Let's disect this 1 by 1, I will highlight what I find problematic with
red
Lukas_ wrote: ↑12 May 2025, 13:42
besides cs2, the only things i have running are faceit ac, teamspeak, msi afterburner (telemetry is fully off, aware of the 9800x3d issue), and
logitech g hub.
Logitech G HUB is application SW with
severe CPU overhead (wastes CPU cycles & causes excessive context switching), which can contribute to macrostuttering (which is what you're describing instead, at least imo)
Use
Logitech Gaming Software (LGS) instead
Lukas_ wrote: ↑12 May 2025, 13:42
my setup:
cpu: 9800x3d
mobo: asus rog strix x870-e (gaming wifi)
ram: g.skill 2x48gb ddr5-6000 cl28-36-36-96 (f5-6000j2836f48gx2-tz5nr)
gpu: evga 3080 ti ftw3
cooler: noctua nh-d15 gen2 (best fit for am5)
psu: seasonic prime 1600w
mouse: logitech superlight 2 (1k polling rate to 8k hz polling rate makes no diffrence currently at 4k)
keyboard: ducky one 60%
display: zowie XL2546K 240 hz
bios settings:
expo: on expo 1
game mode: on (better cs2 performance for me, smt is off)
pbo: advanced, all cores -15
power limit: motherboard
rebar: on
all other drivers (bios, chipset, etc. — except gpu) are up to date.
temps are totally fine, i monitor everything.
system passes aida64 and cinebench without issues.
in heaven benchmark, frame times spike to maybe 20ms max, and there’s way less stuttering than in cs2.
my isp could be at higher down speeds but for gaming its very stable, no or very litter jitter
Cinebench
isn't a "stress testing" user application software to verify stability of a system.
AIDA64 does have a stability testing, where you have the option to choose what you want to stress (cache, FPU, system memory,...), but I haven't checked what algorithms it uses & what specific hardware faults it's good for.
Though, IMO, they are
benchmarking applications, first
I will quote @shrimps, who worded elegantly on what stability entails:
“No amount of memory stress testing is enough, stress testing is just an approximation of stability.
Learn patterns of your memory/imc and apply an overclock that not only passes tests, but also has headroom for when conditions change.”
Here are my recommendations, in terms of hardware:
- Your kit is not on the
QVL of your motherboard, meaning possibility of instability is
extremely possible when enabling EXPO.
- Considering that CS2 does not scale as much with lower R/W RAM latency or higher RAM bandwidth
(source), I would recommend that you experiment with
EXPO disabled and reinstall your OS (23H2 or 24H2) to iron out any potential corruption from a (very likely) unstable RAM profile.
- If you want to go down the RAM stress testing route, use app SWw such as:
RamTestPro (purple icon) for +16h (ideally with a GPU stress test such as Uniengine to simulate a more "realistic" workload, but you can try without; Furmark requires to keep cooling in check, as it's a bit of a "unrealistic" workload),
Karhu,
TM5
ycruncher,
OCCT,
HCI Memtest etc.
Linpack does not work on AMD AFAIK, as it relies on Intel specific libraries to do the linear algebra.
According to anecdotes I've read, Karhu & TM5
seem to miss easy errors on DDR5 as of the time of writing, but it depends on
what type of HW fault we're dealing with.
Every test has it's own strength's & weaknesses and it's good to keep them all as a "tool in a box"
What temps do you consider “fine”? This is very important to note.
You could also provide a ZenTimings screenshot to evaluate other settings of your motherboard.
I find
Game Mode to be redundant, you can
disable SMT instead of using
Game Mode.
Though, I wouldn't do it on AM4/AM5 CPUs, personally
PBO should be avoided at all cost, as it increases jitter from the variable clock rate and thus can incur microstuttering behavior.
Using a wireless mouse incurs
at minimum 1ms of added motion latency overhead, use wired instead.
You also induce severe RF intereference to your environment and possibly even negatively contribute to your overall health.
Higher polling rates exacerbate stuttering as well, due to increased interrupt count. Especially noticable if you're using stock (Microsoft) USB drivers, which are known for increased DPC/ISR overhead.
Lukas_ wrote: ↑12 May 2025, 13:42
os: windows 11 24h2 (latest updates) / game mode: on / hags: off / debloated with chris titus tech utility / power plan: performance
cs2 settings:
4:3, 1280x960, 239 hz / competitive settings / player contrast: on (yeah, it eats some fps but helps in darker scenes) / launch options: -console -high -allow_third_party_software -noreflex
fps_max 0 — currently capped at 400 fps via RTSS using this config: "cs2 subreddit found a fix for bad 1% and 0.1% lows"
(async cap stutters a bit less ms-spikes still present also worried about input lag)
fullscreen optimization off
nvidia driver: 544.16 (newer ones didn’t perform better for me)
settings: max performance,
low latency on, max refresh rate: highest available / scaling: gpu, override scaling of programs and games: checked
what i’m thinking:
something’s just off with the game or
my hardware combo is just unlucky
maybe something’s wrong with the gpu? don’t really think so, it’d probably behave differently if it were actually broken
ram timings might be too tight (cl28), but that’s stock for the kit — no manual oc, just expo 1
or maybe there’s some setting i missed that’s making the 9800x3d stutter — saw some posts about that too
Why are you running -noreflex and
Nvidia's Low Latency setting set to
"On"?
If you are GPU limited with your 3080ti, opt for Reflex (either forced through RTSS or in-game) or use the in-game frame rate limiter to achieve lower GPU usage.
As for using
RTSS Async &
RTSS Back-edge sync as a external frame limiter solution, bear in mind that it adds an additional
1 frame of latency (400fps = 2,5ms), while
RTSS Front-edge(best for frametimes) causes an additional
2 frames of added latency (5ms)
Why are you using FSO (used to be called eFSE) disabled?
Have you verified what
Presentation Mode your game is running in when you disable it, which is done by ticking the setting on CS2.exe's “Properties” window?
I would advise against doing so.
Considering that you are on a XL2546, refer to:
https://boringboredom.github.io/tools/fpscapcalculator
If you are using DyAc, it will exacerbate any macrostuttering or microstuttering you have due to the lower MPRT you're targetting.
Refer to
“How to use strobing competitively and beautifully”
I would also avoid using W11 drivers for your NIC, but I don't have any data on how Realtek's 5GbE NIC behaves.
I would also avoid using the CS2 Benchmark's statistics provided in the console. Use PresentMon-based solutions such as
Nvidia's FrameView,
PresentMon CLI or
CapFrameX,
Hope this helps.
) mircostutters still present
