Industry-Standard CPU/Core Benchmarking Task?

Discussion in 'Programming' started by kmiklas, Feb 28, 2018.

  1. kmiklas


    Hello All,

    What is considered to be the industry-standard task for benchmarking a CPU or core?

    I'm thinking things like sorting a list of 1000000 elements, or something of that nature.

    One thread per CPU; no hyperthreading.

    Thanks for your help.

    Keith :^)
  2. comagnum


    Download the light utility, run the test and see how your puter stacks up against the 7+ million already tested. Shows you strong & weak areas.
    kmiklas likes this.
  3. jharmon


    That sort of thing (checking the speed of a very limited operation) worked well... in the 80s!

    We now use multithreading, GPU offload and real workload simulations.

    What type of workload you actually trying to benchmark?
    kmiklas likes this.
  4. comagnum


    The 80's - dream on - nothing remotely close to a test like this existed back than in the DOS/Floppy/Modem era. Besides it tests up to 32 cores and the GPU. For a quick free PC test it's not bad. Probably did not even look at it.

    Sure you can buy more elaborate benchmark software if you run a server farm or just want to geek out.
    Last edited: Feb 28, 2018
    kmiklas likes this.
  5. jharmon


    Cogmagnum, there were plenty of such tests in the 1980s on DOS.

    There's the old trusty Whetstone and Dhrystone benchmarks (they actually pre-date DOS). I also remember the PC Magazine Labs Benchmark Series test that performed more "real world" tests. There were dozens more too.

    OP seems to be trying to reinvent the wheel.
    kmiklas likes this.
  6. kmiklas


    Perf On a 4-core RH Linux VM for the purpose of tick data monitoring, algo analysis, automated order submission, risk and compliance measures. No GPU available. Kernel settings/bypass accessible to some degree.

    Trying to understand what kind of torque I can get out of this rig, and how to best tune these 4 cores to do so.
    Last edited: Mar 1, 2018
  7. kmiklas likes this.
  8. jharmon


    Tune your routines - forget about the hardware. Do some theory of computing courses to learn about Big O Notation. Figure out whether your methods are O(1), O(n^n) or something in between, and optimize them. Consider whether multithreading has a use.
  9. kmiklas


    I'm doing ULL kernel tuning, and need to analyze and measure raw cpu and core horsepower; asymptotic analysis is important, but off-topic.

    Also, multithreading is intimately tied to hardware; such as how threads are assigned to cores, interrupts, hyperthreading, etc.
    Last edited: Mar 8, 2018
  10. kmiklas


    #10     Mar 8, 2018