LatencyMon Results

[Saphyen]

I am kinda getting discouraged at this point. My hardwear is pretty good. I have a RTX 3080Ti, an i9-13900k watercooled, 32gb of ddr5 ram, with an Asrock z790 pro rs mobo. My whole PC has been overhauled and the only things that remain the same are my watercooler and my gpu. On paper I should not be having any latency issues, but I just simply am and my morale is just tanking. Don't know what it could be and don't know what I can do. I've even paid 3 seperate people like 50 bucks to optimize my pc who are reliable in the gaming scene I'm in and who do pros, but it did not do anything. To show what I mean, this is on a fresh windows 11 install (Was on 10 before)

1.png (33.09 KiB) Viewed 9504 times

[Fpsgamer]

I am kinda getting discouraged at this point. My hardwear is pretty good. I have a RTX 3080Ti, an i9-13900k watercooled, 32gb of ddr5 ram, with an Asrock z790 pro rs mobo. My whole PC has been overhauled and the only things that remain the same are my watercooler and my gpu. On paper I should not be having any latency issues, but I just simply am and my morale is just tanking. Don't know what it could be and don't know what I can do. I've even paid 3 seperate people like 50 bucks to optimize my pc who are reliable in the gaming scene I'm in and who do pros, but it did not do anything. To show what I mean, this is on a fresh windows 11 install (Was on 10 before) 1.png

My results with just high performance power plan and a good nvidia driver are perfect on w10. below 2ms latency.
w11 no matter what you do it's around 8ms. It's kinda inaccurate on w11 plus w11 has more latency overall.
I never found latencymon to work well on w11.

[Saphyen]

It is the same on windows 10. I played on win10 up until 2 days ago when I decided to switch and see if my issues would somehow be fixed

[Fpsgamer]

It is the same on windows 10. I played on win10 up until 2 days ago when I decided to switch and see if my issues would somehow be fixed

The 2 most important things I'd say is power plan,Nvidia driver and drivers in general when it comes to it.
Try a confirmed "good" driver to see if it makes any difference, 537.58 for example. (Make sure there is no GeForce experience and stuff like that)

[kyube]

LatencyMon is not the proper way to evaluate Windows driver DPC/ISR performance.
The LatencyMon driver itself causes severe overhead and thus impacts results severely.
The metrics it displays are also misleading.

Why is LatencyMon not desirable to objectively measure DPC/ISR driver performance?
(a small write-up from different users)

the latencymon dev admitted that he does not know why ISRs are not picked up in latencymon after i sent a output of both the driver tab and a xperf trace, i can show the email

latencymon is dogshit because
1) there is no option for a timeout/delay before recording starts for you to tab in to game which already makes results invalid
2) the devs @ resplendence admitted they don't know why it doesn't pick up isrs (i doubt it picks up dpcs properly either)
3) no detailed stats for individual drivers
4) the KM driver that comes with latencymon (rspLLL64.sys) generates a crap ton of dpcs alone (invalid results)
5) the "average measured" values in the main page is useless/invalid due to reason 4, (how is average calculated anyway?)
there is absolutely no reason not to use xperf for debugging driver latencies

minimizing driver execution times is still important but i agree in regards to latencymon not being a great tool. it doesn't pick up ISRs/DPCs at all for some drivers and the driver for the program (rspLLL.sys) generates DPCs itself which cucks the average
developer doesn't seem to care at all to fix monitoring when I contacted them but there's no alternative so maybe i can look into extracting average/stdev out of xperf
also doesn't measure interrupt to dpc latency which makes it even more useless

Mr.T: anyone actually ever mess with the settings?
Mr.T: yea latencymon doesn't even match xperf readings
Mr.T: regardless of what settings i choose
Mr.T: thought its funny
Mr.T: so many people in here measuring with that garb
Mr.T: i feel like most of the latency is that of the software/driver
Mr.T: its wierd because it says it's using ETLs...
Mr.T: doesn't make sense
Mr.T: lol

[15:25]Deleted User: I don't trust the interrupt to DPC latency (µs) metric either, i've asked tons of people how to go about calculating this and they've said it is not possible
[15:26]Deleted User: including one of the authors of windows internals and a few RE folks
[15:26]vincentyoyoyo: mhm it's very much fucked
[20:02]timecard: only way its possible is by writing the driver that generates the interrupt and getting the timing information (what latencymon does)
[20:10]timecard: on interrupt get high resolution timestamp then get the dpc complete timestamp and get the difference
[22:46]Deleted User: that's not a real-word representation of ISR-to-DPC times since in reality, they could be scheduled on difference cores. that approach assumes a lot of things in regards to the scenario

