Does hardware polling software like RTSS & HWiNFO64 have a discernible impact on performance?

Discussion in 'Rivatuner Statistics Server (RTSS) Forum' started by BlindBison, Jul 1, 2022.

  1. BlindBison

    BlindBison Maha Guru

    Messages:
    1,451
    Likes Received:
    534
    GPU:
    RTX 2080 Super

    They mention in this Hardware Unboxed video that polling software like HWiNFO64 can cause a difference in performance, but is that really true?

    It was my understanding that polling applications like RTSS or HWiNFO64 were almost so lightweight as to be basically negligible.

    Thanks!
     
  2. mbk1969

    mbk1969 Ancient Guru

    Messages:
    12,807
    Likes Received:
    11,008
    GPU:
    GF RTX 3060TI
    Lightweight load applied to 100% loaded CPU can be noticeable.
     
    BlindBison likes this.
  3. BlindBison

    BlindBison Maha Guru

    Messages:
    1,451
    Likes Received:
    534
    GPU:
    RTX 2080 Super
    Thanks, that's helpful to know! In that case what are your thoughts on stuff like mouse polling rate? I personally cannot tell the difference even at high framerates between 500 and 1000 hz. There were some videos from awhile back on YT demonstrating a measurable difference of a couple percentage points on the CPU usage with different mouse polling rates, but that was a bulldozer era CPU iirc.

    Sadly, I don't think I can give up using HWinfo and RTSS, the core data is just so darn useful. Still, it's a shame if they do cause a measurable or discernible decrease in overall system performance during gaming.
     
  4. aufkrawall2

    aufkrawall2 Ancient Guru

    Messages:
    2,097
    Likes Received:
    636
    GPU:
    3060 TUF
    I'd poll only the sensors you are interested in, not the same ones in multiple tools at the same time and not more frequently than each 500ms. Then those tools should cause ~0% CPU load and not have any measurable impact.
     
    BlindBison and Unwinder like this.

  5. Unwinder

    Unwinder Moderator Staff Member

    Messages:
    16,066
    Likes Received:
    3,659
    They don't, if you understand the principles of hardware polling, monitor you system wisely and do not try to do absolutely stupid things like enabling every sensor you can enable and polling them 100 times per second. That's a question of nature of sensors you poll, total sensors count and polling rate you define.
    Each sensor requires some CPU time to be read, some are extremely fast and eat just a few microseconds of CPU time per polling iteration (mostly on-die ones like CPU temperatures or GPU temperatures), some are slower and require accessing external chip via slower I2C protocol (e.g. power sensor on GPUs), such sensors may require a few milliseconds of CPU time per polling iteration. Some sensors are not intended (and not tuned by hardware vendor) for realtime high perforomance monitoring at all (e.g. SMART temperature attribute or any other SMART attribute in general), such sensors may eat a few dozens millisecond per polling iteration. Total polling time for all sensors you enable and polling time to polling period ratio is the key factor defining performance hit, for example if total polling time is approximately 50 milliseconds and you poll sensors once per second (i.e. each 1000 milliseconds), it means that monitoring eats roughly 5% on one CPU core.
    MSI AB monitoring module contain status bar (accessible via context menu), which is showing you all you need to configure it properly: total sensors count, total polling time and polling period. Additionally, there is a performance profiler panel, which is displaying sorted diagram of per-sensor polling times, allowing you to identify the main performance offender (or the slowest sensor) if necessary. HwInfo also provides similar monitoring performance profiling features.

    And yes, all that was discussed a few times and it is reflected in MSI AB context help.
     
    Aserback, boogieman, CalculuS and 2 others like this.
  6. mbk1969

    mbk1969 Ancient Guru

    Messages:
    12,807
    Likes Received:
    11,008
    GPU:
    GF RTX 3060TI
    Since I am not a competitive player and I see no difference between 500Hz and 1000Hz mouse polls in my games I see no point in 1000Hz. Modern CPUs should not have troubles with 1000Hz mouse polling rate but in case rig contains glitching or badly coded device drivers (ISRs and/or DPCs) interfering with mouse interrupts user can notice "stutters".
     
    BlindBison likes this.
  7. aufkrawall2

    aufkrawall2 Ancient Guru

    Messages:
    2,097
    Likes Received:
    636
    GPU:
    3060 TUF
    I think CPU load by high mouse polling rate in idle also appears higher than it really is due to pstates. I remember the reported CPU load declined a lot after disabling SpeedStep in bios, though last time I checked that was years ago.
     
    Last edited: Jul 2, 2022
    BlindBison likes this.
  8. BlindBison

    BlindBison Maha Guru

    Messages:
    1,451
    Likes Received:
    534
    GPU:
    RTX 2080 Super
    Thank you for explaining this so thoroughly, I really appreciate it.
     
  9. BlindBison

    BlindBison Maha Guru

    Messages:
    1,451
    Likes Received:
    534
    GPU:
    RTX 2080 Super
    Thank you very much, I’ll need to look into whether/how I can disable the unused sensors within HWiNFO64.
     
  10. Astyanax

    Astyanax Ancient Guru

    Messages:
    13,998
    Likes Received:
    5,641
    GPU:
    GTX 1080ti
    building on what Unwinder has said here, some HDD's are better than others for SMART polling periods, WD and Seagate tend to be pretty quick at returning the data, where as Samsung and Hitachi have a lot of extra data that contributes to longer smart polling times, all Spinpoint F3's to my knowledge take 3x as long to return data to tools such as hdtune compared to WD and Seasonic, and hitachi/wd enterprise disks don't even have all smart data exposed.

    in the case of the F3's, this means HwInfo actually causes stuttering with every refresh of the smart data, that i cannot replicate with hdd's from other vendors.
     
    Last edited: Sep 5, 2022
    BlindBison likes this.

  11. Unwinder

    Unwinder Moderator Staff Member

    Messages:
    16,066
    Likes Received:
    3,659
    Absolutely correct, it is HDD manufacturer specific and some drives report SMART data much faster than others. In any case, you cannot blame them for that - SMART protocol was initially intended for one-time diagnostics rather than periodic frequent realtime access.
     
    BlindBison likes this.
  12. Unwinder

    Unwinder Moderator Staff Member

    Messages:
    16,066
    Likes Received:
    3,659
    In addition to that, RTSS OverlayEditor's own minimalistic monitoring core also provides similar monitoring performance profiling features. If you open "Add new data sources" window and hold <Ctrl> button pressed, it will display internal HAL's per-sensor polling times instead of sensor values. For example on this screenshot you may see that the slowest sensor on my 3090ti is "GPU1 power percent", which is eating approximately 8ms per polling iteration. The rest GPU sensors are much much faster and eat much less CPU time. Also you may use OverlayEditor's %PollingTime0% and %PollingTime1% macroses to display polling times directly inside overlay, %PollingTime1% is the most important one as it reflects total polling time for all internal HAL sensors you added to your overlay. %PollingTime0% reflects polling time used to extract data from extenral providers (HwInfo, AIDA or windows pertformance counters), it is much less variable.

    upload_2022-9-5_12-28-56.png
     

Share This Page