[Ethernet Onboard-vs-USB-vs-PCIe] I finally found the reason behind my input lag after 6 years

Everything about latency. Tips, testing methods, mouse lag, display lag, game engine lag, network lag, whole input lag chain, VSYNC OFF vs VSYNC ON, and more! Input Lag Articles on Blur Busters.
User avatar
kyube
Posts: 126
Joined: 29 Jan 2018, 12:03

Re: [Ethernet Onboard-vs-USB-vs-PCIe] I finally found the reason behind my input lag after 6 years

Post by kyube » 01 Apr 2023, 10:07

novorei wrote:
31 Mar 2023, 11:32
Chief Blur Buster wrote:
30 Mar 2023, 17:19
novorei wrote:
29 Mar 2023, 16:11
What's the conclusion?
You can see the answer by reading the whole thread (click through the pages 1, 2, 3, 4, 5, 6, 7, 8, 9 and read almost a hundred posts):

TL;DR Conclusion: (A) It's a legit issue; there is a mix of successes and no-successes, on a per-person per-system basis. (B) It's hard to troubleshoot; (C) Far more common than some niche causes such as EMI;
But what are the tools that can measure some effect of it? Preferably, some tracking tool which has a server-client configuration with load processing, timestamping, external capturing of mouse/screen.

DPC latency (high and low, never really felt a difference), ping tracking (is compensated by game netcode), Capframex (only captures stuttering) does not directly translate to the issues of hitreg, heavy mouse, "playing" in past feeling.

Never found the silver bullet. Sometimes feels a bit better, sometimes not, other times nothing has changed and the feeling is different.

Are there "gaming certified" hardware and software platform?
First of all, placebo can be a hell of a drug.

Using a debloated Windows is a good start. Server 2022 is considered as barebones as it gets and it can allow for much granular control of the OS. LTSC Is very barebones too. I don't think W7 and W11 are good to switch to because the former is losing support day by day, while the latter has a few issues for now (I'd rather wait till W11 LTSC gets released) even though it has the +500hz Windows cap fix and supposed 'high polling rate' fix in the Insider build, which could definitely help alleviate issues.

Linux has the potential to be superior in regards to latency than all the above mentioned versions of Windows, sadly lacks developers to implement modern features and a more GUI approach to some features.

DPC latency has a noticable impact on overall system responsiveness and claiming 'never felt a difference' while you can 'feel' your cursor being 'heavy' is kinda ironic.

TCPOptimizer just adjusts registry of Windows, Leatrix fix does nothing.
Minimum rendered frames is a feature from the past and is called ULLM, which is superseeded by Reflex (much superior) and is only used to lower input lag at high GPU usages. Reflex (ease of setup) >= proper frame cap > ULLM / Anti-lag in terms of solutions for GPU-limited scenarios.

Aa few verified ways you can do for 'heavy' mouse feel:
Use xperf (analyzing DPC latency) and tools listed here: https://github.com/djdallmann/GamingPCS ... /README.md can be a a good start to hunt down your issues. (I personally use LatencyMon as well on top of xperf)

Using MouseTester (ver1.4 for +1kHz) to verify whether polling rate is stable is also another one.

Different sensors have different ways of tracking ('most responsive feeling' being the MLT04, 3366, 3360 SROM 5, 3399, 3950 due to the way their tracking works) I have on purposely excluded the Hero sensor (even though it has better circle motion handling), due to variable framerate which some people don't like. Perhaps consider switching out the mouse.

Different games have different ways of calculating mouse inputs. Diabotical is currently the only game I know of that has multithreaded mouse input, a feature that's desperately needed in modern games imo. It also features a very robust and stable in-game framecap, unlike any other game.

Perhaps your monitor just has slow response times?

RAM Tuning has a noticable impact on system responsiveness / cursor 'feel' as well.

Have you considered that humans are variable as well and not consistent? There's a ton of human factor at play (other than the above issues) from a physical and psychological aspect.

I don't want to get into the EMI aspect at all. It can be a culprit but I think there are a ton of other things at play before this has any play.

