Linpack Benchmarking on PVM with CUDA?

I'm working on a project comparing the performance of PVM and MPI in GPU accelerated clusters. Naturally, I turned to the defacto standard for computer cluster benchmarking, LINPACK. The problem I'm running into is finding a version of LINPACK for PVM. It does exist somewhere because there are performance comparisons  of systems using LINPACK and PVM. Ive already installed SCALAPACK and all of the other libraries required for it to work, but I can't figure out how to use it as a benchmark. Additionally, I need the benchmark to be GPU accelerated. Ive looked into HPL, which is GPU accelerated through CUDA, but there are no versions for PVM. Does anyone know of a version of LINPACK that can be run on PVM and could be modified to be CUDA accelerated?

Asked by danielemur 5 years ago


How to use Scalapack as a benchmark?

For a science fair project I am comparing the performance of MPI and PVM on GPU clusters. The biggest problem I have is finding benchmarking software for PVM. Linpack is the most used benchmarking software, and from there I have gone on to find Scalapack, a set of libraries which has the same type of functions as Linpack, but supports both PVM and MPI. From my understanding, Linpack, Specifically HPL , Is a set of functions along with a program which utilizes them for benchmarking. Scalapack, on the other hand, appears to be only the funtions, with no program. Does anyone know how to use Scalapack as a cluster benchmark that measures performance, in flops if possible?

Asked by danielemur 5 years ago