Windows: Line-Based vs. Message Signaled-Based Interrupts. MSI tool.

Discussion in 'Operating Systems' started by mbk1969, May 7, 2013.

  1. narukun

    narukun Master Guru

    Messages:
    212
    Likes Received:
    22
    GPU:
    EVGA GTX 970 1561/7700
    Thank you for replying, I know that my kind of posts are the worst haha but you replied anyways and thats really nice of you, I'm reading the info you sent!
     
    mbk1969 likes this.
  2. EdKiefer

    EdKiefer Ancient Guru

    Messages:
    2,516
    Likes Received:
    241
    GPU:
    MSI 970 Gaming 4G
    well some drivers set them high, like Networking, but I would leave at defaults until you have some kind of issue.
    An example I had issue with the mouse not responding right in BFx games, setting one to higher might help (it didn't really in my case) but others have reported improvements.
    I don't think anyone can give a blanket recommendation as it depends on HW and software in the system.
     
  3. Smough

    Smough Master Guru

    Messages:
    528
    Likes Received:
    89
    GPU:
    GTX 1060 3GB
    More devices on High will decrease DPC latency, but if you set too many on High, it could cause problems and they could not work correctly. Try GPU and LAN on High and test it like that. Generally, networks cars tend to have some latency spikes so this is why most of them come on MSI mode by default and with High priority, to diminish this.
     
    Last edited: Aug 15, 2020
  4. mbk1969

    mbk1969 Ancient Guru

    Messages:
    9,665
    Likes Received:
    6,556
    GPU:
    GF RTX 2070 Super
    If you will have questions after reading linked doc, ask them away...
     

  5. GSDragoon

    GSDragoon Master Guru

    Messages:
    236
    Likes Received:
    77
    GPU:
    AMD Radeon VII
    If everything is a high priority, then nothing is a high priority.
     
  6. shiga94

    shiga94 New Member

    Messages:
    2
    Likes Received:
    0
    GPU:
    msi trio 2080
    I have read some of the messages, I just want to know if based on these two images, there is something wrong, because I still have micro stutter and can't lower the delay on my gpu.

    [​IMG]
    [​IMG]
     
  7. mbk1969

    mbk1969 Ancient Guru

    Messages:
    9,665
    Likes Received:
    6,556
    GPU:
    GF RTX 2070 Super
  8. shiga94

    shiga94 New Member

    Messages:
    2
    Likes Received:
    0
    GPU:
    msi trio 2080
    Not sure if I did it right, would you say this was done properly?:

    --------------------------
    DPC Info

    --------------------------
    CPU Usage Summing By Module For the Whole Trace

    CPU Usage from 22808542 us to 31669194 us:

    CPU 0 Usage, CPU 1 Usage, CPU 2 Usage, CPU 3 Usage, CPU 4 Usage, CPU 5 Usage, CPU 6 Usage, CPU 7 Usage,
    usec %, usec %, usec %, usec %, usec %, usec %, usec %, usec %, Module
    98292 1.11, 25148 0.28, 550 0.01, 410 0.00, 476 0.01, 571 0.01, 1458 0.02, 830 0.01, "Unknown"

    Total = 20818
    Elapsed Time, > 0 usecs AND <= 1 usecs, 4477, or 21.51%
    Elapsed Time, > 1 usecs AND <= 2 usecs, 3012, or 14.47%
    Elapsed Time, > 2 usecs AND <= 4 usecs, 2900, or 13.93%
    Elapsed Time, > 4 usecs AND <= 8 usecs, 4957, or 23.81%
    Elapsed Time, > 8 usecs AND <= 16 usecs, 3929, or 18.87%
    Elapsed Time, > 16 usecs AND <= 32 usecs, 1197, or 5.75%
    Elapsed Time, > 32 usecs AND <= 64 usecs, 330, or 1.59%
    Elapsed Time, > 64 usecs AND <= 128 usecs, 16, or 0.08%
    Total, 20818

