Blur Busters Forums

Chief Blur Buster wrote: ↑

17 Aug 2021, 01:56

(image zoom link)

What games? You say it happens in every game in the topic line? And you're in single monitor mode, right?

Your graphs look really.... interesting. Maybe 4 or 5 different causes of superimposed frametime spikes. So this will be kind of an onion peeling task (of sorts).

So let's focus on this cyclic (frametime spiking by 50% at exact intervals)

weird-cyclics.png

Frametime spikes come from so MANY sources that it's kind of a whac-a-mole killing the sources, but I definitely see a cyclic problem I'd like to see some MSI Afterburner data for offline games to focus on fewer layers of the stutter onion.

-I was going to suggest Call of Duty inefficiency (e.g. texture loads or network) but then you say stutters in all games.
-I was going to suggest a malware scan, but then you say stutters in both Linux and WIndows, thru multiple reinstalls.
-I see your 16 GB RAM maxing out the moment the cyclics immediately begins, but then you say you have 32GB. Off the top of my head...
- I've seen defective power supplies (e.g. intermittent undervoltage sagging) rubegoldberg/dominoe into weird problems (such as cyclics). But you say you tried multiple power supplies?

So you've pre-emptively eliminated a lot of potential causes (if we're paying attention to just the cyclical portion)

I'd like to initially debug a non-network game. Can you confirm you were paying attention to the cyclical portion (always a stutter every exactly 1 second) regardless of game or OS?

Let's laser-focus on just only your cyclicals of your stutter onion.

1. Disable all your RGB software; some of them are poorly optimized (about 25% of them) and were traced to cyclic frametime spikes.
2. Did you ever try underclocking (10%-20%) of all your copmonents in case the cyclic is triggered by something overheating partway or some margins being pushed too aggressive?

This will not solve all frametime spikes, but attempting to co-relate the cyclics with other events occuring.

The stutter onion (that's what we call it -- the soup of bug stutters, textureload stutters, GPU frametime variance stutters, disk performance stutters, and so on, etc) is horrendously complex full of multiple causes of stutters, so dialing back to simpler offline games to start debugging the easiest layers first is possibly a good idea. There will be unsolvable layers (e.g. network stutter, game-specific stutter, etc) but we can work to peel back a few layers. There look like at least 1 or 2 solvable layers (even if we can't fix all).

There are certain subsystems that will intermittently throttle upon duress (overheatings, undervolts, overvolts, etc). The main chips may not be getting hot, but their power supply component might be.

We need to debug one layer at a time in the stutter onion (ignoring most of the stutter onion except the easiest-to-identify stutter patterns) -- and I choose the cyclic (easiest debug). You say you tried both Intel and AMD with similar issues, with both NVIDIA and AMD GPUs, with both Windows and Linux, with different PSUs -- same 1-second cyclic frametime spike issue, no matter what?

Chief Blur Buster wrote: ↑

17 Aug 2021, 08:18

Before we go on with some more tips, I need a super-important confirmation first. To confirm, when you are saying "frametime spikes", you're talking about the cyclic component (a perfectly regular stutter like exactly once every second or such). Right now, we're ignoring all other causes of frametime spikes such as disk access etc.

Nobody245 wrote: ↑

17 Aug 2021, 13:03

You mentioned trying different CPU's, Intel and AMD, that basically means you tried different computers altogether, since no motherboard can take Intel or AMD (as far as I'm aware). At that point, I don't know what else to suggest, maybe have a certified electrician test the outlets in the house?

Still, I'm going to throw an idea in here, from a slightly different perspective. Motion clarity isn't my field of expertise, but I've spent years working with hardware and providing technical support for hardware-related issues.

So, from the point of view of hardware, I would advise several things:

1. For best results, avoid mixing and matching hardware. Buy as many parts as possible from the same manufacturer, motherboard and GPU should be from the same manufacturer. AMD GPU should be used with AMD CPU and Intel CPU should be used with Nvidia GPU.

