The micro-CT of the X-ray imaging beamline at Shanghai Synchrotron Radiation Facility can achieve sub-micron resolution

while the data size for a single set of CT projection is often dozens of giga bytes.With the serials methods

several hours of CPU time are usually needed for processing a single set of data

while the beamline operates 24 hours a day and seven days a week

with dozens of CT data sets being collected by users every day.Conventional methods based on CPU are no longer applicable for processing such massive data sets.GPU is a highly parallel stream processor

which is capable of accelerating the CT reconstruction significantly.In this paper

CUDA programming standard is used to program the paralleled CT reconstruction.The test results show that the program running in a NVDIA GTX295 GPU can accelerate the CT reconstruction by a factor of 150.The reconstructing time could be shortened to 15 minutes or around.The reconstruction speed shall be further improved by a GPU server with more pieces of GPU chips.