Everyone, We've got some big shiny new SMP machines running Cent OS. I'm trying to convince our people that they are slow; about 2 fold slower than my 12 month old MacPro (running either Slackware or OS X). So... to show relative/absolute differences in execution time are there any well-known benchmarking problems that I should be using? I'd like to cover a lot of different bases (e.g. linear algebraic problems, memory intensive, CPU intensive, etc). We'd have to make the usual caveats about diferences in OS's but it would be good to have numbers. If people have good ideas, I wouldn't mind putting a small R package on R-Forge to make the tests easily accessible. Thanks, Max
benchmark problems
4 messages · Steve Lianoglou, Max Kuhn, Dirk Eddelbuettel
Hi,
On Mon, Jun 7, 2010 at 10:35 AM, Max Kuhn <mxkuhn at gmail.com> wrote:
Everyone, We've got some big shiny new SMP machines running Cent OS. I'm trying to convince our people that they are slow; about 2 fold slower than my 12 month old MacPro (running either Slackware or OS X). So... to show relative/absolute differences in execution time are there any well-known benchmarking problems that I should be using? I'd like to cover a lot of different bases (e.g. linear algebraic problems, memory intensive, CPU intensive, etc). We'd have to make the usual caveats about diferences in OS's but it would be good to have numbers. If people have good ideas, I wouldn't mind putting a small R package on R-Forge to make the tests easily accessible.
Does the LINPACK benchmark help here? http://www.top500.org/project/linpack http://www.netlib.org/benchmark/hpl/ Here's a MATLAB program that runs the benchmark: http://people.sc.fsu.edu/~burkardt/m_src/linpack_bench/linpack_bench.html Sorry if that's too obvious of a suggestion, but it's a start. -steve
Steve Lianoglou Graduate Student: Computational Systems Biology | Memorial Sloan-Kettering Cancer Center | Weill Medical College of Cornell University Contact Info: http://cbio.mskcc.org/~lianos/contact
Yes, those are being run. I wanted something from within R to specifically evaluate that build. For example, R built with icc instead of gcc. Also, I want to test a variety of computational problems. Thanks, Max On Mon, Jun 7, 2010 at 7:02 PM, Steve Lianoglou
<mailinglist.honeypot at gmail.com> wrote:
Hi, On Mon, Jun 7, 2010 at 10:35 AM, Max Kuhn <mxkuhn at gmail.com> wrote:
Everyone, We've got some big shiny new SMP machines running Cent OS. I'm trying to convince our people that they are slow; about 2 fold slower than my 12 month old MacPro (running either Slackware or OS X). So... to show relative/absolute differences in execution time are there any well-known benchmarking problems that I should be using? I'd like to cover a lot of different bases (e.g. linear algebraic problems, memory intensive, CPU intensive, etc). We'd have to make the usual caveats about diferences in OS's but it would be good to have numbers. If people have good ideas, I wouldn't mind putting a small R package on R-Forge to make the tests easily accessible.
Does the LINPACK benchmark help here? http://www.top500.org/project/linpack http://www.netlib.org/benchmark/hpl/ Here's a MATLAB program that runs the benchmark: http://people.sc.fsu.edu/~burkardt/m_src/linpack_bench/linpack_bench.html Sorry if that's too obvious of a suggestion, but it's a start. -steve -- Steve Lianoglou Graduate Student: Computational Systems Biology ?| Memorial Sloan-Kettering Cancer Center ?| Weill Medical College of Cornell University Contact Info: http://cbio.mskcc.org/~lianos/contact
Max
Hi Max,
On 7 June 2010 at 20:53, Max Kuhn wrote:
| Yes, those are being run. I wanted something from within R to
| specifically evaluate that build. For example, R built with icc
| instead of gcc.
That is a nice idea! Something that has been suggested in the past was to
simply run the whole regression test which resembles 'normal' load as it
covers a variety of tasks. So you could
cd gcc-build && /usr/bin/time make check-all
cd icc-build && /usr/bin/time make check-all
| Also, I want to test a variety of computational problems.
Simon has a page here
http://r.research.att.com/benchmarks/
Also, we just started an R-Forge project aiming to collect gpu/cpu benchmarks
and cutoff points (ie to answer "who wins when"). That will require some
basic infrastructure building as well. Do you want to join in the effort?
We're at
https://r-forge.r-project.org/projects/gcb
but as I said there isn't much there yet...
Regards, Dirk