        <vbox_window>0</vbox_window>
    </options>
</cc_config>

Post Project reset the 3080ti is now running 14 CPU Einstein Jobs and 2 GPU jobs.  But it was doing that before as well, and then it decided to stop doing GPU work and only do CPU work.  Plus if you look at the number of tasks running I don't understand why it's still running 14 CPU tasks.

I believe your app_config is not setup properly. 
 

you’re mixing different tags without grouping them appropriately. 
 

see examples here, scroll down to “Project-level Configuration”: https://boinc.berkeley.edu/wiki/Client_configuration
 

you need to group these commands under <app> or <app_version> Elements. And make sure you use the right combination of tags for each element type. (ngpus and avg_ncpus are not valid for the <app> element for example) 

_________________________________________________________________________

try this:

this will run gamma ray tasks at 2x per GPU, and gravitational wave tasks at 1x per GPU. there's no benefit or reason to change cpu_usage to anything other than 1.0 since the application will use a full CPU core, and it's better for BOINC to properly account for that. change the <gpu_usage> values to whatever is appropriate for what number of multiples

you can certainly use the "use at most CPU%" section in compute preferences of BOINC Manager to limit total CPU use. but if you want to set Einstein to be limited to a certain number of jobs, you can do add a section for the CPU app (i dont know the proper app name for the CPU FGRP5 app, "hsgamma_FGRP5"? check your client state file for the real name.), or just limit Einstein globally for whatever number of jobs including GPU jobs that you want to run, with project_max_concurrent)

for example, if you want to run 2x gamma ray, and 12 Einstein jobs total, meaning 10 Einstein CPU jobs will run, you can use this as an example:

adjust as needed and pay close attention to syntax and placement of various elements.

_________________________________________________________________________