With　the　development　of　GPU,　the　GPU＇s　float-point　computing　capacity　improves　rapidly.　How　to　apply　the　float-point　ability　of　GPU　to　the　non-graphic　computing　field　becomes　a　highlight　in　the　research　of　high　performance　computing.　Jacobi　is　a　typical　application　in　scientific　computing.　This　paper　designs　and　implements　Jacobi　Algorithm　on　Nvidia＇s　CUDA　platform　and　gets　a　good　speedup　compared　with　the　CPU.　©　2010　IEEE.