ChristophSmaul1337 wrote: ↑21 Jun 2024, 08:32
Hello once more.
As the title might suggest, I've found the culprit in my specific situation and have somewhat come to closure with this ominous problem. I'll leave the solution in my case right here so you don't have to read through the rest of this post:
My PSU died and has been replaced with a higher wattage unit.
My old 850W PSU failed and because of a great deal I've replaced it with a 1200W unit. No more problems in my specific situation since the replacement, no matter the circumstances. Keep on reading, because it's not as simple as it might seem at first and it doesn't make sense to me at all.
Please be aware that this isn't meant to be an universal fix for everyone, as I've seen people try this before and it didn't help them.
The Findings
Recently I've had a spectacular failure happen, which resulted in lots of "magic smoke" inside of my year-long standing PSU, a beQuiet Dark Power Pro 11 850W. My best guess is that a capacitor, possibly even the bulk cap, died, but I haven't opened it up and looked for damage yet. This thing has held up for a long time and I've gotten my money's worth from it by a mile. This is also exactly what should happen when a PSU fails, as it's a rather "simple" device: It either works and delivers the required voltages for the computer to operate, or it doesn't and the computer can't turn on. There is no in between, it can't be half-operational, just like a lightswitch can't be turned on halfway. Thank god this PSU didn't take any components with it while un-aliving itself.
Over the years I've obviously looked into the possibility of my PSU causing the weird problems mentioned in the first post. I've tested upwards of 5 different brands of PSUs, ranging from 450W to 850W. As a reminder, my current system consists of an Intel Core i9-10900k and a RTX 2080 Ti. While overclocked, the CPU draws an absolute maximum of ~250W (lower in anything that isn't prime95) and the GPU is a Gigabyte Aorus, which is limited to 300W, up from the usual 250W. These two add up to 550W at full load, and when factoring in the rest of the system (which is basically only a 64GB RAM kit, a 4TB NVME SSD and 2 CPU cooler fans as it's on an open air test bench all the time) the combined system load should never exceed 650W on paper. Measuring the actual power draw with a kill-a-watt shows about 400W when playing a highly demanding game like Battlefield V or Battlefield 2042. The idea of putting an 850W unit in this machine was only fueled by getting a good deal on this particular PSU, reducing its price to below what the corresponding 750W unit costed. If not for this deal I would've gone for the 750W part, as a 850W PSU seems very overkill on paper, and even more so when measuring real world power draw.
It never crossed my mind for even a second that this PSU could be the root of the problems. Not only is it rated for double the wattage that my PC pulls under even the heaviest gaming loads, it also is a quality product by a reputable brand, is 80+ Platinum rated and is placed in the highest respective tiers in PSU tier lists. And, for what I can conclude for now, this PSU per se wasn't even the cause. As I stated before, I've tried a bunch of other PSUs to rule out a possible defect and/or influence of this device on the problem. Every PSU behaved the same, including other 850W units from different manufacturers. The same behaviour was also observable on 750W, 650W and 500W PSUs, all showing the ever so present desync problems. This was why I originally ruled the PSU out of the equation. If every PSU does it, it can't be the PSU. That's how troubleshooting works after all.
After the beQuiet unit died, I needed to buy a new one. I've decided on a 1200W Corsair Hx1200, as I got this from a friend for a very cheap price. As with the beQuiet one, it's a high quality device with 80+ platinum certification from a reputable brand. Ever since installing this PSU all the desync problems have gone away, and never returned once. I've deliberately waited for about 2 weeks before posting, as I've read over and over again that problems creep back after a while for most people, which didn't happen for me.
This doesn't make any sense whatsoever. I cannot tell you why this works, and I've thought about this every day for the entire 2 weeks now. There is only one possible thing that I could remotely fathom here: Somehow, for whatever reason, everything up to 850W isn't enough wattage, despite theories and actual readings all showing the exact opposite. Even if we factor in the GPU transient load spikes, and assume that these are 2x the usual power draw, this results in a peak transient draw of 600W on the GPU with 250W of peak CPU power draw, still ending up right on the 850W mark that the PSU is certified for. And then, this transient wouldn't last for more than a few microseconds, and therefore wouldn't be a sustained load. When transients become a problem, it's because a PSUs over-current protection is set too aggressively and shutting the unit down because the maximum rated amperage was exceeded for the duration of the transient. This would result in a system crash, but not in instability, desync or other weird problems like horrible hitreg.
The PSU replacement came with an immediately obvious, significant smoothness increase. I could notice this even on the desktop, where the mouse cursor appeared outrageously smooth. Loading up CS2 I was absolutely blown away of how smooth this looks on a 360Hz screen. Enemies are slow, every gunfight is reasonable and deaths are always understandable. Suddenly, opponents with 10k hours become a non-factor and I'm winning the vast majority of head-on fights. All the other games I usually play have this effect, too: Battlefield 2042 is a notoriously horrible game, and the optimization is very, very poor. The game runs on a mere 120FPS on average, but after this change, even this feels very, very smooth and responsive (ignoring the horrible mouse input in that game for now). Running through the map and scoring 200+ kill games is now a thing of absolute ease, with nobody being able to shoot back at me.