    Total = 20818 for module "Unknown"
    Elapsed Time, > 0 usecs AND <= 1 usecs, 4477, or 21.51%
    Elapsed Time, > 1 usecs AND <= 2 usecs, 3012, or 14.47%
    Elapsed Time, > 2 usecs AND <= 4 usecs, 2900, or 13.93%
    Elapsed Time, > 4 usecs AND <= 8 usecs, 4957, or 23.81%
    Elapsed Time, > 8 usecs AND <= 16 usecs, 3929, or 18.87%
    Elapsed Time, > 16 usecs AND <= 32 usecs, 1197, or 5.75%
    Elapsed Time, > 32 usecs AND <= 64 usecs, 330, or 1.59%
    Elapsed Time, > 64 usecs AND <= 128 usecs, 16, or 0.08%
    Total, 20818

    All Module = 20818, Total = 20818, EQUAL

    --------------------------
    Usage From 22809 ms to 31669 ms, Summing In 1 second intervals. Intervals=9

    , CPU 0 Usage , CPU 1 Usage , CPU 2 Usage , CPU 3 Usage , CPU 4 Usage , CPU 5 Usage , CPU 6 Usage , CPU 7 Usage
    Start (ms) End (ms) , (usec) % , (usec) % , (usec) % , (usec) % , (usec) % , (usec) % , (usec) % , (usec) %
    22809-23809 :, 9136 0.91, 497 0.05, 162 0.02, 9 0.00, 28 0.00, 70 0.01, 121 0.01, 103 0.01
    23809-24809 :, 11456 1.15, 3244 0.32, 48 0.00, 41 0.00, 53 0.01, 99 0.01, 183 0.02, 83 0.01
    24809-25809 :, 8947 0.89, 3083 0.31, 73 0.01, 46 0.00, 72 0.01, 59 0.01, 163 0.02, 66 0.01
    25809-26809 :, 7950 0.79, 2842 0.28, 26 0.00, 69 0.01, 59 0.01, 62 0.01, 175 0.02, 131 0.01
    26809-27809 :, 12776 1.28, 3580 0.36, 57 0.01, 79 0.01, 61 0.01, 43 0.00, 143 0.01, 103 0.01
    27809-28809 :, 15529 1.55, 3475 0.35, 27 0.00, 47 0.00, 55 0.01, 60 0.01, 214 0.02, 91 0.01
    28809-29809 :, 14171 1.42, 2999 0.30, 43 0.00, 32 0.00, 55 0.01, 57 0.01, 141 0.01, 64 0.01
    29809-30809 :, 11073 1.11, 3084 0.31, 61 0.01, 47 0.00, 56 0.01, 32 0.00, 154 0.02, 107 0.01
    30809-31669 :, 7254 0.84, 2345 0.27, 52 0.01, 40 0.00, 37 0.00, 89 0.01, 164 0.02, 81 0.01

    --------------------------
    Interrupt Info

    --------------------------
    CPU Usage Summing By Module For the Whole Trace

    CPU Usage from 22808542 us to 31669194 us:

    CPU 0 Usage, CPU 1 Usage, CPU 2 Usage, CPU 3 Usage, CPU 4 Usage, CPU 5 Usage, CPU 6 Usage, CPU 7 Usage,
    usec %, usec %, usec %, usec %, usec %, usec %, usec %, usec %, Module
    55311 0.62, 0 0.00, 0 0.00, 0 0.00, 0 0.00, 0 0.00, 0 0.00, 0 0.00, "Unknown"

    Total = 10165
    Elapsed Time, > 0 usecs AND <= 1 usecs, 4309, or 42.39%
    Elapsed Time, > 1 usecs AND <= 2 usecs, 2016, or 19.83%
    Elapsed Time, > 2 usecs AND <= 4 usecs, 490, or 4.82%
    Elapsed Time, > 4 usecs AND <= 8 usecs, 347, or 3.41%
    Elapsed Time, > 8 usecs AND <= 16 usecs, 1713, or 16.85%
    Elapsed Time, > 16 usecs AND <= 32 usecs, 1222, or 12.02%
    Elapsed Time, > 32 usecs AND <= 64 usecs, 68, or 0.67%
    Total, 10165

