Has anyone tested how a dual CPU board with dual core CPUs scales with SLI?

Discussion in 'Architecture and Products' started by g__day, Mar 12, 2006.

  1. g__day

    Regular

    Joined:
    Jun 22, 2002
    Messages:
    580
    Likes Received:
    2
    Location:
    Sydney Australia
    It might be pricy, but for benchmarking to see just how much CPUs might be holding back the new high end cards in SLI or Crossfire, has anyone yet tested a dual CPU board (e.g. TYAN's high end) using say two dual core AthlonFX 60s on top end dual gpu gear with a benchmark that caters for it?

    I'd like to understand beter how much more headroom dual GPUs have if the CPU were to scale much higher through clockspeed or increased parallelism.
     
  2. Dave Baumann

    Dave Baumann Gamerscore Wh...
    Moderator Legend

    Joined:
    Jan 29, 2002
    Messages:
    14,081
    Likes Received:
    651
    Location:
    O Canada!
    You'll need the applications themselves to make significant usage of increased CPU parallelism before you'll notice much difference from dual core over single, as this isn't "automatic" as it is with graphics. Even then, applications are not necessarily going to scale linearly with numbers of cores (i.e. Tim Sweeney suggested that the likelyhood for Unreal would be that there is a "core" engine thread, a graphics thread, possibly a physics thread (IIRC) and then several "helper" threads).

    In many cases the gains for multi graphics on Dual core stya very similar to single graphics with the drivers drivers shuttling things like the general driver overhead, shader compiliation/optimisation and possibly some vertex shaders on to the second core with single threaded apps.
     
  3. obobski

    Newcomer

    Joined:
    Jul 25, 2005
    Messages:
    117
    Likes Received:
    3
    With more processors all your doing is increasing the total FLOPS of the system, your doing more work in the same time, however most software of today is designed to scale by linear clock speed: a 9GHZ Pentium 4 would be faster than 2 5GHZ Pentium 4's (this is purely theoretical)

    Hopefully with the advent of dual cored processors this changes, and we get software that can utilize higher GFLOP output of the system, without needing it moving faster (in other words, most software appears to want the system to look at things in a scalar and serial way, and in order to get more performance we need to get software that wants parallel (and possibly vector) perceptions, then dual cores and SMP's can really kick ***)
     
  4. g__day

    Regular

    Joined:
    Jun 22, 2002
    Messages:
    580
    Likes Received:
    2
    Location:
    Sydney Australia
    Isn't that just a design question, hand work out using gaming middleware with put/get queues? That way you design becomes inherently parallel friendly and any idle unit simply line up getting work from a queue, doing it then getting more work.

    These principles have been around since the eighties!
     
  5. g__day

    Regular

    Joined:
    Jun 22, 2002
    Messages:
    580
    Likes Received:
    2
    Location:
    Sydney Australia
    Interesting article here, titled Dual CPU or Dual Core?

    http://www.pugetsystems.com/articles.php?id=23

    Conclusion

    As you can see, the move to dual-core is definitely a win for consumers. Since they are more affordable than dual processor systems, but offer the same or better performance, they are becoming the standard for modern computer systems.
     
  6. g__day

    Regular

    Joined:
    Jun 22, 2002
    Messages:
    580
    Likes Received:
    2
    Location:
    Sydney Australia
Loading...

Share This Page

  • About Us

    Beyond3D has been around for over a decade and prides itself on being the best place on the web for in-depth, technically-driven discussion and analysis of 3D graphics hardware. If you love pixels and transistors, you've come to the right place!

    Beyond3D is proudly published by GPU Tools Ltd.
Loading...