Page 2 of 2
Re: This guy is on another level - electrical test
Posted: 20 Feb 2023, 21:06
by Chief Blur Buster
wadge wrote: ↑20 Feb 2023, 11:30
I have another question, can a GPU induce error-correction events inside the PC ? and if yes, what would be a way to fix it ?
(asking because no matter how long i'm using my pc after it has been shut down for a while, i always get perfect hit reg/no desync but it quickly deterior after gpu is on load
)
Even without dirty electricity, the GPU has multiple behaviors that can affect its own performance:
- Thermal throttling (prevent overheating)
- GDDR6 error correction (repeat memory read events)
- Power management (automatically scale back power to save energy)
While most of it is unrelated to electricity quality -- it is possible that mediocre grid electricity can create PC power supply transients electricity, which may propagate onwards to GPU power (in other forms such as minor undervoltages, overvoltages, fluctuating voltages, or less able to provide sudden surge power, etc) -- that can cascade into worse memory performance (increased error correction events) or more overheating (as the onboard VRMs work harder to smooth out inferior PC power supply)
Testing a minor underclock while disabling power amangement can also stabilize things more. Yes, a smidge lower framerate, but possibly fewer stutters (frametime spikes = latency spikes). Use NVInspector to scale back power management a little bit too.
Re: This guy is on another level - electrical test
Posted: 22 Feb 2023, 07:29
by TheKelz
Chief Blur Buster wrote: ↑20 Feb 2023, 21:06
wadge wrote: ↑20 Feb 2023, 11:30
I have another question, can a GPU induce error-correction events inside the PC ? and if yes, what would be a way to fix it ?
(asking because no matter how long i'm using my pc after it has been shut down for a while, i always get perfect hit reg/no desync but it quickly deterior after gpu is on load
)
Even without dirty electricity, the GPU has multiple behaviors that can affect its own performance:
- Thermal throttling (prevent overheating)
- GDDR6 error correction (repeat memory read events)
- Power management (automatically scale back power to save energy)
While most of it is unrelated to electricity quality -- it is possible that mediocre grid electricity can create PC power supply transients electricity, which may propagate onwards to GPU power (in other forms such as minor undervoltages, overvoltages, fluctuating voltages, or less able to provide sudden surge power, etc) -- that can cascade into worse memory performance (increased error correction events) or more overheating (as the onboard VRMs work harder to smooth out inferior PC power supply)
Testing a minor underclock while disabling power amangement can also stabilize things more. Yes, a smidge lower framerate, but possibly fewer stutters (frametime spikes = latency spikes). Use NVInspector to scale back power management a little bit too.
This is interesting and probably what I have, since no matter what GPU I use, it's always stuttering weirdly, even on a freshly installed Windows with drivers only (downloaded from manufacturers site). Also, stutters happen on different builds and even an iMac that I have. Definitely some wonky electricity thing going on.
Re: This guy is on another level - electrical test
Posted: 22 Feb 2023, 21:19
by Chief Blur Buster
TheKelz wrote: ↑22 Feb 2023, 07:29
This is interesting and probably what I have, since no matter what GPU I use, it's always stuttering weirdly, even on a freshly installed Windows with drivers only (downloaded from manufacturers site). Also, stutters happen on different builds and even an iMac that I have. Definitely some wonky electricity thing going on.
I wouldn't jump to conclusion about electricity related stuff at this time, it's a dangerous assumption to make.
However, I never rule it out.
Troubleshooting is very tough when you're dealing with this. Game shader compilation events are also a common cause of stutters in many games, and that is hard to avoid.
Game texture streaming is also another issue but you can mitigate it partially by turning off your Windows swapdrive, using 32 gigabytes RAM (to compensate for lack of a Windows swapdrive), and use a very fast M.2 SSD
with a proper heatsink on it). SSDs have a nasty habit of thermal throttling without a heatsink.
This was one thing that Intel Optane was good for (minimizing texturestreaming stutter), so I'm a bit sad that Optane was never marketted correctly, since it helped a bunch of 0.1% worst-frametimes in some games. But you can just sidestep all of this by doubling up your RAM and disabling the PAGEFILE.SYS
Re: This guy is on another level - electrical test
Posted: 27 Feb 2023, 13:21
by dervu
I was wondering about something that could show some light on those issues people have regarding lag caused unknown source related to electricity.
It is not easy test. Maybe someone with enough experience with electronics and proper equipment like spectrum analyzer could test that on some random worthless laptop at affected place.
You can find on youtube that it is possible to measure that.
On every motherboard either in PC or laptop there is crystal oscillator.
Everyone talks either about interference or some other dark magic. Why not check most basic thing that whole PC relies on?
If there are no errors in any statistics that you can access, why not go with measuring crystal oscillator output to see if it is not affected? Many components use it as reference. If reference is bad, wouldn't that cause inconsistency in calculations resulting in lag/inconsinstent inputs etc?
End result of calculation is there, but it is not where we expect it in time.
Whether it is caused by interefence or ESD or ESD + interence wave coming from it or anything else, any disruption would be visible on oscillator output. From there, if disruption is visible in oscillator output, you can go through things that crystal oscillators are susceptible to and rule out things.
At least that would be some proof in form of graph, something to try to correlate with other measurements of suspected source of your issue.
Re: This guy is on another level - electrical test
Posted: 28 Feb 2023, 09:37
by wadge
dervu wrote: ↑27 Feb 2023, 13:21
I was wondering about something that could show some light on those issues people have regarding lag caused unknown source related to electricity.
It is not easy test. Maybe someone with enough experience with electronics and proper equipment like spectrum analyzer could test that on some random worthless laptop at affected place.
You can find on youtube that it is possible to measure that.
On every motherboard either in PC or laptop there is crystal oscillator.
Everyone talks either about interference or some other dark magic. Why not check most basic thing that whole PC relies on?
If there are no errors in any statistics that you can access, why not go with measuring crystal oscillator output to see if it is not affected? Many components use it as reference. If reference is bad, wouldn't that cause inconsistency in calculations resulting in lag/inconsinstent inputs etc?
End result of calculation is there, but it is not where we expect it in time.
Whether it is caused by interefence or ESD or ESD + interence wave coming from it or anything else, any disruption would be visible on oscillator output. From there, if disruption is visible in oscillator output, you can go through things that crystal oscillators are susceptible to and rule out things.
At least that would be some proof in form of graph, something to try to correlate with other measurements of suspected source of your issue.
Yeap exactly what i was thinking after all my research... maybe there would be something to do with crystal oscillator.
Like all the hardware has to work together in sync and maybe, (due to poor electricity quality or something else) with the time it is going out of sync and cause the issue we are facing.
I tried to force windows to sync with a ntp server stratum 1 every 5sec or so , seemed to help at first but issue is still here.
I thought maybe heat is affecting crystal oscillator but they are rated from -40 to + 85°c something you cant achieve with a good air flow in your pc case.
For me it is clearly something going out of sync but how to measure it.. i dont have the skill for that
Maybe Chief can help us to point in the right direction, i'm pretty sure we are missing a little part of the puzzle