1. This site uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn More.

CPU Real workload stress testing

Discussion in 'Hardware' started by wyx087, 7 Jul 2020.

  1. wyx087

    wyx087 Homeworld 3 is happening!!

    Joined:
    15 Aug 2007
    Posts:
    11,992
    Likes Received:
    711
    What would be a good real world workload stress testing suite?

    I used to use the CPC benchmark on 999 loop and just leave it overnight. But now with 12T, 32GB, it's not very scalable.
    (also, the link is dead: https://bit-tech.net/blogs/download-the-custompc-media-benchmarks-here/1/)



    Reason I ask is because on my new Ryzen 3600 build, with PBO limits increased and AutoOC set to +200 MHz, the voltages look scary, almost reaching 1.5v. I've set offset -0.075v to avoid accelerated transistor aging.

    Memory is also overclocked, the cheapo modules are two single stick 3000 CL16 (£104 for 2x16GB, bargain!). I am running dual channel at 3400 CL16 no problem.

    It all seemed stable with overnight Prime, 4+ hours of AIDA CPU+FPU+cache+memory stress test, overnight Memtest86 and my normal use (web browsing, VR and normal gaming)

    BUT when my wife was editing some kid's video, it bluescreened and crashed. My guess is due to the lowered CPU voltage which caused errors in fewer core higher boost clocks. This use-case was not extensively tested. Changing offset to -0.625v did not see any further crash.


    So, TLDR, is there any mixed test suite that simulates the hardest hitting real-world use-cases? Like CPC benchmark, where it includes photo editing, video transcoding. Workload is hugely varied and causes the CPU to change its boost regularly. But automatically scales with cores and memory size to stress your CPU.
     
  2. damien c

    damien c Mad FPS Gamer

    Joined:
    31 Aug 2010
    Posts:
    3,004
    Likes Received:
    255
    Would Cinebench or Blender be worth running for a specific amount of time?

    I have left most of the bios settings to out of the box settings except enabling PBO, which to be honest doesn't seem to do anything but I have seen my 3950x use 1.43v or there abouts at 4.2Ghz.

    I did mess around and found it seemed to run all cores at 4.2Ghz at around 1.275v but I have not had time to have a proper read up on overclocking these cpu's and trying mine.

    What program was being used to edit and render the videos?

    Previously I tried Sony Vegas Pro 15 on a undervolted 2700x and it would crash but Adobe Premier Pro wouldn't crash.
     
  3. wyx087

    wyx087 Homeworld 3 is happening!!

    Joined:
    15 Aug 2007
    Posts:
    11,992
    Likes Received:
    711
    It's Adobe Premier.

    It would be good to be able to test the CPU at all levels of workload (1T, 2T, 4T; FPU heavy, logic branching heavy etc) overnight on infinite loop. I don't see any program to do that unfortunately. Have I missed anything?
     
  4. GeorgeStorm

    GeorgeStorm Aggressive PC Builder

    Joined:
    16 Dec 2008
    Posts:
    7,019
    Likes Received:
    559
    Geekbench gives a variety of cpu/mem heavy tasks, both single and multithreaded, worth a look.
     
    damien c and wyx087 like this.
  5. wyx087

    wyx087 Homeworld 3 is happening!!

    Joined:
    15 Aug 2007
    Posts:
    11,992
    Likes Received:
    711
    Thanks for suggestion. I had a look, the free tryout version of Geekbench requires clicking start every 2 min or so. In this sense, it's similar to Cinebench, where it's more single run benchmark tool, less stability test tool.

    About 7 years ago, when I overclocked i7 2700k. I didn't do straight 4.4 GHz. I wanted to do 4.2 GHz all cores and single core boost to 4.5 GHz to have a quiet PC yet maximum gaming performance. At that time I also had trouble finding stability testing tool, so I ended up with more modest boost and tested using the 2009 CPC benchmark on 999 loop.

    It's surprising today's testing tools are still just 100% load stress testing. That does not test all processor power states, thus they are unsuitable for stability testing. Considering the popularity in variable boost clocks these days, why isn't such a tool exist?
     
  6. Pete J

    Pete J Employed scum

    Joined:
    28 Sep 2009
    Posts:
    7,242
    Likes Received:
    1,800
    @wyx087 personally, I've found converting Blurays/large video files is the most intensive test.

    Find a 50GB+ video file to download (no, I'm not condoning piracy!) and use Handbrake to convert it to something else. Preferably such that it takes at least 12 hours.
     
    wyx087 likes this.
  7. Gareth Halfacree

    Gareth Halfacree WIIGII! Lover of bit-tech Administrator Super Moderator Moderator

    Joined:
    4 Dec 2007
    Posts:
    17,127
    Likes Received:
    6,714
    Or just loop Handbrake - the PC Pro benchmark suite (and my handcrafted Linux equivalent) uses the command-line version so it's scriptable, and it's the work of seconds to loop.

    Linux:

    Code:
    for i in {1..20}; do HandBrakeCLI -i input.mov -t 1 --angle 1 -c 1 -o output.mkv -fmkv  --decomb -w 1920 --loose-anamorphic  --modulus 2 -e x264 -q 20 --vfr -a 1,1 -E faac,copy:ac3 -6 dpl2,auto -R Auto,Auto -B 160,0 -D 0,0 --verbose=1; done
    
    Windows:

    Code:
    1..20 | % { HandBrakeCLI.exe -i input.mov -t 1 --angle 1 -c 1 -o output.mkv -fmkv  --decomb -w 1920 --loose-anamorphic  --modulus 2 -e x264 -q 20 --vfr -a 1,1 -E faac,copy:ac3 -6 dpl2,auto -R Auto,Auto -B 160,0 -D 0,0 --verbose=1 }
    
    (Maybe, I haven't done any PowerShell in anger in a long time.)

    That'll do 20 iterations, which should be enough to find any flaws.
     
    wyx087 and Pete J like this.

Share This Page