Windows 11 24h2 has worse latency than Windows 10 22h2!

[grvn]

Hello!

I've been testing Windows 11 24h2 and I always come back to Windows 10 22h2 due to various reasons.

Mostly relates to gaming performance, which, in my case, is better and smoother on Windows 10. The 1% and 0.1% are higher on Windows 10 for me.

The one glaring issue I have found is that Windows 11 has terrible latency?

I've installed LatencyMon and played one CS2 competitive match, while LatencyMon was running in the background.

These are the results for Windows 10 22h2:

These are the results for Windows 11 24h2:

Note that the same drivers, same config, same settings etc. have been used on both Windows 10 / 11.

As you can well see, the latency is just atrocious in Windows 11!

	Windows 10 22h2	Windows 11 24h2	Difference (%)
Current Interrupt Process Latency	25.90	78.70	148.23 %
Highest measured Process Latency	241.30	599.00	148.23 %
Highest ISR execution time	50.15	141.46	182.07 %
Highest DPC execution time	291.40	1725.75	492.22 %

 

[Vayra86]

Your hardware plays a part in this too of course; would be nice to know what system you're doing this on.

 

[Ruru]

The latency thing seems to be in overall hella dependant on the hardware used. No issues here with 11 24H2.

-Crosshair VII (BIOS 5502)
-Ryzen 5800X @ PBO +200 CU -20 allcore
-32GB DDR4-3466 16-18-18-36-1T
-RTX 3080 OC
-Win 11 Pro 24H2

 

[Vayra86]

Here's mine while running an installation / file consistency check.

System in specs to the left. Seems like I'm beating your W1022h2 results. You might need to take a longer look at your system altogether.

It does seem like I have a high pagefault count Gonna look into that. But could be related to actually doing an installation.

 

[grvn]

The latency thing seems to be in overall hella dependant on the hardware used. No issues here with 11 24H2.

System in specs to the left. Seems like I'm beating your W1022h2 results. You might need to take a longer look at your system altogether.

Ryzen 5800x3d / RTX 4070 / Audient ID4 Interface

Also try to play a game while LatencyMon is enabled in background. I played one game of CS2 on both OS.

 

[Ruru]

Ryzen 5800x3d / RTX 4070 / Audient ID4 Interface

Also try to play a game while LatencyMon is enabled in background. I played one game of CS2 on both OS.

Which motherboard and do you have chipset drivers installed?

Forgot to mention that I have the newest which are available on today (12.1.)

 

[Bill_Bright]

I don't really see how such a comparison can be done fairly. There are just too many variables at play here.

As others noted, hardware plays a part here. In fact, it plays a HUGE part. And the problem is hardware tends to be optimized to work best with the operating system that was current at the time the HW was designed.

Similar conditions apply to the drivers of such hardware. Yes, developers may be able to create W11 drivers for that 5 year old graphics card, but will it be as efficient as W10 drivers running in W10? I don't know - and because I don't know, how can I say it is a fair comparison? I can't.

My point is, I don't think it totally fair to compare W10 and W11 results when running both OSs on hardware designed for W10. Same applies if running both on W11 hardware.

I just don't see how any test computer can be configured so it will impose the exact same conditions on 2 different operating systems.

And comparing W10 hardware running W10 with W11 hardware running W11 would be comparing apples to oranges. Or at least a 2015 Honda Accord with a 2021 Honda Accord.

Another problem is the operating systems themselves and how they manage resources. Both operating systems, for example, perform various "housekeeping" chores in the background and when we humans are idle. These include indexing, security scans, checking for and installing updates, defragging hard drives or running TRIM and wear-leveling on SSDs.

How would one ensure resource utilization for background tasks and during idle times are exactly equal? I don't see how.

 

[phanbuey]

I don't really see how such a comparison can be done fairly. There are just too many variables at play here.

As others noted, hardware plays a part here. In fact, it plays a HUGE part. And the problem is hardware tends to be optimized to work best with the operating system that was current at the time the HW was designed.