Oh, btw, what PCIE NIC have you used? I'd like to know if you're experiencing any DPC issues with it and if the drivers are good, been on a lookout for a NIC to replace my Atheros with.

neighbours cat
Posts: 3
Joined: 15 Jan 2023, 17:24

Re: [Ethernet Onboard-vs-USB-vs-PCIe] I finally found the reason behind my input lag after 6 years

Post by neighbours cat » 01 Apr 2023, 19:18

My ethernet is Realtek 8168H and after tweaking my system i realized the only latency is coming from my ethernet driver, so i closed all the offload settings etc didn't do anything. Yes disabletaskoffload helped with latency in online games but i've found that if u change "PRIORITY&VLAN" setting to 'VLAN' only the input lag is going away, im super fast in games right now even affects on desktop and everything
if u disable it you will gain weird acceleration, if u only use 'priority' the input lag remains

and 3 settings for low latency from me

Computer\HKEY_LOCAL_MACHINE\SOFTWARE\Policies\Microsoft\Windows\Psched

TimerResolution value "1" this will make your mouse much responsive

Computer\HKEY_LOCAL_MACHINE\SYSTEM\ControlSet001\Control\PriorityControl

Win32PrioritySeparation value "42"

Computer\HKEY_LOCAL_MACHINE\SYSTEM\ControlSet001\Control\Power\Profile\Events\{54533251-82be-4824-96c1-47b60b740d00}\{0DA965DC-8FCF-4c0b-8EFE-8DD5E7BC959A}\{7E01ADEF-81E6-4e1b-8075-56F373584694}

TimeLimitInSeconds default value is "6"

if u lower the value mouse will be much smoother, if u go up mouse will gain a little bit acceleration

you can try regedit settings while gaming, affects instantly



and from msi utilv3 im using realtek ethernet in msi mode with low priority and gpu with msi mode undefined
i can advice to not use your usb host controllers with high priority, adds input lag im using undefined.

naporitan
Posts: 96
Joined: 09 Jun 2021, 06:16

Re: [Ethernet Onboard-vs-USB-vs-PCIe] I finally found the reason behind my input lag after 6 years

Post by naporitan » 08 Apr 2023, 05:24

6yToFindTheAnswer wrote:
30 Oct 2022, 22:02
Hi there! To get rid of this problem you also need to have a usb controller which supports MSI-X. After changing the NIC to marvell aquantia the problem was solved by 50%. As a result I bought a motherboard specifically for low latency gaming and I won't have any more problems. With a usb controller without MSI-X it's like the dpi of the mouse is half of what it used to be.
If you are still stuck on z390, I recommend to buy a motherboard "msi z390 meg ace" with a network card from "KILLER" and I recommend to disable wi-fi in the bios.
Also the usb controller ASMEDIA.
eqwe.PNG
eqwe.PNG (886.64 KiB) Viewed 10483 times

Ragerlis
Posts: 32
Joined: 06 Sep 2020, 09:40

Re: [Ethernet Onboard-vs-USB-vs-PCIe] I finally found the reason behind my input lag after 6 years

Post by Ragerlis » 18 Apr 2023, 06:59

Just leaving this here.

https://www.overclock.net/threads/msi-m ... d.1801208/

There is intel documentation about it inside the post itself, more information about this particular effect.
Intel® Ethernet Connection 82579/I217/I218/I219 Network Adapter families and Receive Side Scaling (RSS)
The Intel® Ethernet Connection 82579, I217, I218, and I219 Network Adapter Families provide support for Receive Side Scaling (RSS) for up to 2 receive queues. This feature allows incoming network traffic to be processed in multiple queues and provides greater receive efficiency.
Because of the internal architecture of these Intel® Ethernet Network Connection families, they do not support Message Signaled Interrupts (MSI-X). Message Signaled Interrupts are required to allow separate receive queues to be processed by separate CPU cores.
The Intel® Ethernet Connection 82579, I217, I218, and I219 Network Adapters’ multiple receive queues are thus handled by a single CPU core. This prevents the use of these network connections in SMB-Multichannel configurations.
Other Intel® networking products, including the Intel® I210 Gigabit Network Connection, do support MSI-X and can be used in SMB-Multichannel configurations.
TLDR: i219V drivers comes with MSI mode on when the chip itself is not, fucking things up, turning it off worked for the OP.

cheerful
Posts: 3
Joined: 16 Oct 2021, 15:39

Re: [Ethernet Onboard-vs-USB-vs-PCIe] I finally found the reason behind my input lag after 6 years

Post by cheerful » 24 Apr 2023, 14:51

I have Gigabyte Z370, and Intel I219V-V, same issue. This is some crazy information, gotta get a new motherboard soon, mine seems to be cursed :D

If I play an online game, I don't use my LAN cable for years since I first saw that dramatic difference with Ethernet cable in n out, only 4g/LTE modem. Also, I have a problem with my superlight, It shows my frequency is 970-980 not 1000, which might be a reason why it is.