This finding also comes with more implications. I was always of the opinion that software settings, and even BIOS settings, could not influence the problems regularly experienced here. When resetting the BIOS to stock, and freshly installing Windows, your game
must feel correct. If you do have problems with all-stock settings, it
has to be a hardware level defect. If this wasn't the case, everybody using Windows on all kind of hardware must also have problems, which they don't. And this all but confirms it, at least in my situation.
My previous post might contain false information, as the mentioned workaround isn't needed on this system at all. I've therefore deleted it, as I deem the information unhelpful. Right after installing Windows, even in the installation screen, the mouse already feels perfect and no tweaking - apart from disabling mouse acceleration of course - is needed. If this isn't the case for you, you likely have a hardware level defect somewhere. What I can now say for certain is that these problems - whatever may cause them - is something that comes from the outside and isn't related to settings or software.
You cannot break this again, too. No matter how many of the "best practices" I ignore, nothing will make the desync return. I can completely "Express" install the NVIDIA driver, without a care about GeForce Experience getting installed and running. I can enable the GFE overlay. I can disable XMP in the BIOS, even disable the CPU overclock entirely. I can disable turbo boost. I can use the Realtek ethernet chip, and I can also use the Realtek on-board audio with the Windows default driver, which were also suspected at one point. I can have multiple background programs running, I can download a game while playing. Heck, I can even play CS2 on maximum settings with artificial 40% packet loss set in pfSense. Nothing which is considered "bad practice" re-surfaces the desync and hitreg problems. Sure enough, most of these reduce the framerate by a decent margin, and if you're going for maximum performance this is actually bad to do, but it's not influencing the desync. I even went out of my way to try and break it, doing ridiculously stupid things like limiting the PCIe slot of my GPU to x4 bandwidth, while having the GPU on a riser card right next to the PSU for maximum interference. Not a single thing I did had an actual effect on synchronization, hitreg or mouse input. All these tweaks do is change the framerate, but as long as it doesn't drop below about 100FPS, the mouse input feels great. The only reliable way to break this again is to replace the new PSU with any of the other units I have laying around.
To address the elephant in the room: How does this make sense with my experiences at other places? And to be honest, I don't know. Maybe, if I knew why a significantly higher wattage PSU helped in the first place I could, but all of this is astounding to me. The 850W unit should be enouh, end of discussion. Theoretical limits and measurements show this for a fact. The only real speculation I have is a correlation with interference from either the mains directly or over the air with the PSU wattage, so you'd need a higher wattage PSU the more interference there is. That would explain why at my home, the 850W unit is causing issues while in the house 2 streets down it all is working flawlessly. But then, this goes straight against anything I know about electronics, PCs and how components draw power. The 850W beQuiet device is a PSU of highest quality, it should have ample filtering circuits to deal with virtually any kind of interference.
Testing
This finding sparked a flurry of tests that related to PSU wattage and component power draw, trying to get to the bottom of this as best as possible. I haven't found out the cause, but I can clearly see a pattern at least. For example, when doing component testing with lower power draw parts, it becomes apparent that any PSU I have laying around eventually becomes "good" and doesn't produce any problems. For example, even the 500W unit works perfectly fine when paired with an i5-4670 and a GTX 750 Ti. As soon as I swap the CPU for example back to my 10900k, the same problems on the 500W PSU return. Same happens when using a high power GPU, such as a GTX 1080 Ti, or my daily-driver 2080 Ti. Swapping in a high-wattage part "pushes" the threshold upwards, requiring a higher wattage PSU to make all the weirdness disappear again.
The really weird thing about this is that this threshold doesn't corellate with any numbers that would make any kind of sense. I want to give one more example here using an i5-4670 and a 1080 Ti, to make clear why this should absolutely not be working, but still somehow is. The two components combine for a power draw under full load of about 334 Watts (CPU: 84W, GPU: 250W). Leaving headroom for the RAM, fans and other stuff, consistent with the usual recommendations, would suggest that you would never need anything more than 500W at most for this setup. Using this setup with a 500W PSU though shows the usual atrocious problems with desync. Even a 650W isn't enough, as this still shows problems. Only a 750W unit and anything above can ensure flawless gameplay without any weird desync issues. To put this into numbers: Assuming the threshold is at exactly 750W, this would be another 67% increase over the usual recommended wattage, which in itself already factors in a 67% headroom over the theoretical maximum wattage of this system (334W).
As you can see, this doesn't make any sense and I cannot explain these results. I've appended the excel sheet with all of the tests I've run so far for everyone to take a look at. I don't plan on doing any more testing as the problem is resolved in my case.
I want to state with the highest emphasis that this isn't meant to be a universal fix for everybody. Please don't run out and buy a new PSU after reading my post, because it is likely an entirely different problem in your case. If you still decide to try it, make sure the place you're buying a new PSU from has a good return policy.
I want to thank the admins once again for allowing this forum to exist, all the users who contributed to this topic and I wish all of you good luck in finding your culprit, too. This thread can be marked as solved and closed. Thank you again.
Cheers
I think this should be posted on the main input lag forum too.