In this paper we designed a remote sensing image data fusion algorithm on GPU (Graphics Processing Unit) using the programmability of GPU which is a parallel vector processor. Both of the forward IHS and inverse IHS transform computation were mapped into GPU. We realized parallel rendering and output of the three components of the IHS, and the inverse IHS transformation using RTT (Render To Texture) and MTR (Multiple Render Targets) method to speed up the computation. The result shows that this algorithm on GPU is much faster than the previous CPU-based IHS transforming algorithm in the case of large data. And with the fusion images getting bigger the advantage of the velocity is more obvious in GPU implementation.