I've done something similar. I used the GPU (actually WebGL) and made several images of my geofences where each one was simply drawn as a bitmask (black and whi

I've done something similar. I used the GPU (actually WebGL) and made several images of my geofences where each one was simply drawn as a bitmask (black and white). Then, I wrote a fragment shader to determine if a point is within a geofence by looking the pixel value. I rounded the lat/long pairs to a precision that matched the resolution of my image. Can't remember the exact results but throughput was much higher than what I was able to achieve using highly optimized PostGIS queries.