cpus (real > simulated)

Discussion in 'Hardware' started by bestminipc, Jan 18, 2019.

  1. bestminipc

    bestminipc Efficiency Noob
    Thread Starter

    #1 bestminipc, Jan 18, 2019
    Last edited: Jan 18, 2019
    cpus (real > simulated) is sectioned out of the main post to prevent clutter from the main post
    https://smallformfactor.net/forum/t...good-at-building-small-pcs.10061/#post-132142


    real > simulated

    simulated tests doesnt seem very helpful relative to actually testing real software

    simulated tests so inaccurate to real software that uses 20-80% of cpu
    https://www.quora.com/Why-isnt-the-...-core-score-multiplied-by-the-amount-of-cores

    simulated tests are aggregated so not useful to specific uses
    https://www.quora.com/Can-we-compare-desktop-and-mobile-performance-based-on-Geekbench-scores

    even in the specific case of only photoshop: even the ppl tha know ps far more than other ppl do not know what is best cpu for photoshop

    "Multiple cores make a difference for things like preview building, exporting, merges, etc. But many things have other constraints that prevent Lightroom from maxing out the CPU."

    https://feedback.photoshop.com/phot...best-suit-editing-mac-mini-vs-imac-vs-waiting

    but most ppl use more than just and only 1 software

    prob hard to say what cpu is good for a multitude of future unknown softwares and versions of current software
    there's no software taht's equally best for a multitude of things
    there's no chopsticks or other utensils that are equally best for a multitude of things
    i dont know/think there's any computers that's equally best for a multitude of things
    and i dunno anything about cpus

    but i'd guess there's no cpus that's good for a multitude of software that most ppl wont be using anyway, and thus any claimed value is completely lost

    as that value was never used consistently or benefically or effectively or most of the time

    actualy test things, not simulations
    https://www.quora.com/Which-is-bett...-playing-games-or-using-benchmarking-software

    all these sites is garbage, inaccurate, unhelpful
    https://www.quora.com/In-your-opini...rking-review-site-for-video-cards-and-or-CPUs

    billions of things, ill maybe have to organise them all one day, but this is trivial topic that no time should be wasted on


    but if we have no good simulated tests (which there are none?), and if we have no good real tests, then how are we able to decide well on cpus? or any other performative parts?

    dunno, i've no idea

    if there's any supa helpful info to this problem feel free to link them here
     
  2. Choidebu

    Choidebu King of Cable Management

    This is utter nuts.

    well glad we got that sorted out. Do you want to know? I'm pretty sure there are books about it. Maybe start at 8086?

    How old are you if I may ask? Background? Studying anything? Building something?

    I mean, in this modern world, you can totally have no idea how something works but still good at using it. That's called standing at the shoulders of giants, as we all are.
     
    kotproger likes this.
  3. Choidebu

    Choidebu King of Cable Management

    Dude. Actual people writes, votes for, comments on those answers. Sorry if their time worth f*k all to you.

    *sigh* certainly not trivial enough that you don't need to google it.

    Tests/benchmarks are not perfect, yes, but that gives us a starting point to compare parts. You can't even correlate a benchmark number result with how good your experience in using , say, photoshop.

    Look, I'll give you a general idea here. And maybe show you a bit of the complexity involved.

    Software devs make softwares. They make it work, then they (can) make it work better/faster. How? They optimize with current gen instruction sets that comes with newer cpus.

    Cpu engineers make cpus. They make it more efficient at executing codes. How? They increase clocks. They create instruction sets, hardware implementation blocks for popular algorithms. They streamline pipelines, improve branch predictions, reduce cache faults and misses. They add more cores.

    Now you see this is a chicken and egg problem. Devs create algorithms, it gets popular, then maybe it gets enough traction cpu engineers put hardware implementation for it. But then to take advantage of it the devs then need to change their software. Which means consumers also need to change their hardware and their software to take advantage of it.

    And the benchmark now needs to decide if they'll give a better score for prime generation block set (tm) for a new cpu....

    Now, don't get me started on cores, and the complexity of multi threaded programming.

    Tldr; the cpu is a beast. Software is complex.
     
    kotproger likes this.