Announcement

Collapse
No announcement yet.

Comparing Ratings

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

  • Comparing Ratings

    For my bachelorthesis i need the difference in performance of two processors. i found them on this site, and their ratings are 2253 and 5074. does that mean that when running an algorithm, the second one iwll do it approximately twice as fast? or doesnt it say anything? I know it will be different for every algorithm, but i only need an approximate.
    Thanks in advance

  • #2
    Yes it will be different for different algorithms.
    It will be especially dependent on
    - if the code is threaded
    - if it makes use of SIMD instructions that are only available on one of the CPUs.
    - if the code is CPU bound or makes use of a lot of RAM or the disk.

    But very broadly speaking the 2nd one is more than twice as fast as the 1st.

    Comment


    • #3
      and what if;
      i made an algorithm in c++ not threaded that takes allot, say +2GB of RAM and ran it on intel duo core; http://www.cpubenchmark.net/cpu.php?...00+%40+3.00GHz

      and someone else made the same algorithm in java 1.5, and DID thread it and ran it on the faster http://www.cpubenchmark.net/cpu.php?...30+%40+2.80GHz with 4 cpu's.
      Would my gestimate that that computer would run that algorithm 8 to 10 times faster be defendable?

      Thanks for the quick reply

      Comment


      • #4
        Depends on the JVM, how good the two coders were, how dependent the algorithm was on RAM access speeds, the compiler in use, if you are doing managed C++ in .NET or unmanaged, how much RAM was in the machine, it is was 32bit or 64bit, if the algorithm scales in a linear fashion across multiple cores, etc...

        For example if the algorithm could be coded to make use of SIMD instructions then I am sure I could write C++ code that was 2 or 3 times faster than Java. Negating the advantage of the faster CPU. You also get much finer control of memory management in C++. So a good coder can exploit this when using large data sets to get better performance.

        Further in C/C++ you can write code to run on the GPU as well as the CPU. So you might get 10x the performance from C++ if you put the time in and know what you are doing. But in this case you would need to thread the code.

        So I think your gestimate makes too many assumptions, and isn't really defensible.

        Comment


        • #5
          Thank you very very very much

          Comment

          Working...
          X