Abstract: To solve load imbalance problem among the nodes of rendering clusters, a sort-first rendering cluster was proposed. The deferred shading was used as major rendering method in the graphics applications of the proposed rendering cluster. Algorithms of deferred shading were classified as visibility testing, global pixel shading and local pixel shading. These algorithms were used to predict rendering load and converted to fragment shadings of rendering. Referring to the parallel computation performance of graphics processing unit (GPU), the sort-first rendering cluster predicted per-pixel rendering load within the whole screen space. On the base of the rendering load distribution, the proposed system subdivided rendering tasks and sorted them to each node of the cluster. According to the experiment results, the rendering load distribution can be predicted in real-time, the rendering tasks can be subdivided by the proposed sort-first rendering cluster. The higher and stable rendering performance of the system was obtained by the accurate load prediction.