GVF Snake is one of the most widely-used edge detection algorithms, nevertheless subject to its slow computation. This study reveals the bottleneck and transfers the time-consuming part of this algorithm to the GPU for better performance. In detail, this algorithm is decomposed into three parts, (1) GVF Computation, (2) inversing a circulant matrix and (3) curve deformation. All of these parts are analyzed and designed to run on the GPU via suitable data structures and corresponding operations. With the help of parallel computational power of the GPU, our improved algorithm could be about 15 times as fast as is executed on the CPU.