Opposed to some other GPU sciences, WCG's HCC-GPU project needs a CPU core for the largest portion at it's disposal for the start, middle and end section [when it uses 100% of the spare cycles on a processor thread]. Depending on the settings, what could happen is that the GPU tasks, each time it grabs the CPU thread again, the CPU only task gets suspended, and unloaded [which is default], and if that happens continuously, with such short tasks, it could lead to the client considering too many interruptions for a task and cancelling the job. Nothing gained, rather a bad task added to the repair cycle. [My theory]