Using an external USB PCIE card won't solve on some mobos

[Vocaleyes]

So i've just discovered that on some chipset architecture even an external USB card with it's own dedicated chip won't solve certain issues.
This is due to certain chipsets having shared DPC queues and non-deterministic USB interrupt routing, even across PCIe add-in cards.

This means even if the mouse is on an isolated USB host, activity from the keyboard on the motherboard USB may still inject latency spikes into the same DPC queue handling mouse movement.

That irregularity can cause microstutter (even if FPS is stable), result in slight drifting, missed frames, or inconsistent cursor velocity & be especially noticeable in fast swipes or slow pixel-precise movement.

Thought i'd share that info with you all.

[Slender]

do you try mcs9990 with usbport.sys driver?

[kyube]

So i've just discovered that on some chipset architecture even an external USB card with it's own dedicated chip won't solve certain issues.

What “issues”?

This is due to certain chipsets having shared DPC queues and non-deterministic USB interrupt routing, even across PCIe add-in cards.

This means even if the mouse is on an isolated USB host, activity from the keyboard on the motherboard USB may still inject latency spikes into the same DPC queue handling mouse movement.

That irregularity can cause microstutter (even if FPS is stable), result in slight drifting, missed frames, or inconsistent cursor velocity & be especially noticeable in fast swipes or slow pixel-precise movement.

You're free to provide a source for these claims or repeatable testing data based on ETW.

[Vocaleyes]

So i've just discovered that on some chipset architecture even an external USB card with it's own dedicated chip won't solve certain issues.

What “issues”?

This is due to certain chipsets having shared DPC queues and non-deterministic USB interrupt routing, even across PCIe add-in cards.

This means even if the mouse is on an isolated USB host, activity from the keyboard on the motherboard USB may still inject latency spikes into the same DPC queue handling mouse movement.

That irregularity can cause microstutter (even if FPS is stable), result in slight drifting, missed frames, or inconsistent cursor velocity & be especially noticeable in fast swipes or slow pixel-precise movement.

You're free to provide a source for these claims or repeatable testing data based on ETW.

Quite literally listed those "issues" in my initial post? you're free to read it again.

Anyway...

All HID devices (keyboard, mouse, etc.) use:

HIDCLASS.sys
HIDPARSE.sys
USBPORT.sys or USBXHCI.sys

Even if routed through separate USB controllers, these devices still go through shared kernel-mode stack layers, which may introduce:

Prioritization issues
DPC queue pressure
Delays in mouse input processing when keyboard input is active.


Observable with tools such as:

LatencyMon (High DPC from USBPORT.sys, HIDCLASS.sys, or Wdf01000.sys)
MouseTester (Irregular polling during keyboard input)
USB Tree Viewer (Shows controller separation but cannot diagnose DPC routing)

If there's anything else I can further help you understand, don't hesitate to ask.

People are more than capable to run the aforementioned tests themselves to check, but I personally have already posted and shared my DPC results, so if your next response is "do it again then", good luck with that.

[kyube]

Observable with tools such as:
LatencyMon (High DPC from USBPORT.sys, HIDCLASS.sys, or Wdf01000.sys)
MouseTester (Irregular polling during keyboard input)
USB Tree Viewer (Shows controller separation but cannot diagnose DPC routing)

Hope you know LatencyMon is completely useless in evaluating this.
MouseTester is also a toy, comparatively.
What you're looking for is Microsoft's Media eXperience Analyzer.
https://github.com/djdallmann/GamingPCS ... /README.md
https://learn.microsoft.com/en-us/windo ... experience
https://learn.microsoft.com/en-us/windo ... exercise-3
https://learn.microsoft.com/en-us/windo ... exercise-4

If you want something more digestible: https://github.com/valleyofdoom/xtw

Have fun digging.

[skdxkiller]

How does one fix said mouse jitter/stutter??

Have it across two systems

[Slender]

how much i fucking hate kyube.
[Note Added by Moderation Team: Please follow forum rules when criticizing. This forum is not social media.]

instead of asking the person how he found out what he writes (because it is really interesting), this person, starts to bring his favorite set of latencymoon nonsense, and his fucking stories about ram tuning, as if he is the only one of all of us who solved this problem. fuck off.

[kyube]

how much i fucking hate kyube.

You've necro'd a thread made in July 2025 to vent your frustrations towards me, impressive.
Emotions don't care about objective facts though
You're better off investing your time in reading books or 1st party articles in regards to Windows troubleshooting than spouting nonsense around here

instead of asking the person how he found out what he writes (because it is really interesting), this person, starts to bring his favorite set of latencymoon nonsense, and his fucking stories about ram tuning,

99% of users commonly browsing here don't know how to properly troubleshoot software & hardware problems on these forums using 1st party methods.
MouseTester is a data visualization of bInterval (or at least it should be)
LatencyMon is a 3rd party tool which doesn't capture DPC & ISR usage properly.

The 1st party alternatives are:
Windows Performance Analyzer (WPA)
Windows Media eXperience Analyzer (MXA)
XTW (not proper 1st party, but relies on ETW)

I don't care about random ramblings based on erroneous benchmarking methods, nor do I care about 'feel' of users which cannot understand human psychological bias & the daily variance in human physiology.
I have no favorites, I only care about what's objectively right & wrong.

[Slender]

how much i fucking hate kyube.

You've necro'd a thread made in July 2025 to vent your frustrations towards me, impressive.
Emotions don't care about objective facts though
You're better off investing your time in reading books or 1st party articles in regards to Windows troubleshooting than spouting nonsense around here

instead of asking the person how he found out what he writes (because it is really interesting), this person, starts to bring his favorite set of latencymoon nonsense, and his fucking stories about ram tuning,

99% of users commonly browsing here don't know how to properly troubleshoot software & hardware problems on these forums using 1st party methods.
MouseTester is a data visualization of bInterval (or at least it should be)
LatencyMon is a 3rd party tool which doesn't capture DPC & ISR usage properly.

The 1st party alternatives are:
Windows Performance Analyzer (WPA)
Windows Media eXperience Analyzer (MXA)
XTW (not proper 1st party, but relies on ETW)

I don't care about random ramblings based on erroneous benchmarking methods, nor do I care about 'feel' of users which cannot understand human psychological bias & the daily variance in human physiology.
I have no favorites, I only care about what's objectively right & wrong.

if mousetester shows delta 995-1005, and after some changes it show 999-1001, and it happens every time when i change thing, I think that instrument shows true.
same with latency moon.

[krisphi0]

Interesting point about the precision of those deltas. I've been trying to replicate that level of visualization in a browser-based environment so people don't have to download the .exe.

I built HardwareTest org using performance.now() to capture timestamps, but I'm constantly fighting against the browser's event loop overhead.

Since you seem to have a solid baseline with the native MouseTester app, would you mind giving the web version a quick spin? I'm curious if the "Peak Hz" and stability on the web tool align with what you're seeing in the native app (999-1001), or if the browser introduces too much jitter to be reliable for that specific 1-2Hz variance.

Any feedback on the accuracy would be appreciated.