GPU clients also use some CPU time. Sometimes it helps to let a GPU client take all of one CPU, though giving it more usually doesn't help. Often, giving it one hyper-threaded core is plenty. Changing the number or percent of cores used is better than changing the % CPU time, as that just makes the entire CPU idle periodically.
In your case, you have a 690 which is really two GPUs. So you should experiment to see whether your particular project uses enough CPU that giving it two cores helps or not. And whether those cores need to be real or hyper-threaded.