View source for Using the Intel LINPACK Benchmark to stress the hardware

''Via [http://web.archive.org/web/20120201010316/http://kb.fusionio.com/KB/a62/using-linpack-benchmark-to-stress-test-your-system.aspx http://kb.fusionio.com/KB/a62/using-linpack-benchmark-to-stress-test-your-system.aspx]''

* Download [http://software.intel.com/en-us/articles/intel-math-kernel-library-linpack-download/ Intel LINPACK Benchmark]

* Reboot to the terminal (don't start X), unload unneeded drivers, stop unneeded applications, turn off swap

* Go into the <code>.../linux/mkl/benchmarks/linpack</code> directory

* Compute the job size using a python script:
<pre>
#!/usr/bin/python
import commands
import math
print(math.floor(math.sqrt((float(commands.getoutput("free -b|grep Mem|awk '{print $2}'"))-1073741824)/8)/16)*16+8)
</pre>

* Creat a new <code>lininput_xeon64</code> input file using the number <code>XXXXX</code> calculated above:
<pre>
Sample Intel(R) Optimized LINPACK Benchmark data file (lininput_xeon64)
Intel(R) Optimized LINPACK Benchmark data
1 # number of tests
XXXXX # problem sizes
XXXXX # leading dimensions
1 # times to run a test
4 # alignment values (in KB)
</pre>

* Run <code>time ./runme_xeon64</code>

* Adjust problem size and leading dimensions if needed

* Compute the number of passes needed for the test (not less than 10 passes)

* Run <code>./runme_xeon64</code> and then check <code>lin_xeon64.txt</code>

[[Category: Hardware]]