FLOP performence of the Mac Pro 8-Core?

Discussion in 'Mac Pro' started by mattthemutt, Apr 5, 2007.

  1. mattthemutt macrumors regular

    mattthemutt

    Joined:
    Jun 13, 2004
    Location:
    Ontario, Canada
    #1
    With the release of the new 8-Core mac pro, the question arises about how many gigaflops this computer is capable of.

    Is there any way of telling?
     
  2. gnasher729 macrumors P6

    gnasher729

    Joined:
    Nov 25, 2005
    #2
    I'd estimate 1 double precision floating-point operation per cycle per core sustained with hand-optimised, vectorized 64 bit code that is really well suited to that kind of thing; that would be 24 GFlops per computer. The theoretical limit is probably four times that, but really really hard to achieve. Single precision should be twice that.
     
  3. mattthemutt thread starter macrumors regular

    mattthemutt

    Joined:
    Jun 13, 2004
    Location:
    Ontario, Canada
    #3
    I pulled this from Wikipedia for comparison purposes. I realise that this may just be marketing, but realistically speaking, does this mean that in terms of FLOPs, the Mac Pro is less powerful?
     
  4. finiteyoda macrumors member

    Joined:
    Apr 10, 2007
    #4
    I think the gigaflop performance of an 8-core should be around 54 gflops, actually. I saw a 4-node dual-proc Xeon 5160 post 136 gflops on linpack, that's 17 gflops per dual-core Xeon 5160. Intel is reporting a 1.6X increase on Linpack with clovertown, that would be 27 gflops. With two clovertowns, that's roughly 54 gflops for an 8-core.

    Comparing PS3 to a mac/pc is apples and oranges. A PS3 is about 15 gflops (double precision) from the Cell CPU, and roughly 2 teraflops from the GPU. A mac pro is maybe 50 gigaflops from the CPU's, and maybe 500 gigaflops from the ATI card (for nvidia cards I don't know).

    Also, the PS3 has a slow hard drive and only 256MB ram. Programming its 6-core powerpc-based CPU effectively is much harder than the quad-core xeons. Writing code to use GPU hardware is even a magnitude harder. If your app of choice can be written to take advantage of stream processing on a GPU and is fine with 60GB disk space and 256MB RAM (folding@home is a good example) then the PS3 is far more cost-effective. But for most people, the mac pro (or any x86-based workstation) will be far more useful.
     
  5. gnasher729 macrumors P6

    gnasher729

    Joined:
    Nov 25, 2005
    #5
    Realistically speaking, this is just marketing. The Cell processor is a pure vector processor. Think of it as an Altivec processor, which is a pain in the ass to write code for, without the general purpose processor, which makes it a much bigger pain, restricted to a rather small amount of memory, which makes it an even bigger pain, and then you have seven of them, so you need to parallelize your code by hand, which squares the pain.

    Also, the numbers given are for single precision floating-point. That is about six or seven decimal digits of precision, nowhere near enough for many purposes. Double precision floating point is much much slower.
     
  6. Chaszmyr macrumors 601

    Chaszmyr

    Joined:
    Aug 9, 2002
    #6
    As far as the PS3's Cell architecture is concerned, everyone knows that peak potential is through the roof, but that definitely doesn't mean you'd want to replace your x86 Xeon architecture with a Cell architecture for personal computing.
     

Share This Page