Windows 10 DPC latency issue with ndis & tcpip.sys

I'm going crazy here.

I have tried contacting both realtek and msi in case they knew something which they didn't. Through MS support it got escalated to their tier 2. A guy had a 30min remote session with my machine, and couldn't find anything wrong at all. He even told me that it was very rare for him to remote a machine that felt so responsive for him, he was used to SFC /scannow taking up to 45min, but my machine did it in like 10min.

But the stuttering from the DPC issue continues. Clean installs have been made multiple times, system file checks, driver updates and downgrades, BIOS CPU settings disabling c-states, throttling, HPET on & off and more.

Yesterday i even installed a new network adapter hoping that would fix it, but nope. Still having DPC issues with ndis & tcpip.sys. The onboard network adapter is realtek, the new one is intel. So 2 different brands.

Searching threads like:

http://www.tenforums.com/drivers-hardware/28578-random-stuttering-dpc-latency-nightmare.html

http://answers.microsoft.com/en-us/windows/forum/windows_10-performance/very-high-dpc-latency-on-win10-in-ndissys/2c523e49-e2a0-45f0-8233-b6435dbbe905

http://answers.microsoft.com/en-us/insider/forum/insider_wintp-insider_perf/win10x64-dpc-latency-issue-ndissys-tcpipsys/1d49821b-7e21-4498-82e2-3d36926d3a3a

And many more yields no results, only people with the same problem and no solution, besides knowing that it is defiantly network related.

The only conclusion i can come to, is that there is a software issue in windows 10 with their network drivers. Their support does not appear to be aware of the problem. And from talking to MS support several times, i've learned they have no clue as to what, how or why.

Problem didn't exist in windows 7, for me at least. This is specific to windows 10. I've tried almost everything and it's driving me nuts.

 

Question Info


Last updated October 17, 2019 Views 7,082 Applies to:

* Please try a lower page number.

* Please enter only numbers.

* Please try a lower page number.

* Please enter only numbers.

Hi,

I would request you to check the link mentioned below as a reference:

DPC Latency USBport.sys
http://social.technet.microsoft.com/Forums/windows/en-US/4667aefd-5756-4ce2-9866-2bcb42668246/dpc-latency-usbportsys?forum=w8itproperf

Thank you.

Did this solve your problem?

Sorry this didn't help.

Great! Thanks for marking this as the answer.

How satisfied are you with this reply?

Thanks for your feedback, it helps us improve the site.

How satisfied are you with this response?

Thanks for your feedback.

Thanks for your answer. The thing about RST is interesting, but my c: is just an ssd so it doesn't apply to me. Besides that i'm not really getting much from that thread, the general stuff i've already tried. Not entirely sure where you were going with it.

But right now, ndis.sys just made my machine stutter with an execution time of 158ms.

Did this solve your problem?

Sorry this didn't help.

Great! Thanks for marking this as the answer.

How satisfied are you with this reply?

Thanks for your feedback, it helps us improve the site.

How satisfied are you with this response?

Thanks for your feedback.

This is of course another issue of the countless problems which are part of Windows 10. No one @ MS cares about it, there is of course again no fix for it at all. I have tried nearly everything that's possible except reinstall (which will not fix it). This occurs on two of my machines regardless what board or network card. It seems to be a bug in the OS and for me it's easy to replicate...as soon as there is enough load on the tcp/ip or ndis network driver something seems to break which results in a dpc latency above >50ms sometimes even 100 or 200ms.

There are tons of threads discussing this issue. But i never read anything helpful from the MS Staff, except the super commands DISM and SFC...but they will not fix this problem. I tried every available driver for all of my internal devices, I deactivated and reinstalled every single device on my machine, changed energy settings, fixed cpu clock, fixed speedstep, changed every bios / uefi setting. Replaced the Network card with a USB Dongle. Uninstalled sound driver, replaced every driver with the defaults from microsoft. Uninstalled every application that is somehow involved in the driver process...nothing. It always occurs in exactly the same way. Of course some setting like 100% CPU reduces the overall DPC and latency by 60us - 120us but that doesn't matter because the tcpip.sys and ndis.sys latency will cause a peak which is at least 10³ higher so that little change doesn't do any overall benefit, great!

For me it occurs regardless of the network card.

On Windows 7 everythings fine...It's exactly how you decribed it. This is a Windows 10 Issue and I wrote a simple C# application which will trigger this problem instantly...what does this application? It simply scans a network range e.g. 10.0.0.1 - 255 (multithreaded) thats enough to break the tcpip.sys....yeah nice one!

Oh and by the way on my Windows 7 machine nothing happens, no stutter no unusual DPC peak no extreme latency, I can run the application 50 times in 2seconds and nothing happens not a single stutter. On my Windows 10 machine 1-2 instances are enough to break the drivers...

I sugget that some MS technicians should be involved in the community process because reposting the same community generated s...stuff over and over again will not fix anything. Things that are clearly broken can't be fixed with solutions that are no solution at all...that's the thing that really annoys me because the moderators simply repost threads over and over again which are also not solved or unrelated...so the user is simply delegated until he finally gives up...is that serious??!?

