Jump to content
HWBOT Community Forums

Dead Things

Members
 View Profile  See their activity

  • Posts

    241

  • Joined

  • Last visited

 Content Type 

Posts posted by Dead Things

    Bug reports : Multiple instances of Xeon X5460

    in Support

    Posted · Edited by Dead Things

    Ticket ID: 1547

     

    Priority: Low

     

    The Xeon X5460 appears to be listed on the site twice, once as a 3160 MHz part, and once as a 3166 MHz part (the latter of which is correct). Multiple instances can be seen in this list: http://hwbot.org/hardware/processors#key=s771

     

    Processor info: http://www.cpu-world.com/CPUs/Xeon/Intel-Xeon%20X5460%20-%20EU80574KJ087N%20-%20AT80574KJ087N%20%28BX80574X5460A%29.html

    Dead Things - 4x Xeon E7 8860 @ 2267MHz - 5131.5 MPT-score UCBench 2011

    in Result Discussions

    Posted

    Good question. Long answer. Apologies in advance.

     

    Windows uses something called "processor groups" to accommodate systems with more than 64 logical processors. With 80 logical processors, the system would be divided into two processor groups comprised of 40 logical processors each. You can manually edit the BCD to change the processor group membership characteristics, but one thing that cannot be changed is the 64-processor ceiling per group.

     

    While this processors group implementation works for parallelized tasks, it does not work for symmetric tasks meaning that you're essentially limited to 64 cores in Windows for any one particular process.

     

    Because UCBench is actually really well-coded, each thread does not leave much processing headroom for hyperthreading, so it benefits way less from HT than, say, wPrime. As such, my score with 32 cores enabled and HT on was way lower than my score with all 40 enabled and HT off.

     

    I had some hope that Server 8 would do away with the clumsy processor group implementation, but that was not the case last time I checked. By comparison, Ubuntu has supported >64 threads per symmetric task since 10.04.

    R4.2.4/R4.3: HWBoints finetuning

    in HWBOT Rev.4

    Posted

    I think the case of my E7-8860 rig, though, proves that there should be variable awards for top spots, as suggested. Having a rare machine like this and benching it should be worth something, but it should not be what is essentially a bye to first place in the EL.

    UC Bench 2011 Processing Thread Limit

    in Benchmark software

    Posted

    Figured it out! You have to use the -noaffinity flag (or untick the "Use affinity locking" checkbox in the advanced options). That's the only way unrar.exe will scale past 32 threads.

     

    So my command line is: "unrar bench -cf=ssse3 -cpus=X -noaffinity test.rar" where X is a number larger than 32.

     

    Thanks for all the input guys!

    UC Bench 2011 Processing Thread Limit

    in Benchmark software

    Posted · Edited by Dead Things

    I'm playing with a new multi-socket Westmere rig and trying to run UC Bench with two CPU's installed but running into a bit of an issue... It appears the bench limits itself to using 32 processing threads even though there should be 40 available.

     

    Can anybody confirm or deny this limit? Is 32 cores a hard ceiling for UC Bench or is there a way to manually force the number of max threads the bench can use? I know that you can force the instruction set used with the -cf flag, but I don't see any documentation regarding #threads.

     

    Any input would be appreciated. Cheers!

×
×
  • Create New...