There is only 1 GPU task running using 1 core of an 8 thread CPU. I'll just leave it running for now. I switched projects here so I'll give it time to figure it out. Still seems weird to me.

Project not being highest in fetch order and the described processing order disparity are different 'issues'. Seen the HP processing many times myself, caused by the relation of share time and projected runtime, 5% v 95% a good example. How do you squeeze 12 hours of work in a 5% share per day with a deadline just few days away, adding a large cache as further pressure point. Typically if with 7 and having Minimum buffer set at ~2 days, all WCG repair jobs will jump queue and process high priority... perfect when Beta testing ;].

Long as there is no idle resource [no mention], no reason the fetch work from a project that has had it's share. Suspend GPUGRID task fetch and you have a workaround, else set various log flags such as work_fetch debug and post log. It will confirm what the summary message "Project not highest priority" already signals. Advertising v7 [saw a note to that effect by OP poster]... I'd be careful, since WCG's own kit is testing at 7.0.68 and seeing dev notes that prepping for boarding 7.2 is underway... the scheduler had more overhauls. Running 7.2.5 [in test] myself and WCG 7.2.7 is out... test it if you like and report to the alpha mail list on finding... maybe it's a lead up to finally getting a formal test at WCG [Lawrence, we may be able to stop sighing ;>]

Edit: The WCG kit only comes in 32 bit for Windows. Makes no difference as it supports 64 bit sciences anyhow.