Modern programmable graphics processing units (GPUs) provide increasingly higher performance, motivating us to perform general-purpose computation on the GPU (GPGPU) beyond graphics applications. In this paper, we address the problem of resource selection in the GPU grid. The GPU grid here consists of desktop computers at home and the office, utilizing idle GPUs and CPUs as computational engines for compute-intensive applications. Our method tackles this challenging problem (1) by defining idle resources and (2) by developing a resource selection method based on a screensaver approach with low-overhead sensors. The sensors detect idle GPUs by checking video random access memory (VRAM) usage and CPU usage on each computer. Detected resources are then selected according to a matchmaking framework and benchmark results obtained when the screensaver is installed on the machines. The experimental results show that our method achieves a low overhead of at most 262 ms, minimizing interference to resource owners with at most 10% performance drop.