[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: testing hardware - use what software ?



max bianco wrote:
On Tue, Jun 3, 2008 at 3:41 PM, Roger Heflin <rogerheflin gmail com> wrote:
Robin Laing wrote:

Compiling up something called HPL (with something called MPI) at least does
nicely at finding that you have a memory/overheat/internal CPU issue.  If
the results corrupt or the machine crashes something is really wrong,
typically it won't tell you what is wrong, but if it successfully runs for a
long time then you can expect most things to be correct.    Generally it
will at least crash the machine several times faster than most other
applications.

It won't find IO/PCI/Video issues unless they are really severe, though
generally most of the issues fall into what it does test.


Do you happen to know what the latest version is? I have turned up a
version 1.0a dated Jan 20, 2004. Do you know if that is the latest
version available.

Max


It does not change much, so that is the latest version.

You will also need a few other pieces of software, none of which will likely be in a rpm to get it to fully compile and run. You will need mpich and either libgoto or atlas or AMD MKL or Intel MKL. I have built it all 4 ways, the AMD and Intel libraries are probably the best choice, though the makefile will need to be adjusted to point the the library. Both of those MKL libraries were the last time I checked available for free download.

And then you will have to define a HPL.dat file for your machine, once you know how to do it, really the only a couple of things in the file change when you move things to a new machine (size of the run), and it hpl can be ran across a network on multiple machines.

                           Roger


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]