UPDATE: Got z390 UD (Realtek Gaming GbE), it's been a week already, guys this is insane. I don't have electric input lag as bad as it was at my old apartment and It got even better than it was on z370 with LTE internet. I feel dumb that after 3 apartments and 5 internet providers still blame the internet and not the motherboard. It's super enjoyable to play any game, I play on regular wired internet, which used to blame for delays. Hitreg is great, sometimes I shoot next to the head and it still counts in the head. About my old superlight - even on a new motherboard still is not 1000hz on testers. Tested 703 wireless and deathadder elite - both are stable 1000hz. Also bought new red superlight - also not stable 1000hz. I give up on that.
Last edited by cheerful on 05 May 2023, 03:45, edited 1 time in total.

User avatar
RamenRider
Posts: 56
Joined: 25 Jul 2018, 07:14

Re: [Ethernet Onboard-vs-USB-vs-PCIe] I finally found the reason behind my input lag after 6 years

Post by RamenRider » 01 May 2023, 22:35

I put on an Intel i340-T4 server NIC I had laying around and I can confirm it's slightly better, at least initially.

However I do not know if DisableTaskOffload on or off was better cause I did that as well. viewtopic.php?f=10&t=8993&p=71548&hilit ... oad#p71548

I did also enable Hardware Offload on my EdgeRouterX but didn't really see much of a difference. https://www.youtube.com/watch?v=jlPrRqri54o

Turning off XHCI off hand and Legacy felt like my mouse movement got better but honestly until I find a way to disable HPET on MOBO I still got lag.

ilyastroy
Posts: 4
Joined: 26 May 2023, 06:48

Re: [Ethernet Onboard-vs-USB-vs-PCIe] I finally found the reason behind my input lag after 6 years

Post by ilyastroy » 26 May 2023, 06:52

Нello everyone. I have a similar intel i219v chip . when disconnecting the Internet cable and disconnecting the intel adapter in the bios, my sensitivity instantly changes for the better. the model even moves faster. Mobo: ASRock B560M PRO4.
sorry I communicate with the help of a GPT translator

ilyastroy
Posts: 4
Joined: 26 May 2023, 06:48

Re: [Ethernet Onboard-vs-USB-vs-PCIe] I finally found the reason behind my input lag after 6 years

Post by ilyastroy » 26 May 2023, 11:52

I bought a realtek pcie adapter today. I will write soon how

User avatar
kyube
Posts: 126
Joined: 29 Jan 2018, 12:03

Re: [Ethernet Onboard-vs-USB-vs-PCIe] I finally found the reason behind my input lag after 6 years

Post by kyube » 26 May 2023, 15:22

I've got a I210-AT expansion card (Axagon PCEE-GIX, to be specific) and I must say, after enabling MSI-X mode and adjusting my network settings, it's substantially better.
LatencyMon also reveals this, ISR spam got reduced to 0, games feel much more fluid.

People should start exploring the DPC Latency realm (xperf, LatencyMon & using iperf to test DPC under load) and post results instead of touting how one adapter 'feels' superior over the other.

User avatar
kyube
Posts: 126
Joined: 29 Jan 2018, 12:03

Re: [Ethernet Onboard-vs-USB-vs-PCIe] I finally found the reason behind my input lag after 6 years

Post by kyube » 30 Aug 2023, 09:37

6yToFindTheAnswer wrote:
16 Nov 2022, 22:48
Yes, yes, yes, totally agree with you here. Found a friend with a msi Z390 having the i219-v chip, so i asked him (without any previous explanation) to unplug his ethernet cable and see if there is a difference in mouse feeling... guess what? Yep he was losing his mind, as i personally felt on my rig, he felt the same on his; huge difference in mouse movements :shock:

I bought a bunch of PCie to ethernet card and one USB-C to ethernet here are my findings :

2nd card :arrow: : chip : Realtek 8111H : Now this one is a very very good surprise, it was super stable and seems like it handles small packets very well making it a superb chip for competitive online gaming. I got 2ms less ping overall on every servers on it was a VERY stable card.


