In order to solve the problem of video memory limitation during GPU-based visualization of large scale volume data, a compressing and rendering strategy was proposed. Volume data was compressed by combining wavelet transform with vector quantization algorithm, and then transferred to GPU totally. When rendering, a GPU-based ray casting algorithm was adopted and only few current needed data was decompressed and transformed. In order to realize real-time interactive visualization, a multi-resolution rendering method was designed. Experimental results based on CUDA show that this strategy can decrease the video memory consumption effectively and obtain better image quality while ensuring interactive rendering frame rate.