16 people were helped by this reply

·

Did this solve your problem?

Sorry this didn't help.

Great! Thanks for marking this as the answer.

How satisfied are you with this reply?

Thanks for your feedback, it helps us improve the site.

How satisfied are you with this response?

Thanks for your feedback.

I installed win8.1 which works quite fine with classic shell. And i've been running that ever since with 0 issues. I have no reason to attempt win10 again before every game requires dx12, but i don't see that happening for another year. Maybe then things will be different.

But yes, the conclusion from MS support was "we don't know what is wrong, and we don't know how to fix it."

Did this solve your problem?

Sorry this didn't help.

Great! Thanks for marking this as the answer.

How satisfied are you with this reply?

Thanks for your feedback, it helps us improve the site.

How satisfied are you with this response?

Thanks for your feedback.

Hey Nicolaj

it's great to hear that at least Win 8.1 works fine concerning the dpc peak issue but sadly rolling back to a previous version is no option for me. It's time consuming to do this on my two machines which are already configured so I've to stick o/a find a fix (at least hope for one).

The real problem is that's so difficult to communicate a real issue with the support and get it to the devs because genereally it's the users fault. I'm pretty sure that a dev can directly investigate and find the issue with the information that i can provide. It's a common problem and I have an application which direclty and instantly triggers the problem to 100% on two completly different machines on the same build.

The users are having the same problem 100 times but the issue is not escalated to the next layer. Feedback Hub does in the current way not work quite well. It's a generation tool of useless content. Technical detailed description are ignored because there are so many useless tickets which only describe a problem in 10 words.

MS must find a better way to report bugs, srsly.

Did this solve your problem?

Sorry this didn't help.

Great! Thanks for marking this as the answer.

How satisfied are you with this reply?

Thanks for your feedback, it helps us improve the site.

How satisfied are you with this response?

Thanks for your feedback.

That actually surprised me a bit. I thought that they'd collect information on the problem to escalate it. Because now their support had encountered an issue they didn't know about, nor could they solve it. But they didn't. So i'm more or less completely sure that this isn't an issue that's being worked on.

Did this solve your problem?

Sorry this didn't help.

Great! Thanks for marking this as the answer.

How satisfied are you with this reply?

Thanks for your feedback, it helps us improve the site.

How satisfied are you with this response?

Thanks for your feedback.

After a little more investigation I'm pretty sure this is a bug, I don't know when they introduced it but I also asked a friend to replicate the bug with my tool and indeed it also occurs on a fourth unique machine with the latest Windows 10 build.

It was tested with LatencyMon and he also gets a DPC Peak above 70ms for tcpip.sys but he has a quite powerful new machine. Its very difficult for the user because there is no way to see if there is already an open ticket in the development process which is linked with an actual issue. So the users are completely left alone.

There is no way of interaction on a problem, no real responses, no information. Every 1 man GitHub project works better...so the next build will possibly only be fancy again but no real world fixes, i'm very disappointed

Did this solve your problem?

Sorry this didn't help.

Great! Thanks for marking this as the answer.

How satisfied are you with this reply?

Thanks for your feedback, it helps us improve the site.

How satisfied are you with this response?

Thanks for your feedback.

thexyz, could you share the source code of your program? I have written one like you described, but it does not trigger the problem.

Did this solve your problem?

Sorry this didn't help.

Great! Thanks for marking this as the answer.

How satisfied are you with this reply?

Thanks for your feedback, it helps us improve the site.

How satisfied are you with this response?

Thanks for your feedback.

Sure ;), here is the C# class. You have to change the base ip to your local subnet...credits are not on my side, I took most of the code from stackoverflow because it's linked to an application were i needed it. Only slightly modified. But this triggers the problem on four different devices that I tested!

Code: http://pastebin.com/VUrVASMh


One instance triggers an abnormal peak on my side 2-3 instance let it escalate to around 80-200ms. After that more instances would not significantly add more dpc latency. But you can compile a debug exe and run it 5 times in a row and you are on the safe side to trigger the issue ;)

PS.: I forgot that there is the Bag Collection with the corresponding Host object simply remove that stuff or create a dummy it will work in both cases

Credits for the C# Snippet: Tim Coker @ Stackoverflow

Did this solve your problem?

Sorry this didn't help.

Great! Thanks for marking this as the answer.

How satisfied are you with this reply?

Thanks for your feedback, it helps us improve the site.

How satisfied are you with this response?

Thanks for your feedback.

I’m not sure, but it is strongly recommended to remove events and dispose disposables before exit. But it does not help a lot. I tried.

This code infinitely pings 300 random hosts.

http://pastebin.com/in26fJKD

I can run it forever, I can stop it whenever I want, and I can start and stop it many times.

But if I make only 254 loops and exit (after cleanup and extra sleep) several times in a row, bad things happen. I’ll try to find out why.

Did this solve your problem?

Sorry this didn't help.

Great! Thanks for marking this as the answer.

How satisfied are you with this reply?

Thanks for your feedback, it helps us improve the site.

How satisfied are you with this response?

Thanks for your feedback.

* Please try a lower page number.

* Please enter only numbers.

* Please try a lower page number.

* Please enter only numbers.