Autotuning and -a 6 on single GPU

[Mangix]

I have an Nvidia Geforce GTS 450(slow) and I have a question about autotuning. On IRC I asked atom why -a 6 caused autotuning to tune the -n parameter from 8 to 1 and he answered because the GPU did not have enough to do. Here are comparisons between -a 0(mask preapplied) and -a 6. I used the -u 128 parameter since -n was autotuned.

Code:

cudaHashcat-plus64.exe -m 2500 -a 0 handshake.hccap regular.dict
12404/s

cudaHashcat-plus64.exe -m 2500 -a 0 -u 128 handshake.hccap regular.dict
13239/s

cudaHashcat-plus64.exe -m 2500 -a 6 handshake.hccap nonumbers.dict ?d?d?d
9991/s

cudaHashcat-plus64.exe -m 2500 -a 6 -u 128 handshake.hccap nonumbers.dict ?d?d?d
11254/s

Does this mean that atom was correct and the GPU actually does not have enough work or is the limitation intended for multi-GPU systems(which I have read is where the high -n values are problematic)? GPU Utilization drops with -a 6 as well according to the built-in HWMonitor.

As an aside, I've noticed that -a 6 causes a capacitor squeal on my card whereas -a 0 or -a 3 do not. Folding@home does this as well but this is the first time I've noticed it in cudaHashcat.