When you start mixing and matching, you're opening the door to all sorts of unpredictable issues. Manufacturers are supplied components by different suppliers and it could be something as low level as a difference in the quality of semiconductors of different parts that is causing issues.

2. Only buy parts from the top manufacturers, ASUS, MSI, EVGA and Corsair. Avoid Gigabyte, for some reason, whenever I heard or saw a hardware horror story, it was (almost) always Gigabyte... Also, their components are just, well, flimsy...

3. Only use 80 PLUS Platinum PSU from either Corsair or EVGA.

4. Regarding RAM, I always recommend Corsair, but I can't say anything bad about Kingston either.

So, for best results, your computer should look something like this: Intel CPU, ASUS motherboard and GPU(nvidia), Corsair RAM, PSU and AIO, Samsung SSD. Or, for example, AMD CPU with AMD GPU manufactured by MSI, on an MSI motherboard with EVGA PSU, Corsair RAM and AIO, and Western Digital SSD.

Now, this isn't necessary, but this will substantially reduce the probability of issues.

Avoid using extensions, plug directly into an outlet. if you absolutely have to, buy the highest quality extension possible and preferably only plug you computer into it.

Finally, slotting in different parts like you described isn't a proper way of diagnosing or troubleshooting anything. If you are down to attempting to identify the issue by trying different parts, you have to be systematic and methodical about it, otherwise there is just no point. When trying to find a combination to a lock, you have to start at 0.0.0.0. and then work from there, 0.0.0.1. 0.0.0.2., etc... otherwise, you're just relying on luck. You have to be systematic about it in order to narrow the issue down to the culprit (faulty part) by means of elimination of possibilities.

If you want to be paranoid about it:

1. When reinstalling OS, you should use software like Active@ KillDisk to erase the OS disk between installs.

2. During testing you should only have 1 disk installed, and it should be installed in accordance with the motherboard manufacturer's recommendations.

3. During testing, you should only install the low level drivers necessary for the computer to operate, nothing else.

4. Ideally, a CPU should only be installed once, on a new motherboard, after visually inspecting the socket, and never removed. It should be changed together with the motherboard. CPU sockets are extremely delicate things and unless it's absolutely necessary, a CPU should not be taken out/replaced.

Yuiiski wrote: ↑

17 Aug 2021, 19:55

1) I have done this, I posted here like a year ago that I had built a completely new computer with completely different parts yet the issue persisted, a year later I am still trying to fix it.

Chief Blur Buster wrote: ↑

18 Aug 2021, 16:21

Yuiiski wrote: ↑

17 Aug 2021, 19:55

1) I have done this, I posted here like a year ago that I had built a completely new computer with completely different parts yet the issue persisted, a year later I am still trying to fix it.

I'm sure you already changed this setting, but doublecheck this setting, and then I need more screenshots.



As a troubleshoot challenge:

1 -- Screenshot www.testufo.com/animation-time-graph with nothing on your computer running, no browser tabs except that one. Stutters can be caused by both hardware/software, but web browsers are unusually sensitive to background software, so this will help me determine if background software is contributing.

2 -- Screenshot the frametime graph of the simplest, most low-resolution, oldest game that still shows the cyclic stutters even during VSYNC ON. Perhaps Half Life 2 or some other very old offline source-engine game. Or even go further back to GLQuake. Do separate screenshots for the following sync technologies:
2a. VRR
...check monitor has VRR *enabled*, and make sure to use NVIDIA Control Panel VSYNC ON, and in-game settings VSYNC OFF -- opposing settings produce the best VRR fluidity
2b. VSYNC OFF
...check monitor has VRR disabled
2c. VSYNC ON
...check monitor has VRR disabled

All with the same game.

Four screenshot challenge. Ideally, no game newer than ~2007-ish.

The name of the game is to remove a lot of layers of the stutter onion (causes of stutter) and focus on things that still stutter even in these situations. I have a much higher chance of debugging these kinds of stutters -- then when these are fixed, things tend to be slightly smoother for newer games (even if they still stutter).

Thank you!