Instead, use ETL-based scripts such as xperf or xtw
https://github.com/valleyofdoom/PC-Tuning (.bat file is inside)
https://github.com/valleyofdoom/xtw
ETL-based measurements are approved by Microsoft, they are the official way to troubleshoot driver problems.
LatencyMon is a 3rd party, a shoddy one at that.
https://forums.guru3d.com/threads/simpl ... rs.423884/

For optimal driver performance, what you're looking for is:
1.) ISR's should always be <1µs
2.) DPC's should be as tight as possible (1µs)

Other than this, you can also use Intel VTune or AMD's µProf to evaluate performance of your hardware.

Before attempting any kind of benchmarking, you need to understand what you're testing.
If you are testing USB devices (drivers), then simple circles with your mouse in-game or in idle should reveal you one part of the story.
If you are testing your NIC drivers, use iperf or in-game of choice
If you are testing audio drivers, test in-game or browser

Of course, you can supplement this data with MXA (Media Experience Analyzer, Microsoft's tool) to figure out thread durations & more granular information which make up the total system latency chain.

For more information, read through:
https://github.com/valleyofdoom/PC-Tuning
https://github.com/BoringBoredom/PC-Optimization-Hub
https://calypto.us/ (he doesn't use latencymon for drivers as far as i'm aware, even though he has a image of it in his guide)
https://github.com/djdallmann/GamingPCSetup

You should also read through Intel's documentation in regards to what's good & what isn't in terms of BIOS setting choices.
For example, Intel's SpeedShift is required for Windows (W11 kernel only) to properly utilise the new heterogenous architecture on LGA1700 socket.
HT can also impact DPC/ISR performance and lead to a latency impact.
e.g. of how much it can impact performance:
https://community.intel.com/t5/Software ... d-p/959467
https://www.intel.com/content/www/us/en ... dware.html


Hope this helps.

[Saphyen]

LatencyMon is not the proper way to evaluate Windows driver DPC/ISR performance.
The LatencyMon driver itself causes severe overhead and thus impacts results severely.
The metrics it displays are also misleading.

Why is LatencyMon not desirable to objectively measure DPC/ISR driver performance?
(a small write-up from different users)

the latencymon dev admitted that he does not know why ISRs are not picked up in latencymon after i sent a output of both the driver tab and a xperf trace, i can show the email

latencymon is dogshit because
1) there is no option for a timeout/delay before recording starts for you to tab in to game which already makes results invalid
2) the devs @ resplendence admitted they don't know why it doesn't pick up isrs (i doubt it picks up dpcs properly either)
3) no detailed stats for individual drivers
4) the KM driver that comes with latencymon (rspLLL64.sys) generates a crap ton of dpcs alone (invalid results)
5) the "average measured" values in the main page is useless/invalid due to reason 4, (how is average calculated anyway?)
there is absolutely no reason not to use xperf for debugging driver latencies

minimizing driver execution times is still important but i agree in regards to latencymon not being a great tool. it doesn't pick up ISRs/DPCs at all for some drivers and the driver for the program (rspLLL.sys) generates DPCs itself which cucks the average
developer doesn't seem to care at all to fix monitoring when I contacted them but there's no alternative so maybe i can look into extracting average/stdev out of xperf
also doesn't measure interrupt to dpc latency which makes it even more useless

Mr.T: anyone actually ever mess with the settings?
Mr.T: yea latencymon doesn't even match xperf readings
Mr.T: regardless of what settings i choose
Mr.T: thought its funny
Mr.T: so many people in here measuring with that garb
Mr.T: i feel like most of the latency is that of the software/driver
Mr.T: its wierd because it says it's using ETLs...
Mr.T: doesn't make sense
Mr.T: lol

[15:25]Deleted User: I don't trust the interrupt to DPC latency (µs) metric either, i've asked tons of people how to go about calculating this and they've said it is not possible
[15:26]Deleted User: including one of the authors of windows internals and a few RE folks
[15:26]vincentyoyoyo: mhm it's very much fucked
[20:02]timecard: only way its possible is by writing the driver that generates the interrupt and getting the timing information (what latencymon does)
[20:10]timecard: on interrupt get high resolution timestamp then get the dpc complete timestamp and get the difference
[22:46]Deleted User: that's not a real-word representation of ISR-to-DPC times since in reality, they could be scheduled on difference cores. that approach assumes a lot of things in regards to the scenario

Instead, use ETL-based scripts such as xperf or xtw
https://github.com/valleyofdoom/PC-Tuning (.bat file is inside)
https://github.com/valleyofdoom/xtw
ETL-based measurements are approved by Microsoft, they are the official way to troubleshoot driver problems.
LatencyMon is a 3rd party, a shoddy one at that.
https://forums.guru3d.com/threads/simpl ... rs.423884/

For optimal driver performance, what you're looking for is:
1.) ISR's should always be <1µs
2.) DPC's should be as tight as possible (1µs)

Other than this, you can also use Intel VTune or AMD's µProf to evaluate performance of your hardware.

