Chief Blur Buster wrote: ↑02 Jun 2022, 14:23
Most People's Rare/EMI Problems Are Totally Different From Each Other, Despite Similarities
dervu wrote: ↑01 Jun 2022, 11:36
Also it is interesting if difference in power quality outside of PC would make any difference in error correction/latencies in PC with worst cable management setup + eventually worst EMI shielding on cables.
Unfortunately it's hard to prove this or otherwise, but it's quite possible -- one-off longshot type situations.
It's possible, but it almost feels harder to predict than quantum mechanics. You literally have to budget millions of possible EMI along the entire signal path of every single circuit wire (and every path on a chip, and every path inside a component, etc, etc, etc).
Think of this like a S/N budget -- any penny worth of EMI added to the mess can stack up to S/N noise floors weak enough to trigger error-correction storms that add latency. A slightly better cable routing (1 extra millimeter of separation between cables) versus a slightly worse ambient external EMI regime (e.g. 2% difference in strength), can turn this into a practical coin flip.
Importantly, one needs to understand
Signal to Noise Ratio (S/N).
Sometimes it behaves like a piggy bank. The S/N margin piggy bank can improve/worsen in real time along the path. A superior power supply can add more to the S/N piggy bank and a worse external EMI can subtract from the S/N margin piggy back, and so on. Because the superior filtering pushes the N down (less noise). Other times it's a great amplifier (inflate S without inflating N). But remember there can be millions (billions) of different piggy banks. A different cable routing can add more to the S/N piggy bank for one part of the computer, but subtract from the S/N piggy bank for a different part of the computer.
Enough bad stuff goes on, you've emptied the S/N piggy bank and the problems finally begin unexpectedly. Also, sometimes certain S/N piggybanks are withdraw-only (never possible to deposit additional margin again) -- it depends on the type of the interference and how the S/N can improve/worsen through successive filtering stages.
But remember in S/N, the S and the N are two independent variables. If your S drops, you also have to drop N to keep your S/N ratio constant. S can be amplified with amplifier stages, but N can also amplify too. And in some stages, S can only decrease while N can never decrease. So if you understand S/N (Signal-to-Noise), every single component (of the billions of components, including every transistor on a silicon chip) have their own separate S/N behavior. That's why overclocking sometimes fails well before the overheating stage -- overclocking reduces S/N ratio of certain transistors and when just ONE transistor fails, the computer crashes. Perhaps it's because a too-fast-switched transistor didn't have enough current through its gate, and the signal flowing through that one transistor was a low S going through the N ratio, and the S/N fell below spec, and so on.
Whether it's a 2 mile long DSL connection, a 1000 mile fiber connection, or a 100 micrometer path on a silicon chip, all of them have S/N ratios. The S is fighting against N. You can overwhelm massive S to nearly non-existent N (e.g. push half a million volts along a power transmission cable). But for chips, your S is super-weak, fighting against N of interference. And there's billions of super-weak S. Only one dominoe needs to fall! Just one. You don't have billions of re-amplifiers on a silicon chip because there's just no room. We made chips so fast and operating so close to the noisefloor, that it's less immune to inference (that's why we don't send powerful GPUs to outer space, and still rely on many old IBM RAD 600 chips -- there's a lot of radiation interference in space outside atmosphere). But likewise, down on Planet Earth, we have enough various kinds of radiation sources to cause problems to computers -- if you're accidentally near enough them, even if it's just a simple defective fridge/dryer/microwave oven running in the next room behind the wall inches away from the radiation-transparent side of your glass computer. Things like that happen.
But even smaller EMI sources like computer monitor power bricks can be a problem when placed too close to a non-metal part of a computer, etc. Or even the 500-watt wire connecting between power supply and GPU -- that can inject some serious EMP to do a bit flip on a SSD chip if the 500 watt wire is routed touching an M.2 SSD, for example. So, pay attention to power wire routing to your GPU and motherboard, don't route those specific wires touching your electronics -- add at least a centimeter or preferably an inch of air gap. That can be all you need (inverse square law is your best friend!)
The big problem is there's practically uncountable number of different S/N ratios in a whole computer (every single circuit path and transistor on every single silicon chip -- not just circuit boards or external wires), and it takes only ONE weak link to start a dominoe effect sometimes. The well designed modern parts will error-correct without crashing (and only give you error-correction latency of various kinds, depending on whether it's a one-off error correction or a complete blizzard of error-correction).
You can focus on the brute hammers (e.g. biggest potential EMI sources), that's where you will win the most in your battles against EMI.
But you can do nothing about those adjacent silicon paths inside your chip, can't reroute everything, not possible. A low-S/N-margin chip (not highly overclockable) could perhaps be more sensitive to external EMI than high S/N-margin chips (e.g. chips that are highly robust, stable, and highly overclockable), so it's also an incentive to add some margins throughout -- hit the low lying apples like buying the right parts (good power supply, good power routing for high-wattage wires inside and outside of the computer, chips that don't crash when overclocked by 5% and then keep them un-overclocked, etc, etc) and you might reduce your odds of random EMI problems by only 50%-75% because you can't control the other billions of S/N ratios. But you can attack the low lying apples specific to your situation -- but be reasonable, don't overdo it.
It's easy to mis-focus on trying to improve an apartment's external EMI regime when simply redoing cable routing may sometimes fix the issue, but it is also easy to mis-focus on trying to fix cable routing, when an intense external EMI regime will overwhelm even the best cable routing. It's easy to roll the wasted-troubleshoot-time dice badly in a blind gamble of troubleshooting EMI without a lab full of equipment.
For almost a five-sigma of population, it is usually a waste of time to spend more than a few hours on EMI mitigation measures (i.e. attacking the low lying apples), the rest is usually just attacking those likelihoods far to the right of the decimal point. And EMI problems are like snowflakes -- because of the infinite (>billions, >trillions) of different S/N weak links inside a computer, inside a circuit board, inside every single one of those chips, etc -- the only hope is to hit the low lying apples and statistically lower our likelihood of EMI-related issues. The "bright beacons of EMI like a 500 watt unshielded wire next to an unshielded data wire" versus those "tiny things that deplete the S/N piggy bank for one single transistor in the middle of a chip".
It's fun to ask to do tests and fun to speculate, but one has a greater appreciation of chip engineers doing a grand fight to S/N ratio that caused Intel to fail to clock faster than 5 GHz and caused GHz to stop increasing... If you didn't train HEAVILY in the laws of physics of S/N, then one can never hope to understand this post even partially.
Again, this is not a "Because it successfully worked for me, it will work for you" type of problem. No two EMI issues by users are identical, although the high-likelihood ones (the ones with more than 1% chance of being the culprit, and/or the outlier ultra-weak S/N dominoes like the LG 5K recall from WiFi-interference complaints) -- by different users might birthday-attack against each other and occasionally two users have the same EMI issue. But more often than not, it's completely separate unique EMI issues. So solution by user A will typically not easily fix issues by user B.
Most EMI work is pre-emptive, by the chip designers and circuit board manufacturers. S/N is a basic staple of electronics circuit-design teachings. It's often the weaker S/N dominoes that slip through, and those unexpected interference sources that tips them.