if any driver , firmware level or os level depends on information on device that use bus , and is polled and slow frequency then i think it can lead to decreased responsiveness of system, either voltage or clock adjustments. Also , if this transaction is being handled as system management interrupt, there you have your answer for why we cant track this issue by latency/benchmarking tool, because os is not aware of that and thoose microseconds spend in smm is literally wasted cpu cycles that could be used to execute code.
Psu is not issue for sure, im kinda tired of repeating it every single post but i find very important to keep that in mind, unlike many new users posting mostly unrelated stuff, or things that has been tested like 5 years ago and except hyote or slender i can't see "old" users being active so cycle of placebo will repeat. To clarify what i have been talking about when i told i keep repeating this is i know exact same moment when this issue started for me back in 2011 while i was still playing cs 1.6 , basically i have grown on fps games like quake or whole goldsource engine games like cs , hl, tfc, or even tf2 and had identitcal setup from 2007 to 2011 until change has been made. During that era , it was golden times, every thing ran buttery smooth and responsive even if some fps drops occured. Actually this fps drops sometimes drove me mad during random pcw matches so i've tried to overclock my cpu and bricked my motherboard, at that time it was already old platform but i was broke teenage and my dad althrough he has been tech enthusiast he did not have money to buy me whole new platform from scratch, but got me new motherboard but it already needed new rams and new gpu (ddr400 to ddr2 and agp to pci-e switch) , psu, disks, CPU and even os (win xp service pack 3 ) remained same and well the first match i've played i instantly noticed there was something off, i knew has vsync lag is percieved , it's more like delayed mouse movement but still u can do perfectly crisp turns , your bullet spread is little bit slower because obviously you are having 75 fps instead of 101 at that time but despite that u just disable vsync, maybe set your pre-rendered frames to 1 and u didnt need to touch anything really, but it was different , too many things affected at once, the most noticable that drove me crazy is not even the input latency itself but how slow and "heavy" your characters feels, especially when you try to strafe there was literally time that my finger needed to spend on keyboard to move instead of action being instant, and even when you finally get your character to move it felt out of time, i even think that it take more time to stop, i watched my old demos i was collecting from matches when there was no problem and i could see a lot more agressive movement and less time i needed to strafe burst strafe burst shoot, i tried to imitate it on local lan server and it was obvious that time signature of whole system is off, i couldn't replicate that movement because my brain couldn't adjust to it, if i tried hard it looked like random strafe and random shoots and that's the most obvious sign of that you have this problem instead of random input lag caused by bad driver or mouse sensor it was obvious from first plays that this is something more serious. If system generates too much SMI event's it stalls your cpu for microseconds and also it stops tsc from counting when in smm mode. The rest of the story is like i went on my life , gaming was always my most loved loner hobby , i continued in school , i got a job and eventually in free time i was reading articles, i was trying different regedit stuff, whatever ,just trying to take a wider image. Then from money i earned i bought completly new amd platform, same there , and since then i have been on like 7 or 8 platforms both amd and intel, different motherboard brand always high or mid-high category, different ram kits, deticated sound cards, usb cards, network cards, with slight or no improvement, in 2013 i just stopped playing cs and started playing dota with friends from high school, still ocassionaly messing with os parameters, 5 years ago i was kinda into emi theory but back in my head i was sceptical because from 2006 to 2011 i wasnt event aware that something that serious even exists, and suddenly when i change the motherboard and other stuff i'm supposed to believe that some random electrical event occurs in my area affecting my devices? even when i moved to different location ? on every platform? why do people report (including me) that some random voltage value makes things noticable better or worse, why locking every value at fixed value , not using single auto value doesn't make any difference? It's like i want to dismantle whole system on parts , part by part , rams, imc, gpu, bus which gpu sits on, bits u can toggle and and can be found on platform specific documentations published by amd or intel directly, thoose are very specific and are not exposed in bios , not even in hidden part. I'm not even mad about this problem anymore even when i spend so much money on it, it's driven by both, curiosity and madness because paying for high end hardware and we're having terrible performance while we blaming game devs. So.... there you got your answer why i think it's motherboard problem, or to be specific motherboard firmware problem and if the architecture is not the issue you've gotta look on other stuff on motherboard which in this case is sensors, embeded controllers and acpi firmware and i'm more like embracing to focus on how power / thermal telemtry works than focusing on actual voltage settings. Altrough there can be some issue or more like mechanism implemented on gpu firmware that causes this ( actually lot of stuff dynamic power saving stuff
https://screenshot.cz/BYTLB/ ) it still doesn't explain why this happens on iGPU also. And psu? Yeah, i've already had many psu's.
Despite spi bus and smi theory, this things can be suspect too. but before some soc engineer dont take this seriously and start investigating we will never have answer. This is way too low-level .
PCI-E: chipset routing and switch arbitration: depends purely on motherboard, according to avaiable informations single switch can adds up to 60 to 300 ns latency in one transaction= request + process + response. Keep in mind that msi interrupts are pci-e based transactions too and on modern motherboards it doesn't matter if you turn msi off , modern motherboard don't have physical irq pins so they are emulated as legacy interrupts but still handled by pci-e.
PCIE-E: high completion timeouts + transaction reoders.
RAM/IMC: Command reordering, wrong ram training or wrong on-die termination causing jittery data eye, on-die termination was introducted in DDR2, DDR400 systems were using physical termination resistors on board.
The complexity of this problem is ridiculous, for me personally i completly switched from " i want most of the stuff handled by bios/firmware" to " i want most of the stuff handled by OS"
Sorry for bad grammar, too tired to double check this post