    Total = 10165 for module "Unknown"
    Elapsed Time, > 0 usecs AND <= 1 usecs, 4309, or 42.39%
    Elapsed Time, > 1 usecs AND <= 2 usecs, 2016, or 19.83%
    Elapsed Time, > 2 usecs AND <= 4 usecs, 490, or 4.82%
    Elapsed Time, > 4 usecs AND <= 8 usecs, 347, or 3.41%
    Elapsed Time, > 8 usecs AND <= 16 usecs, 1713, or 16.85%
    Elapsed Time, > 16 usecs AND <= 32 usecs, 1222, or 12.02%
    Elapsed Time, > 32 usecs AND <= 64 usecs, 68, or 0.67%
    Total, 10165

    All Module = 10165, Total = 10165, EQUAL

    --------------------------
    Usage From 22809 ms to 31669 ms, Summing In 1 second intervals. Intervals=9

    , CPU 0 Usage , CPU 1 Usage , CPU 2 Usage , CPU 3 Usage , CPU 4 Usage , CPU 5 Usage , CPU 6 Usage , CPU 7 Usage
    Start (ms) End (ms) , (usec) % , (usec) % , (usec) % , (usec) % , (usec) % , (usec) % , (usec) % , (usec) %
    22809-23809 :, 5115 0.51, 0 0.00, 0 0.00, 0 0.00, 0 0.00, 0 0.00, 0 0.00, 0 0.00
    23809-24809 :, 6554 0.66, 0 0.00, 0 0.00, 0 0.00, 0 0.00, 0 0.00, 0 0.00, 0 0.00
    24809-25809 :, 5838 0.58, 0 0.00, 0 0.00, 0 0.00, 0 0.00, 0 0.00, 0 0.00, 0 0.00
    25809-26809 :, 5317 0.53, 0 0.00, 0 0.00, 0 0.00, 0 0.00, 0 0.00, 0 0.00, 0 0.00
    26809-27809 :, 5995 0.60, 0 0.00, 0 0.00, 0 0.00, 0 0.00, 0 0.00, 0 0.00, 0 0.00
    27809-28809 :, 7788 0.78, 0 0.00, 0 0.00, 0 0.00, 0 0.00, 0 0.00, 0 0.00, 0 0.00
    28809-29809 :, 7266 0.73, 0 0.00, 0 0.00, 0 0.00, 0 0.00, 0 0.00, 0 0.00, 0 0.00
    29809-30809 :, 6492 0.65, 0 0.00, 0 0.00, 0 0.00, 0 0.00, 0 0.00, 0 0.00, 0 0.00
    30809-31669 :, 4946 0.57, 0 0.00, 0 0.00, 0 0.00, 0 0.00, 0 0.00, 0 0.00, 0 0.00


    Distribution of number of 2000 ms intervals w.r.t. DPC/ISR usage:

    , CPU 0, CPU 1, CPU 2, CPU 3, CPU 4, CPU 5, CPU 6, CPU 7
    DPC/ISR Usage %, DPC ISR Combined, DPC ISR Combined, DPC ISR Combined, DPC ISR Combined, DPC ISR Combined, DPC ISR Combined, DPC ISR Combined, DPC ISR Combined
    >= 0 AND <= 1, 2, 5, 1, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5
    > 1 AND <= 5, 3, 0, 4, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
    > 5 AND <= 10, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
    > 10 AND <= 20, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
    > 20 AND <= 40, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
    > 40 AND <= 60, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
    > 60 AND <= 80, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
    > 80 AND <= 100, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
    ---
    Total: , 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5
     
  9. mbk1969

    mbk1969 Ancient Guru

    Messages:
    9,665
    Likes Received:
    6,556
    GPU:
    GF RTX 2070 Super
    That looks strange because we see only one module in trace. But trace shows no problems.
     
  10. shuvo030

    shuvo030 New Member

    Messages:
    4
    Likes Received:
    0
    GPU:
    RTX 2080
    Hi. Does enabling both MSI Mode and Hardware Accelerated GPU Scheduling (in WIndows 10) cause any issues or conflicts? HAGS give me better performance in games so I have to keep it ON. Since both are GPU latency related, I'm wondering if I should enable MSI Mode or not.
    Rig: i7 4790, RTX 2080
     

  11. Astyanax

    Astyanax Ancient Guru

    Messages:
    7,658
    Likes Received:
    2,549
    GPU:
    GTX 1080ti
    MSI's and HAGS are entirely unrelated in function.
     