4th :arrow: is a USB-C to Ethernet : chip : REALTEK 8153 . I don't know what dark magic went into this one but as soon as i started testing my ping was 3ms lower on every server and it never, lowest amounts of jitter in download / uploads between all the cards. Never seen such a stable chip. My download speed was 2mbp/s slower than on the other cards but i will take it everyday for the stability it provides.

Which adapter and which driver did you use with your RTL8153?
I've recently got Lenovo USB Ethernet adapter with the HWID of USB\VID_17EF&PID_7205&REV_3000 on the 10.54.20.608 driver and It's very lackluster in terms of features and speed.
The adapter doesn't even do a gigabit, at least according to waveform's bufferbloat test (I've seen some slight 184mbps upload spikes, perhaps I've got a dud?)
The adapter doesn't seem to support any kind of interrupt moderation settings, let alone adjusting transmit/receive buffers, RSS etc.

If your adapter behaves similar to my experiences, I don't see a reason to recommend it at all.

The 8111H, however, does support interrupt moderation and is ~2015-16 era. I'm assuming it suffers from the RSS table issue the 8125B does.

UPDATE: 02/2024 - Check below for more information

In general, if anyone sees this, I'd recommend the following LAN chips (listed by speeds)

1GbE → Intel I210-T1, Intel I210-AT, I211-AT, I340, I350 (i217, i218, i219 should be fine too, depending on Windows version)

2.5GbE → 8125B (NOTE: don't get the PCIe adapter version of it, only use it if it's on-board.) (2.5GbE Intel can work too, check below)

10GbE → Aquantia AQC107, Intel X520, X540, X550, Intel X710-T2L, Mellanox or Solarflare NICs (such as the SFN8522, with "sub-microsecond TCP and UDP latency", requires Windows Server however or might not work with Windows entirely)
There are a few LGA1700 mobos which feature the Aquantia AQC113CS, no info on it as the date of writing.

Preferably stick to x1 and x4 PCI-E cards, as a x8 add-on card would gimp the GPU lanes.


UPDATE 02/2024: NOT AN ISSUE ANYMORE FOR 2.5GBE REALTEK, FIX FOUND BELOW
A issue this adapter, found by tharos:
RSS off, NIC assigned to core1. yet, ndis.sys DPCs stay on core0. only ISRs move to core1.
While RSS on, NIC not assigned will only spread the DPCs over the rssq cores while ISRs stay on core0
in addition RSS on realtek seems fundamentally ducked as it sometimes just breaks with reboot or nic restart


To adjust core affinities & RSS, you do the following: (thanks to Timecard, Amit, JerkoTAi, Duckling, calc and others)
For the 8125B & 8111H [steps for this are the same except driver] (fixing RSS and core affinities in particular):

Download this driver (for 8125B)
https://www.realtek.com/en/component/zo ... s-software (in particular, the Win10 Auto Installation Program (NDIS)
In windows, it's shown as
Driver Date: 11/21/2023
Driver Version: 10.69.1121.2023

- Change limit from MSIUtil from 1 to 4
- Gointerruptpolicy - set "Realtek PCIe GbE Family Controller" from "IrqAllCloseProcessor" to "IrqPolicySpreadMessagesAcrossAllprocessors" or "irqPolicySpecifiedProcessors"
- RSS on, which you do by going in Control Panel\Network and Internet\Network Connections\ and right-clicking the adapter, pressing Properties => Configure => Advanced and turning on "Receive Side Scaling"
- Queues to 2
- RSSBaseProcessor on 2 in registry (this in particular)

