Re: IDL Matrix Multiply and Dual-Core CPUs

oops, I have written IDL 7 time to the ATLAS test. Corrected below.

lajos



I have run some tests on a quad-core Intel Core2 Q6600 / linux 64 bit 
machine.

On Fri, 9 May 2008, s.haenger@[EMAIL PROTECTED]
 wrote:

  Hi,

  I have a Problem with IDL 7.0
  We have to multiply large matrices. With some matrix sizes, the CPU
  usage is 100% but for most of the matrices it is 50%. (I'm runnning it
  on a Intel T7250 (Dual Core, 2GHz, 2MB L2 Cache))


  The CPU System Variable is configured like this:
  IDL> print, !CPU
  {           0           0           2           2
  100000           0}


  Now we do this:
  matA = randomn(42, 2000, 2200)
  matB = randomn(43, 2020, 2000)
  matIdl = matA##matB

  So now i've got a CPU usage of 100%


   # of threads    IDL 7 time
         1          12.476210
         2           6.5931890
         3           5.2085290
         4           4.9191489

it scales well for two cores, so the CPU usage should be near 100% for two

threads.

  but with this:
  matA = randomn(42, 2500, 2500)
  matB = randomn(43, 2520, 2500)
  matIdl = matA##matB

  the cpu usage is around 50%-60%

   # of threads    IDL 7 time
         1         22.034877
         2         11.681226
         3          9.7771089
         4          9.3093379

again, CPU usage should be near 100% for two cores.


Just for comparison, ATLAS (http://math-atlas.sf.net)
times:

   # of threads    ATLAS time
         1          4.4285851
         4          1.1784132

and

   # of threads    ATLAS time
         1          7.8148808
         4          2.1345751

regards,
lajos
>