Low computational efficiency is the largest obstacle hindering practical applications of Monte Carlo particle transport simulation. We develop a new MC scheme for photon transport that employs Metropolis-Hasting sampling algorithm to conduct the simulations via a path-by-path scheme in this GPU-based Metropolis MC (gMMC) package. By using the Metropolis algorithm, gMMC is able to sample an entire path of a photon each time, extending from the x-ray source to the detector. The sampled paths over a long run will follow a distribution governed by the particle transport physics. gMMC was benchmarked against an in-house developed GPU-based Monte Carol simulation tool gMCDRR that performs simulations in the conventional particle-by-particle scheme. An example problem of x-ray scatter calculation was studied. For pencil beam case, speed-up factors of 200 to 300 times for first order scatter and 12 to 18 times for second order scatter can be achieved, while the average differences comparing to the conventional approach are within 1%.