For the i225/i226, look at this

To change affinities and make them stick for other 1GbE Intel NICS (i217. i218, i219 in particular):
- Disable RSS if you're using older nic driver
- Set rssqueue to 1
- Set rssbaseprocessor to whatever core u want
UPDATE 02/2024 - Tried 8156b (USB-to-Ethernet adapter) using ASmedia controller
test scenario: bufferbloat test (probably overkill but was easiest to do, yes ik its tcp)
UPLOAD
Total = 25970 for module asmtxhci.sys
Elapsed Time, > 0 usecs AND <= 1 usecs, 24257, or 93.40%
Elapsed Time, > 1 usecs AND <= 2 usecs, 1666, or 6.42%
Elapsed Time, > 2 usecs AND <= 4 usecs, 26, or 0.10%
Elapsed Time, > 4 usecs AND <= 8 usecs, 7, or 0.03%
Elapsed Time, > 8 usecs AND <= 16 usecs, 12, or 0.05%
Elapsed Time, > 16 usecs AND <= 32 usecs, 2, or 0.01%
Total, 25970

Total = 199 for module ndis.sys
Elapsed Time, > 0 usecs AND <= 1 usecs, 176, or 88.44%
Elapsed Time, > 1 usecs AND <= 2 usecs, 1, or 0.50%
Elapsed Time, > 2 usecs AND <= 4 usecs, 6, or 3.02%
Elapsed Time, > 4 usecs AND <= 8 usecs, 0, or 0.00%
Elapsed Time, > 8 usecs AND <= 16 usecs, 0, or 0.00%
Elapsed Time, > 16 usecs AND <= 32 usecs, 5, or 2.51%
Elapsed Time, > 32 usecs AND <= 64 usecs, 8, or 4.02%
Elapsed Time, > 64 usecs AND <= 128 usecs, 3, or 1.51%
Total, 199

ISR
Total = 25970 for module asmtxhci.sys
Elapsed Time, > 0 usecs AND <= 1 usecs, 25887, or 99.68%
Elapsed Time, > 1 usecs AND <= 2 usecs, 59, or 0.23%
Elapsed Time, > 2 usecs AND <= 4 usecs, 13, or 0.05%
Elapsed Time, > 4 usecs AND <= 8 usecs, 10, or 0.04%
Elapsed Time, > 8 usecs AND <= 16 usecs, 1, or 0.00%
Total, 25970

DPC
Total = 1 for module asmthub3.sys
Elapsed Time, > 0 usecs AND <= 1 usecs, 0, or 0.00%
Elapsed Time, > 1 usecs AND <= 2 usecs, 1, or 100.00%
Total, 1
DOWNLOAD
DPC
Total = 57961 for module asmtxhci.sys
Elapsed Time, > 0 usecs AND <= 1 usecs, 57570, or 99.33%
Elapsed Time, > 1 usecs AND <= 2 usecs, 328, or 0.57%
Elapsed Time, > 2 usecs AND <= 4 usecs, 29, or 0.05%
Elapsed Time, > 4 usecs AND <= 8 usecs, 11, or 0.02%
Elapsed Time, > 8 usecs AND <= 16 usecs, 16, or 0.03%
Elapsed Time, > 16 usecs AND <= 32 usecs, 6, or 0.01%
Elapsed Time, > 32 usecs AND <= 64 usecs, 1, or 0.00%
Total, 57961

Total = 27 for module ndis.sys
Elapsed Time, > 0 usecs AND <= 1 usecs, 9, or 33.33%
Elapsed Time, > 1 usecs AND <= 2 usecs, 0, or 0.00%
Elapsed Time, > 2 usecs AND <= 4 usecs, 6, or 22.22%
Elapsed Time, > 4 usecs AND <= 8 usecs, 0, or 0.00%
Elapsed Time, > 8 usecs AND <= 16 usecs, 7, or 25.93%
Elapsed Time, > 16 usecs AND <= 32 usecs, 5, or 18.52%
Total,

ISR

Total = 57961 for module asmtxhci.sys
Elapsed Time, > 0 usecs AND <= 1 usecs, 57883, or 99.87%
Elapsed Time, > 1 usecs AND <= 2 usecs, 44, or 0.08%
Elapsed Time, > 2 usecs AND <= 4 usecs, 17, or 0.03%
Elapsed Time, > 4 usecs AND <= 8 usecs, 17, or 0.03%
Total,
Last edited by kyube on 10 Feb 2024, 10:57, edited 5 times in total.

Post Reply