4 Replies Latest reply on Jan 19, 2015 4:55 AM by yurtesen

    OpenCL Linpack

    yurtesen

      Hello,

       

      I was looking at this article:

      http://developer.amd.com/community/blog/2013/12/17/students_win_sc13_commodity_cluster_challenge_with_amd_apu/

       

      I was wondering from where one can obtain the linpack code they used?

       

      Thanks!

        • Re: OpenCL Linpack
          pinform

          Feel free to get in touch with Neel and Tushar directly!

          1 of 1 people found this helpful
            • Re: OpenCL Linpack
              yurtesen

              Thanks! Tushar sent the link to the wrapper they used

              cl-linpack-wrap - wrappers for clAMDBlas - Google Project Hosting

               

              I managed to compile everything and I am able to run original hpl-2.1 on the nodes but when I link with your wrapper I am getting following problems. I am trying to figure out why this is happening. Other OpenCL programs appear to work fine on the nodes. I wonder if it has anything to do with the input file...any ideas?

               

              $ ./xhpl
              Result amax: 2058
              Result amax: 345
              Result amax: 2248
              Result amax: 2247
              clblasSgemm() failed with -1009
              clblasStrsm() failed with -1011
              clblasSgemm() failed with -1010
              Result amax: 1483
              Result amax: 1752
              Result amax: 1707
              Result amax: 670
              clblasSgemm() failed with -1009
              clblasStrsm() failed with -1011
              clblasSgemm() failed with -1010
              Result amax: 1707
              Result amax: 415
              Result amax: 1710
              ...

                • Re: OpenCL Linpack
                  dnieto

                  Hello,

                   

                  Did you solve those issues?

                   

                  I'm having the same problems

                    • Re: OpenCL Linpack
                      yurtesen

                      I think I resolved this issue but I don't remember what was the problem. I think it was related to arrays being too large to fit in the buffers. (I am not sure if this was the case) or maybe I was suppose to use float's but HPL is using doubles (perhaps this was it because no way they could get so high results with doubles in that competition). But the HPL was too slow anyway. I tried it with ACML6. AMD suggested mHPL but I didnt try it. Here is a thread you may be interested in:

                      Re: ACML 6 and A10-7850K HPL performance

                      Maybe you should try to use it with mHPL? Please update the thread if you succeed!