Windows 10 Esports Quick Guide

[imprecise]

I decided to reinstall Windows 10 and test with pre-461 drivers. I did all the power settings from this guide and only installed nvidia drivers with no geforce experience. It looks like the lag is gone, who knows how long until it shows up again. I haven't done any registry tweaks either. I tested with nvidia sound through the monitor and realtek sound through optical, both working without lag. I can still tell the difference in buffer settings on the network card, but they only make a minimal impact now. I also turned XMP mode on and it seems to be working fine so far.

[imprecise]

The lag started creeping in again with XMP on, but I found how to manually configure the timings and it appears to be gone again with XMP enabled. When I had XMP off, the CPU wasn't running at the full 1600mhz FSB, only 1333mhz, which ran pretty smoothly but felt weird desync slowness, which makes sense if the FSB isn't running at the CPU's intended speed.

With the lag problem, pressing tab will cause a stutter, a flashbang will cause a big stutter, switching weapons while running caused a small stutter. After manually configuring the XMP settings, I was able to hold tab while running and with flashbangs popping without a single stutter. I could even start a spray in the wrong location and swipe it over smoothly to the target.

I used HWINFO to get the timing specs from the RAM, and used a decimal to hexadecimal converter to find the correct numbers to input in BIOS. It was still a bit of a guess since they aren't clearly labeled in BIOS, but it looks like I'm pretty close to optimal settings. I set voltage to 1.35v. I did have to manually configure it, because when just turning on XMP the timings were not set correctly.

xmp.png (14.24 KiB) Viewed 2998 times

[imprecise]

I have been doing more tests and can no longer recommend using GPU-Accelerated scheduling. I will update the guide soon.

[Shade7]

I have been doing more tests and can no longer recommend using GPU-Accelerated scheduling. I will update the guide soon.

I agree- Myself and a few others have noticed that it causes issues in Valorant.

Appreciate the effort you put into formatting the guide/images- keep up the goodwork

[Traveler]

.

[imprecise]

I think I've narrowed down the Win10+AMD CPU issue. Any app that sets to above normal priority has the potential to steal resources from your input responsiveness and/or your network responsiveness. I noticed this first with Discord, but there are several other apps that can interfere as well.

Open Task Manager, go to details tab, right click on dwm.exe and set to below normal priority. I'm already noticing flickers on my screen but it's worth it if I can actually play competitive games smoothly.

priority.png (15.61 KiB) Viewed 2541 times

Other apps that have caused this issue for me:
Epicwebhelper
Steamwebhelper
faceit
discord
firefox/chrome

Big props to Fluffy for this post:
viewtopic.php?f=5&t=4512

I'm still on MPO maxplanes 4 if anyone can assist in dropping it to 1.

[Shade7]

Interesting,

I'll have to look into this (I have a 5600x at the moment).

Why set DWM to "below normal" though. Why not just "normal".

Also, wouldn't introducing "flickers" be too much of a downside to running this tweak for a daily setup.

And, quoting RealNC's reply from that thread- DWM doesn't seem to be a concern for fullscreen games.

[timecard]

And, quoting RealNC's reply from that thread- DWM doesn't seem to be a concern for fullscreen games.

Yes and no, exclusive full screen games bypass DWM (GPU rendering) but DWM still is part of the CPU processing chain (Win10) as is CSRSS (handles raw keyboard/mouse input) and another reason why optimizing CSRSS priority (using process hacker) changes your experience.

If you use Windows Performance Analyzer, do a CPU specific recording you'll see DWM still uses CPU cycles, it's probably the top 2-3 process consuming time.

Windows is always processing cursor data primarily Windows Client/Server Runtime Subsystem (CSRSS/csrss.exe) even when it's not present on the screen, you can see this with this capture.

https://github.com/microsoft/busiotools ... aster/usb/
viewtopic.php?f=10&t=10642

The CPU indicated in these captures is actually the CPU CSRSS was running on and changing the affinity of CSRSS influences these results like, changing that along with USB controller interrupt affinity to reduce context switches. There's pro's and cons to this, pros desktop cursor probably feels great then you go ingame and now the CSRSS/USB adapter affinity doesn't match the affinity of the thread in-game causing more context switches instead of "averaging" out due to dynamic affinities/nature of multi-threaded processing.

[imprecise]

Interesting,

I'll have to look into this (I have a 5600x at the moment).

Why set DWM to "below normal" though. Why not just "normal".

Also, wouldn't introducing "flickers" be too much of a downside to running this tweak for a daily setup.

And, quoting RealNC's reply from that thread- DWM doesn't seem to be a concern for fullscreen games.

I think normal may work as well, I haven't tested yet. But games should be elevated to above normal/high if DWM is on normal, which could cause issues if the game doesn't run well outside of normal priority.

I will be updating the guide soon with process priority being a necessary step, as it is the cause of the major desync issues. I think Intel CPUs don't suffer the same problem due to difference in architecture or something else.

[Chief Blur Buster]

I will be updating the guide soon with process priority being a necessary step, as it is the cause of the major desync issues. I think Intel CPUs don't suffer the same problem due to difference in architecture or something else.

I would nuance and say "....it is one of the many possible causes of the major desync issues..."

There's a really giant optimization rabbit hole there.

As for AMD and Intel, there are AMD-specific tweaks that can help reduce the difference between AMD and Intel. I've seen really crappy Intel systems and spectacular AMD systems, so there's some overlap. There are some really crappy motherboards out there too. Whereupon installing their driver adds lots of latencies (DPC or other) on Intel.

Also, there are left-field optimizations. One of the Threadrippers has shockingly terrible inter-core latencies across distant parts of the dies. So disabling cores in the other half of the Threadripper package, (e.g. use only 8 cores instead of 16 or whatever) can suddenly lower inter-thread latency by a gigantic amount on an AMD CPU.

Likewise, Intel tweakers often turn off the efficiency cores of their brand new processor to keep only the performance cores, to prevent the problems associated with game engines thread-thrashing between the efficiency cores and performance cores.

Oh, and further down the rabbit hole, optimizing the clockspeed helps a lot too -- if you overclock a bit, you sometimes get too much thermal throttling or frequency-switching effects that make the CPU end up having worse latency than just avoiding an underclock. Heck, sometimes (once in a while), people have found better latencies in a very slight (~5%-ish or ~10%-ish) underclock, but that's very rare. I wouldn't call these general-purpose optimizations, but there's frequently system-specific optimizations that affects specific people.

There's Intel-specific optimizations and AMD-specific optimizations, both often very left-field and unexpected.

And an optimization that has big effect on one motherboard, does squat on another motherboard.

It's been VERY, VERY hard to have one general-purpose optimization guide.

Great start and I also want to keep the trollcount down as much as possible.

BTW -- I also used to write "eSports" until this 2017-2019 press change of using "esports" and "Esports". I've synchronized the topic title to the newer capitalization standard used in English-language venues and press. I also made your signature clickable, as fully-clickable URLs are normally not allowed in signatures anymore for forum members but an exception is made on an invitation-only basis.