https://community.intel.com/t5/Intel-oneAPI-Math-Kernel-Library/dpptrf-performance/m-p/986053#M17767 <DIV>Hi,</DIV> <DIV></DIV> <DIV>We are currently developping an application that needs to solve a linear systems of equations.</DIV> <DIV></DIV> <DIV>As the matrix A is a symmetric positive definite matrix, we call the dpotrf and dpotrs methods,butwhen the size of the matrix is too big (order 15.000 and more), we use the packed form of the matrix (dpptrf and then dpptrs methods are called).</DIV> <DIV></DIV> <DIV>However, when we measure the performances, we are observing thatfor big matrices, the packed method is fours time slower than the non packed method .</DIV> <DIV></DIV> <DIV>Does anybody have an idea on how to improve the performnce when using the packed methods?</DIV> <DIV></DIV> <DIV>Thanks,</DIV> <DIV></DIV> <DIV>Daniel</DIV> Wed, 06 Jul 2005 19:42:33 GMT danrod 2005-07-06T19:42:33Z https://community.intel.com/t5/Intel-oneAPI-Math-Kernel-Library/dpptrf-performance/m-p/986053#M17767 <DIV>Hi,</DIV> <DIV></DIV> <DIV>We are currently developping an application that needs to solve a linear systems of equations.</DIV> <DIV></DIV> <DIV>As the matrix A is a symmetric positive definite matrix, we call the dpotrf and dpotrs methods,butwhen the size of the matrix is too big (order 15.000 and more), we use the packed form of the matrix (dpptrf and then dpptrs methods are called).</DIV> <DIV></DIV> <DIV>However, when we measure the performances, we are observing thatfor big matrices, the packed method is fours time slower than the non packed method .</DIV> <DIV></DIV> <DIV>Does anybody have an idea on how to improve the performnce when using the packed methods?</DIV> <DIV></DIV> <DIV>Thanks,</DIV> <DIV></DIV> <DIV>Daniel</DIV> Wed, 06 Jul 2005 19:42:33 GMT https://community.intel.com/t5/Intel-oneAPI-Math-Kernel-Library/dpptrf-performance/m-p/986053#M17767 danrod 2005-07-06T19:42:33Z https://community.intel.com/t5/Intel-oneAPI-Math-Kernel-Library/dpptrf-performance/m-p/986054#M17768 <DIV>Hi, Daniel.</DIV> <DIV></DIV> <DIV>Working on the packed matrix is by defaultslower than on unpacked. If you have enough memory to keep the matrix unpacked, it'sgood idea to apply DPOTRF/DPOTRS instead of DPPTRF/DPPTRS.</DIV> <DIV></DIV> <DIV>Michael.</DIV> Mon, 18 Jul 2005 21:18:49 GMT https://community.intel.com/t5/Intel-oneAPI-Math-Kernel-Library/dpptrf-performance/m-p/986054#M17768 michael_chuvelev 2005-07-18T21:18:49Z