Unfortunately you can't really measure how long the GPU takes using CPU timers. You can do that to know how long is spent on the CPU in order to process a frame (and you should definitely do this!), but for GPU performance you need to measure the time using the GPU itself.

I recommend having two timers in your engine/game and displaying/logging both: one for your CPU frame time, and one for your GPU frame time. This will help you to immediately know whether you're CPU-bound or GPU-bound, and will give you the right info to make decisions for dynamic resolution. With your current setup you would drop the resolution in the case of something like a physics pile-up taking a large amount of time, even though the GPU might be humming along.

If you use Vsync, you just want to push the resolution up until the frame time suddenly jumps too high. If you can't make your resolution high enough to miss a frame, then great; your game runs at peak frame rate on that system!