HPL
From FarmShare
(Difference between revisions)
| (4 intermediate revisions not shown) | |||
| Line 2: | Line 2: | ||
Trying to run HPL on markland cluster. | Trying to run HPL on markland cluster. | ||
| - | CPUs are E5-2670 which are supposedly ~155GFLOPS per socket. [1] Another source says 330 GFLOPS. [2] Let's be conservative and say 150GF per CPU. Trying one run of xhpl with gotoblas, I get about | + | CPUs are E5-2670 which are supposedly ~155GFLOPS per socket. [1] Another source says 330 GFLOPS for a 2S server. [2] Let's be conservative and say 150GF per CPU. Trying one run of xhpl with gotoblas, I get about 26 GFLOPS for one socket, or about one sixth of the max theoretical. Clearly not the best BLAS for this CPU. |
| - | + | ||
==Links== | ==Links== | ||
| - | * run with [http://www.open-mpi.org/doc/v1.4/man1/mpirun.1.php | + | * run with [http://www.open-mpi.org/doc/v1.4/man1/mpirun.1.php mpirun]: http://www.rocksclusters.org/rocks-documentation/4.3/running-linpack.html |
* generate your HPL.dat: http://www.advancedclustering.com/faq/how-do-i-tune-my-hpldat-file.html | * generate your HPL.dat: http://www.advancedclustering.com/faq/how-do-i-tune-my-hpldat-file.html | ||
* get the right BLAS for Sandy Bridge: http://software.intel.com/en-us/articles/performance-tools-for-software-developers-hpl-application-note/ | * get the right BLAS for Sandy Bridge: http://software.intel.com/en-us/articles/performance-tools-for-software-developers-hpl-application-note/ | ||
| + | * AMD page: http://developer.amd.com/documentation/articles/pages/HPCHighPerformanceLinpack.aspx | ||
| + | * Intel pre-compiled binaries: http://software.intel.com/en-us/articles/intel-math-kernel-library-linpack-download/ | ||
==references== | ==references== | ||
#http://forums.anandtech.com/showpost.php?p=33585052&postcount=1 | #http://forums.anandtech.com/showpost.php?p=33585052&postcount=1 | ||
#http://novatte.com/blog/2012/03/how-to-calculate-theoretical-peak-performance-of-a-cpu-based-hpc-system/ | #http://novatte.com/blog/2012/03/how-to-calculate-theoretical-peak-performance-of-a-cpu-based-hpc-system/ | ||
Latest revision as of 13:52, 3 August 2012
markland
Trying to run HPL on markland cluster.
CPUs are E5-2670 which are supposedly ~155GFLOPS per socket. [1] Another source says 330 GFLOPS for a 2S server. [2] Let's be conservative and say 150GF per CPU. Trying one run of xhpl with gotoblas, I get about 26 GFLOPS for one socket, or about one sixth of the max theoretical. Clearly not the best BLAS for this CPU.
Links
- run with mpirun: http://www.rocksclusters.org/rocks-documentation/4.3/running-linpack.html
- generate your HPL.dat: http://www.advancedclustering.com/faq/how-do-i-tune-my-hpldat-file.html
- get the right BLAS for Sandy Bridge: http://software.intel.com/en-us/articles/performance-tools-for-software-developers-hpl-application-note/
- AMD page: http://developer.amd.com/documentation/articles/pages/HPCHighPerformanceLinpack.aspx
- Intel pre-compiled binaries: http://software.intel.com/en-us/articles/intel-math-kernel-library-linpack-download/
references
- http://forums.anandtech.com/showpost.php?p=33585052&postcount=1
- http://novatte.com/blog/2012/03/how-to-calculate-theoretical-peak-performance-of-a-cpu-based-hpc-system/
