Precision and speed are the two aims in the field of computational fluid dynamics (CFD). In order to obtain a physically accurate solution of compressible flow, the common way is to use a highly resolved mesh within the whole flow filed, such as Direct Numerical Simulations (DNS) which uses large uniform grids to resolve all the present scales within the flow. In this processing, High Resolution Schemes have already be employed to modify and increase the amount of numerical dissipation in the neighborhood of a discontinuity, but actually they are very expensive and often cost several weeks, even months to finish an industry simulation. As we know, the compressible flow field has a very severe inhomogeneity and the high gradient just exists in very small regions, so uniform grids may cause unnecessary refinement in most flow regions which can lead very low efficiency, especially for multidimensional problems. Therefore, adaptive simulation techniques, such as Adaptive Mesh Refinement (AMR), or multi-resolution techniques using Wavelets have been introduced to dynamically adjust the resolution of different regions or cells according to error analysis of state values from different length scales. Although with the same precision, these adaptive grid methods can significantly improve the speed, it’s still far from the interest of industrial applications. With the development of computer hardware, almost all kinds of processors (CPU and GPU) start to support parallel computing, which can execute many threads at the same time. Several researchers have demonstrated that GPU offer one to two orders of magnitude speedup over respective CPU implementations, and is an expert in the filed of DNS. However, all the adaptive methods contain a lot of logical events like checking special spot and updating dynamic flow structure of the whole domain. So the best choice for CFD is to combine the adaptive methods with parallel computing together, which can not only save computational time but also Data Storage Media. It will make simulations on desktop or laptop come true. In this project, we would develop a new solver, with using multi-cores in charge of logical events and employing multi-GPUs to handle the expensive high resolution numerical computation, to divide a huge domain into several sub-zones and deal with them on different devices together in order for industrial applications. Our design will be case-oriented, and contain several popular high resolution numerical schemes.