Before attempting any kind of benchmarking, you need to understand what you're testing.
If you are testing USB devices (drivers), then simple circles with your mouse in-game or in idle should reveal you one part of the story.
If you are testing your NIC drivers, use iperf or in-game of choice
If you are testing audio drivers, test in-game or browser

Of course, you can supplement this data with MXA (Media Experience Analyzer, Microsoft's tool) to figure out thread durations & more granular information which make up the total system latency chain.

For more information, read through:
https://github.com/valleyofdoom/PC-Tuning
https://github.com/BoringBoredom/PC-Optimization-Hub
https://calypto.us/ (he doesn't use latencymon for drivers as far as i'm aware, even though he has a image of it in his guide)
https://github.com/djdallmann/GamingPCSetup

You should also read through Intel's documentation in regards to what's good & what isn't in terms of BIOS setting choices.
For example, Intel's SpeedShift is required for Windows (W11 kernel only) to properly utilise the new heterogenous architecture on LGA1700 socket.
HT can also impact DPC/ISR performance and lead to a latency impact.
e.g. of how much it can impact performance:
https://community.intel.com/t5/Software ... d-p/959467
https://www.intel.com/content/www/us/en ... dware.html


Hope this helps.

I did as you said and ran xtw for more accurate results and here they are. I ran it when I was playing my main game.

Screenshot 2025-04-21 221133.png (71.89 KiB) Viewed 8968 times

Screenshot 2025-04-21 221105.png (31.34 KiB) Viewed 8968 times

[kyube]

What version of Windows 11 are you on? You can verify this by using the “winver” command in the Run window (CTLR+R)
Have you done any optimisations to the OS?
What Nvidia driver version are you using? What other drivers (NIC, storage) have you installed? I can tell that your USB drivers are the stock MS ones (wdf0100.sys)
What SSD are you using? What are your current RAM timings? Send a Asrock Timing Configurator or MemTweakIt screenshot. What PSU are you using? Are you using any RGB? What peripherals (incl. audio gear) are you using?
What game have you tested this in? Which scenario have you tested (what were you doing while testing—mouse movement, audio, speaking into mic...)

Are you noticing any anomalies in this main game of yours which made you go down the troubleshooting route?

A few notes on your ETL results:
- This is the first time I've seen these 2 entries: symcryptk.dll & KsId.sys (or Ks1d.sys?) Figuring out what they are could help
One source I found was: https://forums.tomshardware.com/threads ... y.3858828/

- The ntfs.sys DPCs last for a longer duration than usual. After you give details needed above, I can give you recommendations for this.

[Saphyen]

What version of Windows 11 are you on? You can verify this by using the “winver” command in the Run window (CTLR+R)
Have you done any optimisations to the OS?
What Nvidia driver version are you using? What other drivers (NIC, storage) have you installed? I can tell that your USB drivers are the stock MS ones (wdf0100.sys)
What SSD are you using? What are your current RAM timings? Send a Asrock Timing Configurator or MemTweakIt screenshot. What PSU are you using? Are you using any RGB? What peripherals (incl. audio gear) are you using?
What game have you tested this in? Which scenario have you tested (what were you doing while testing—mouse movement, audio, speaking into mic...)

Are you noticing any anomalies in this main game of yours which made you go down the troubleshooting route?

A few notes on your ETL results:
- This is the first time I've seen these 2 entries: symcryptk.dll & KsId.sys (or Ks1d.sys?) Figuring out what they are could help
One source I found was: https://forums.tomshardware.com/threads ... y.3858828/

- The ntfs.sys DPCs last for a longer duration than usual. After you give details needed above, I can give you recommendations for this.