  12. shuvo030

    shuvo030 New Member

    Messages:
    4
    Likes Received:
    0
    GPU:
    RTX 2080
    But don't both of them reduced latency GPU hardware wise? So they're related in that sense
     
  13. mbk1969

    mbk1969 Ancient Guru

    Messages:
    9,665
    Likes Received:
    6,556
    GPU:
    GF RTX 2070 Super
    They act in completely different fashion/field. If your rig manages to reduce latencies from both settings you are lucky.
     
  14. MIET4S

    MIET4S Member

    Messages:
    14
    Likes Received:
    0
    GPU:
    RX 5700 XT Nitro+ S
    Howdy,

    I found some weird results when it comes to IRQ and MSI methods on my modern gaming rig.
    Using MSI mode when its possible (nowdays u can set MSI to almost everything) I wasnt able to reach stable mouse polling no matter what i did.
    Using IRQ and eliminating conflicts for USB controllers (not even GPU) polling/mouse control was noticeably better (also in games). Can someone explain why, did anyone experience this? Im at work atm, will upload some screens with polling and MSI settings/IRQ conflicts later.
     
  15. mbk1969

    mbk1969 Ancient Guru

    Messages:
    9,665
    Likes Received:
    6,556
    GPU:
    GF RTX 2070 Super
    What exactly do you mean by "stable mouse polling", "polling/mouse control was noticeably better"?
     

  16. MIET4S

    MIET4S Member

    Messages:
    14
    Likes Received:
    0
    GPU:
    RX 5700 XT Nitro+ S
    Polling Hz variation with IRQ method dropped from 40us to 5-10 us but still with occasional spikes. Mouse input became less floaty (thought it was placebo but graphs looked different). Overall DPC latency slightly dropped aswell. This is mind blowing for me since i can't really comprehend why this happens, should be other way around - improvements should take place when switching to MSI mode.

    Can't really discuss this without any screenshots, I will upload them later.
     
  17. mbk1969

    mbk1969 Ancient Guru

    Messages:
    9,665
    Likes Received:
    6,556
    GPU:
    GF RTX 2070 Super
    MSI mode is not about DPC stage, it affects the ISR stage which precedes the DPC one.
    I appreciate the visualization made with appropriate software, but you can misinterpret the data. Or you can be misinformed by people claiming something about those visualizations.
    Do you feel any difference in mouse?
     
  18. MIET4S

    MIET4S Member

    Messages:
    14
    Likes Received:
    0
    GPU:
    RX 5700 XT Nitro+ S
    I do feel the difference and im quite sensitive to those changes.

    I gotta admit I am confused atm because I thought Ryzen systems are just worse when it comes to latency(a lot of web data about this one) and got over it since I've tried everything to achieve stable mouse polling with no success.
    Lately i've unchecked everything from MSI mode just to see what happens and got some weird results which might lead me to further experiments(Would be great if you take look at it since im tired of tinkering all the time). Will post them later.
     
  19. mbk1969

    mbk1969 Ancient Guru

    Messages:
    9,665
    Likes Received:
    6,556
    GPU:
    GF RTX 2070 Super
    Better do experiment with one device at a time. This way you will get more precise results.
    I will wait screenshots with explanation. I have seen one time a clip where guy demonstrated the software which captures the mouse stats, but the guy never answered my questions there.
     
  20. shuvo030

    shuvo030 New Member

    Messages:
    4
    Likes Received:
    0
    GPU:
    RTX 2080
    Hardware Accelerated GPU Scheduling + MSI Mode: ON = 300-ish FPS
    Hardware Accelerated GPU Scheduling + MSI Mode: OFF = 350+ FPS
    Tested on Wolfenstein II: The New Colossus (Vulkan)
    Have to test on other games in GPU bound scenarios.
    Will give updates later.
     

Share This Page