Anybody out there aware of Linpack performance test for Windows-7 64-bit to give performance results for multi-GPU configurations. Nice if the software worked within a single node and or cluster.
Seems they are geared to Linux OS’s.
Sandra by SiSoftware and a few other packages have GPU test, but wanted to compare to the TOP500 list.
Also wanted to see how Linpack compares to other performance test software (do the TFLOPS agree with each other?)
I am not aware of any multi-GPU Linpack implementations for Windows. The codes I am familiar with (disclosure, I wrote one of them) use POSIX threads for implementing multi-GPU BLAS calls. That is a pretty major roadblock to porting to a Windows platform.
Yes, boost threads gets rid of the portability problems (and the threading requirements to implement a multi-gpu version of HPL are not very complicated). On the other hand, boost introduces C++ dependencies to HPL, which greatly complicates the build process and makes the port/patch a lot more invasive. Using POSIX threads and CUBLAS lets you get away with a pretty straightforward patch which basically doesn’t disturb either MPI or the rest of the HPL codebase or build at all.