Suspend Project, preferably GPU's, Certain Hours During Weekdays

Seeing as you are running Linux, you can easily write small scripts to get boinccmd to independently control CPU and GPU.  Look in the documentation for the separate control options --set_run_mode and --set_gpu_mode.  These options take a single argument, one of { auto | always | never } with a further optional [duration].  If you give a duration, the mode change gets reverted at the end of that duration.  So you could set up a cron job to launch the script at 3pm (using 'never' to suspend crunching) and with the appropriate duration (18000) included, it would resume crunching at 8pm after 5 hrs had elapsed.

I have a little bit of experience at this sort of thing but not using cron.  Maintaining a cron script on a very large number of individual machines is not a very appealing prospect.  Running a script on a single machine and having that script talk to all machines on the LAN over ssh connections is much more efficient for me.  I do it for temperature control during heat waves.  I don't use aircon to keep machines cool - just forced ventilation.  If the ambient in the machine room gets above 38C, a script gets launched on a server machine which suspends all crunching.  The forced ventilation works quite quickly to bring the temperature down to a more reasonable level.

I could be wrong but I suspect that you may see a better power saving by stopping CPU crunching rather than GPU crunching.  Your 1050Tis would be quite productive (I'm guessing from what others say - I don't own any) and quite miserly on power used.  In particular, your 8 core 8350 would use lots of power when crunching CPU tasks on all available cores.  Unfortunately 750Tis are not very productive with current GPU tasks.  I suspect that you would find that a single 1050Ti would use much less power and produce more output than 2 750Tis.

Do you happen to have access to a kill-a-watt style power meter?  If you do, you should try using it to measure the power saving you could achieve.  Start off by measuring the power consumption at the wall with everything running normally.  Then select and suspend all onboard 'ready to start' tasks (both CPU and GPU) so that new ones can't start when you suspend running tasks.  If you then suspend the running GPU tasks, you will be able to measure how much you are saving by stopping GPU tasks.

When you have that new figure, suspend the running CPU tasks so that nothing is now crunching.  You will see a further drop in power consumption which may give you enough information.  You could then resume just the GPU tasks that were running previously and measure that power increase before resuming everything and going back to full normal crunching.  From all the figures collected, you should be able to work out your best option for saving power without crippling your completed task throughput.

Remember, there is a lot more science being done in a FGRP GPU task than a FGRP CPU task.  You can tell this from the different credit awards.

You can't suspend BOINC without interfering with all crunching.  You suspend a particular science application but BOINC must continue to run to look after apps that aren't suspended and to manage things when a suspended app gets resumed.  BOINC itself consumes very little power anyway.