My windows version is 24H2 build 26100.3775. This is a fresh installed OS and nothing has been done to it, I was also on win10 up until about 3 days ago or so and upgraded to hopefully fix any issues. Using Nvidia driver version 576.02. I have installed my motherboard drivers which are Realtek high definition audio driver ver:9381.1, INF driver ver:10.1.19600.8418, Intel Management Engine driver ver:2340.5.36.0_CONS, and Realtek Lan driver ver:10.060.0615.2022.My ssds are CT2000P5PSSD8 2TB and PNY CS2130 1TB.
My PSU is the Corsair RM850e and I have rgb case fans, gpu has some rgb, and my cpu cooler has rgb. Razer Viper v2 pro mouse, Razer Huntsman tournament edition keyboard, GOXLR, and a Cyberpunk themed Steelseries arctis 1 headset that came with the Cyberpunk bundle thing. The main game I play is Apex Legends so I ran the thing in it while moving my mouse around. I have also decided to redo the test as I did it in the firing range before and now I show it in an actual live match. It feels like my input constantly just cuts out and cuts back in over and over sometimes.

Screenshot 2025-04-22 131321.png (205.2 KiB) Viewed 8668 times

Screenshot 2025-04-22 132851.png (75.26 KiB) Viewed 8668 times

Screenshot 2025-04-22 132831.png (30.68 KiB) Viewed 8668 times

[kyube]

My windows version is 24H2 build 26100.3775. This is a fresh installed OS and nothing has been done to it, I was also on win10 up until about 3 days ago or so and upgraded to hopefully fix any issues. Using Nvidia driver version 576.02. I have installed my motherboard drivers which are Realtek high definition audio driver ver:9381.1, INF driver ver:10.1.19600.8418, Intel Management Engine driver ver:2340.5.36.0_CONS, and Realtek Lan driver ver:10.060.0615.2022.My ssds are CT2000P5PSSD8 2TB and PNY CS2130 1TB.
My PSU is the Corsair RM850e and I have rgb case fans, gpu has some rgb, and my cpu cooler has rgb. Razer Viper v2 pro mouse, Razer Huntsman tournament edition keyboard, GOXLR, and a Cyberpunk themed Steelseries arctis 1 headset that came with the Cyberpunk bundle thing. The main game I play is Apex Legends so I ran the thing in it while moving my mouse around. I have also decided to redo the test as I did it in the firing range before and now I show it in an actual live match. It feels like my input constantly just cuts out and cuts back in over and over sometimes.

I can give you a few recommendations:
1.) Switch to W11 23H2 or a earlier Windows 10 version (1803, 1809, 1909, don't go above or below these)
24H2 had quite a few issues which can impact overall performance.
Refer to the guides I've sent in my 1st post to you, PC-tuning has a easy-to-do .reg file installation process.
Reading through them will help understanding the why
You install a stock W11 23H2 ISO (with Ventoy) without the Ethernet jack plugged in and then run the .reg file afterwards. Then, you'll have a clean slate to work off.
W11 is preferred (with SpeedShift enabled), if you'd like to keep using e-cores to their fullest potential, as W11 has specific kernel additions to make the heterogeneous architecture work.
Disabling HT might lead to a performance increase in Source-based games.
Source engine-based games don't benefit as much from memory bandwidth increases & offer slight benefits from total memory latency decreases, meaning that chasing any manual RAM timings might be detrimental.
I would even go as far as disabling XMP if your RAM kit is not in the QVL of your motherboard and only then reinstalling your OS

2.) Install previous branches of Nvidia drivers
A few examples of decent versions include: 551.23, 566.36
The recent R572 branch & newest R576 branch seem very buggy & causes a performance decrease.
It's difficult to give you the best possible driver for your CPU+GPU combination, considering nobody tests for this (especially with the sheer difference in Windows versions)
I would also recommend only installing the NIC drivers (W10 version, not W11 one) & GPU drivers, nothing else.
Opt to download both from Realtek & Nvidia, not from Asrock's support page.

3.) Consider turning off or removing the RGB if possible, they add unnecessary EMI to your system.

4.) Consider purchasing a 2nd USB controller to alleviate USB-based interrupts from your GOXLR or keyboard
ASMedia-based USB controllers seem to be good. You can experiment with the stock MS one or the Asmedia-specific one and see which you prefer.
e.g.: https://www.aliexpress.com/item/1005005010778591.html

I'd argue the first two mentions, provided you haven't messed with your BIOS heavily and/or degraded your HW components and/or have some instability, will solve your problem.
The last 2 can be helpful, but not as important as the other ones.

[Saphyen]

Would you happen to know what the massive spike in nvlddmkm.sys in the new test results I recorded could be? It seems to be an Nvidia thing. Also to note that about a year back I used Nvidias own latency tester frameview and in the latency section it was recorded as all over the place like that. At the time I thought this was just a bug in the Nvidia frameview software and paid no note to it since it's a weirdly high number. Could this be a much deeper problem with my gpu?