College of Geo-Exploration Science and Technology, Changchun 130026, China

Large-scale gravity 3D interpretation depends on efficient and high-resolution 3D inversion processing of massive data.The authors applied the conjugate gradient method with minimum support function and prior model constraint to reduce multi-solutions in gravity inversion.Based on the parallel programming and computing platform NVIDIA CUDA with C++language,we achieve fast 3D gravity inversion by adopting GPU parallel technique into the most time consuming part relating to sensitivity matrix.The results of theoretical model show that the abnormal body can be clearly located and the inversion speed is improved greatly.Comparing with inversion speed of the Matlab program,speed of inversion with GPU parallel technique has improved more than 100 times under the hardware condition of Geforce GTX 1060 graphics card.