Hi there,
Part a) what do I need to do to ensure GPU tasks keep getting loaded ahead before running out? After a few hours of starting a project GPU tasks are not being loaded in front of the GPU, and the GPU sits idle . The scheduler does not seem to keep 1 day ahead of both GPU *and* CPU separately - so GPU does nothing even when I update the project.

Store at least 1.0 days, plus an additional 1.0 days
Run always=1
Use GPU always = 1
cc_config
<use_all_gpus>1</use_all_gpus>

Part b) there is no separate max_concurrent for GPU and CPU tasks where projects have the same work unit name (app name) so max-concurrent applies to (GPU+CPU) - this ends up running CPU only tasks and does not start GPU tasks unless CPU tasks stop loading.

The whole GPU vs CPU scheduling seems to not work properly, given GPU is much faster than CPU for most projects. I would like to configure my GPU's so they never run out of work. Thanks.

No, you could use a max_concurrent statement separately for each task type in an app_config.xml file. Just put in two separate <app_version> statements.
https://boinc.berkeley.edu/wiki/Client_configuration#Application_configuration