a_c_r_e_a_l wrote: ↑08 Nov 2021, 17:42
Chief's post is interesting, but I've been asked many times by electricians, how would any interference go throught PSU filtering? They are like 100% sure it's not possible and digital signal is so difficult to interrupt.
<rant>
Not directed directly at you -- but this is a rant to old-fashioned analog-era-only-minded electricians who try to argue against the seriousness of certain digital-interference problems. Not directed at you, but frustrations directed to those analog-era electricians who say "big whoop".
Please ask the analog house-wiring electricians if they understand
all the possible methods that interference can happen (including over-the air, bypassing power filters). Power filters help but it's only a few stone-bricks in a castle fort protecting you.
Please ask the analog house-wiring electricians if they have any university training experience with digital electronics.
Please ask the analog house-wiring electricians if they are familiar with Claude Shannon's Law.
Please ask the analog house-wiring electricians if they have any familiarity with some of the EMI engineering that NVIDIA does.
Please ask the analog house-wiring electricians if they are familiar with
The Amazing Human Visible Feats of The Milliseconds, and how sometimes a millisecond cascades to human visible effects.
Please ask the analog house-wiring electricians if they are familiar with advanced digital error correction algorithms (forward error correction) running near the Shannon Limit such
LDPC or
Turbo Code or
Reed-Solomon, or others. This is done to transmit information faster with less power without needing an upgraded medium. Now, when a medium is packed to a theoretical limit of the maximum data per unit of bandwidth & unit of power (e.g. metaphorically like a nearly overflowing river that can't widen anymore, or a pipe about to burst, etc), it is super-sensitive to disturbances. When such data is disturbed by minor interference, then it requires retroactive error correction (e.g. retransmissions, rereads, etc). That incurs a latency overhead. So interruption = latency, instead of cutouts.
Please also tell them
I am in more than 20 peer reviewed research papers.
I will not remotely give the liberty of bothering to answer your electrican garbage advice until you confirm.
While digital signals are difficult to interrupt, they do cut out, much like a HDTV broadcast pixellating or a packet loss on a digital connection. When digital signals are operating close to the digital noise floor, having to use a combination of Reed-Solomon Error Correction Codes, or even Turbo Code error correction, or LDPC (Low Density Parity Check) error correction, the digital signals suddenly become several orders of magnitude easier to interrupt with the faintest possible EMI. One cable medium or a circuit path might see only a speed burp like 1 gigabits per second falling to 800 megabits per second (with temporary jitter) from minor interference, or whatnot. We don't notice because streaming merrily goes on, videos keeps playing, software keeps running, etc. But when in microscope, we're seeing micro-latencies from error correction. Maybe to THEIR OWN GODDAMN SEMANTICS, "a millisecond of lag is not an interruption at all", but that's a fat assumption (ASS out of U and ME) about how important a millisecond is to them, versus to us. So micdrop.
By "interrupt", the signal is not interrupted completely but simply error-corrected (creates lag = jitter). The first packet disappears, and a new packet has to be transmitted later.
Tell them modern USB is now packetized like an Internet backbone.
Tell them modern PCIe 4+ is now packetized like an Internet backbone.
Tell them modern SATA is now packetized like an Internet backbone.
Packet loss DO happen and it happens transparently; interference = error correction = simply retransmitted packets = latency.
Sorry, silly analog-era electricians that make assumptions of the unimportance of interference to digital, or that "inteference = loss of data", not understanding error correction. And the creation of latencies resulting from error correcting behaviours of computer components. Because the components are operating extremely close to Shannon's Theorem. This is happening because we've tried to push more data per second using the same amount of power as before, over the same copper wires, etc.
</numbertwo>
</toiletflush>
Thank you.
(I really mean it, unceremoniously, to self-righteous analog-educated people who don't quite understand the digital domain. They need to realize that EMI can go through multiple mediums instead of PSU filtering, as explained here. EMI is several orders of magnitude more complex than many electricians think. Mathematically. there's more possible theoretical simultaneous patterns/combinations of EMI than there are atoms in the universe, including over-the-air EMI that bypasses power filters (just LOOK, click the goddamned links, you old-era analog electricians). It is not necessarily a single frequency. Old-era analog-minded-only electricans need to intern as an NVIDIA circuit design engineer's assistant to be baseball-bat humbled by the major cascade effects of milliseconds worth of error corrections that cascades to stutters/latency/jitter/etc, and that different people are more sensitive to different things than others)
</rant>