Similar conditions apply to the drivers of such hardware. Yes, developers may be able to create W11 drivers for that 5 year old graphics card, but will it be as efficient as W10 drivers running in W10? I don't know - and because I don't know, how can I say it is a fair comparison? I can't.

My point is, I don't think it totally fair to compare W10 and W11 results when running both OSs on hardware designed for W10. Same applies if running both on W11 hardware.

I just don't see how any test computer can be configured so it will impose the exact same conditions on 2 different operating systems.

And comparing W10 hardware running W10 with W11 hardware running W11 would be comparing apples to oranges. Or at least a 2015 Honda Accord with a 2021 Honda Accord.

Another problem is the operating systems themselves and how they manage resources. Both operating systems, for example, perform various "housekeeping" chores in the background and when we humans are idle. These include indexing, security scans, checking for and installing updates, defragging hard drives or running TRIM and wear-leveling on SSDs.

How would one ensure resource utilization for background tasks and during idle times are exactly equal? I don't see how.

This. Also latency mon varies wildly with run to run variance for software and priority control -- For instance if I set it to realtime priority, I will get a different result than if I leave it at normal. If i watch a youtube video vs don't touch the machine etc.

Add to the variable priority and IO boosting that the OS does on foreground apps and its almost impossible to know why something is happening. Real world benchmarks (a good variety of them) are better in this regard. If you have latency issues on your foreground app, LatencyMon is good at ferreting out what the offending driver is, but it's not a reliable platform benchmark.

 

[Ruru]

I don't really see how such a comparison can be done fairly. There are just too many variables at play here.

Exactly. Though it's still interesting to see the differences between different hardware.

edit: should've read your entire post as you said practically the same

 

[Onasi]

Oh good, we are back to using LatMon to assess whether or not one system is “snappier” than the other version. In short - yes, 11, at least in its NOT heavily debloated form on average does have more latency than 10. Just like 10 had higher latency than 7 - more things going on in the OS will do that. It’s also hardware dependent, as already mentioned, a lot of people with NV cards see higher latency because of the NV driver. NV claimed to “fix” it, but it seems to be hit and miss, so the actual result will vary. It also has run to run variance because of background tasks.

But most importantly - it’s an irrelevant metric. It has its uses for its intended purpose of seeing if something messes with a PC meant for audio work, but it’s irrelevant for gaming. It’s not showing what you think it does. You want to prove actual tangible latency difference in game - get an LDAT or at least an NV Reflex Analyzer screen and mouse to actually measure end to end system latency. You probably won’t see anything meaningful there, though.

 

[grvn]

Oh good, we are back to using LatMon to assess whether or not one system is “snappier” than the other version. In short - yes, 11, at least in its NOT heavily debloated form on average does have more latency than 10. Just like 10 had higher latency than 7 - more things going on in the OS will do that. It’s also hardware dependent, as already mentioned, a lot of people with NV cards see higher latency because of the NV driver. NV claimed to “fix” it, but it seems to be hit and miss, so the actual result will vary. It also has run to run variance because of background tasks.

But most importantly - it’s an irrelevant metric. It has its uses for its intended purpose of seeing if something messes with a PC meant for audio work, but it’s irrelevant for gaming. It’s not showing what you think it does. You want to prove actual tangible latency difference in game - get an LDAT or at least an NV Reflex Analyzer screen and mouse to actually measure end to end system latency. You probably won’t see anything meaningful there, though.

For me is simple as showing that same config / same drivers / same settings / both Windows debloated to a certain degree provide inherently different results and Windows 10 just provides more consistent / smoother performance for me.

Besides the random frame-spikes I have on Windows 11 24h2, the alt-tabbing issues I also have on 24h2, Windows 10 doesn't have this or if it has is at a much lower degree.

How would one ensure resource utilization for background tasks and during idle times are exactly equal? I don't see how.

At this point I ready don't care because it's the same configs, same debloat so it's up the the OS. Not my fault Windows 11 for me... is just worse than Windows 10 on most aspects. Even desktop navigation is snappier on 10. Windows 11 seems... "heavy" with those animations etc but that's another discussion.

Which motherboard and do you have chipset drivers installed?

Aorus Elite X570 - latest chipset drivers.