Announcement

Collapse
No announcement yet.

Floating Point Math test not same results as GLOPs ?

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • Floating Point Math test not same results as GLOPs ?

    Hello.
    I has running PerformanceTest Linux.
    In test Floating Point Math the result is 7575 Million Operations/s
    CPU is AMD FX-6300.
    Have anothers benchmarks showing result 48 GFLOPs.

    PerformanceTest Linux floating point score are changed to operations.
    How convert the results for GFLOPs ?
    Any chance in next version add an option to benchmark using GFLOPs ?
    I hard to understand floating operation/s.

    Have an nice day.

  • #2
    GFLOPS = Billions of floating point instructions per second.
    MFLOPS = Millions of floating point instructions per second.

    To convert one to the other divide by 1000.

    But both MFLOPS and GFLOPS are very poorly defined.

    A few of examples.
    1) 32bit float + 32bit float is a floating point operation.
    2) 128bit float / 128bit float is another floating point operation. But this one might be 8 times slower to execute.
    3) If you work on a small data set (e.g. 8KB) it might fit entirely in the CPUs cache. If you work on a large dataset (e.g. 16GB) it might be so large that it doesn't even fit in main RAM and must swap to disk. The speed difference might be 10,000x faster in the 1st case compared to the second.

    Some benchmarks like to show the biggest numbers possible. We've tried to use a more realistic mix of instructions and data sizes.




    Comment


    • #3
      Thanks David for your another reply.

      CPU tested is FX-6300 and Passmark floating point test was 7575 Million.
      Strange detail is that value 7575 Million is almost the same value for GFLOPs in each CPU core.
      Example = others benchmarks using 6 threads > 8 GFLOPS in each core = 8x6=48 > total 48 GFLOPs

      Performance Test is using only one core for testing result ? If yes I only need > test result value X 6 = total performance ?

      Comment


      • #4
        By default all cores are used for the floating point test in PerformanceTest (you can change this from the preferences window however).

        You can find a more detailed discussion of the instruction mix and operations per second here.

        Comment


        • #5
          THANKS !
          The link has very good information !
          Have an nice week.

          Comment

          Working...
          X