Re: Power/EMI - Is there a way to detect "retransmits" in PCI-Express, etc?

[TooManyPixels]

I'm posting this as a separate thread as it's a separate question that I didn't want to get lost in the noise of that otherwise excellent epic thread.

In response to the "engineer's update" from earlier this month, they made a mention of this:

We don't know what specific slowdowns were caused by -- flood of IRQ interrupt noise or error-correcting behaviours on SATA cables or noise on ECC DRAM buses or some other mysterious EMI-based slowdown effects, or a PCI Express bus trying to retransmit data during noise errors -- probably error correcting by the millions, like death by a million nanoseconds or microseconds that added up to continually visible slowdowns (game stutter, Window-dragging stutter, scrolling stutter, etc). Even when nothing was running in background (CPU/GPU near 0%), and scanners found no viruses.

Now, I believe that my computer is encountering the same issue - I've been experiencing unexplained USB-to-display input-lag for over a year now and I've exhausted every possibility I can think of - but this recent update to the thread made me wonder how I can have access to PCI Express bus retransmit stats - surely this must be exposed through some low-level hardware functionality such as at least a JTAG port, or through SMBus?

I took a look through Windows' Performance Counters in the PerfMon MMC snap-in but I didn't see anything relevant to PCI Express at all.

[Chief Blur Buster]

I'm posting this as a separate thread as it's a separate question that I didn't want to get lost in the noise of that otherwise excellent epic thread.

[Aside: Normally I like this separate thread idea, but EMI is currently too outlier/tinfoilhatty to many people (even though EMI is a genuine computer problem that can sometimes dominoe (sometimes in a "Rube Goldberg" / "Chaos Theory Butterfly" way) into creating input lag. So to prevent distracting users from more common lag problems -- I may merge this EMI topic into that thread until the EMI topic is more mainstream. But to keep it easy for you, I'll do my first reply here and make sure the thread has completed its destiny before I merge it into the EMIverse of that thread...]

Now, I believe that my computer is encountering the same issue - I've been experiencing unexplained USB-to-display input-lag for over a year now and I've exhausted every possibility I can think of - but this recent update to the thread made me wonder how I can have access to PCI Express bus retransmit stats - surely this must be exposed through some low-level hardware functionality such as at least a JTAG port, or through SMBus?

I took a look through Windows' Performance Counters in the PerfMon MMC snap-in but I didn't see anything relevant to PCI Express at all.

This generally is not Blur Busters speciality to do EMI testing; I leave that to the purview of other sites. However, if other users want to chime in (but only after the other readers read Post #1, Post #2, and Post #3 to educate themselves about EMI-derived latency issues), I welcome them to mention error-correction monitoring tools.

Some of the error correcting is on a per-peripheral basis; such as the custom error corrections that an NVIDIA GeForce card does nowadays.

<RedHerringWarningIndicator>
As a usual disclaimer, I must remind readers about about the wild-goose-chase red-herring factor. A specific latency problem is probably not the PCI Express retransmits or USB retransmits. But I will often be proven wrong here when people dump analysis data and show actual EMI-derived lag data; it's definitely happened somewhere else, to someone, at sometime in the past. For example, someone trying to troubeshoot EMI-derived problems often found out that the EMI problem affected something else (even something as simple as a poorly shielded mouse cable). Even mudane things like poll contention (e.g. 1000Hz USB device randomly slowing down another 1000Hz USB device), fixable via a $20 PCI Express USB card giving that specific peripherial a dedicated uncontended lane straight to the CPU.
</RedHerringWarningIndicator>

[TooManyPixels]

I did find this article that talks about PCI Express retransmissions and it explains how you can configure a Linux OS to get information about PCI Express retransmissions: http://billauer.co.il/blog/2011/07/pcie ... yer-error/ - I'll do some more reading on this and report back (if this post is merged back into the original thread by then I'll report it there).

I'll also see if there's a way to do this on Windows.

[TooManyPixels]

Also, I ordered an EMF detector from Amazon (just a cheap $30 thing: https://www.amazon.com/gp/product/B0754VVW4W ) and it arrived today. I used it around the house and it always read absolutely zero μT (so I assumed it was broken - surely there must be measureable EM fields everywhere?) but then I held it next to my computer case...

When held against the side where my Lian-Li's glass panel goes (currently the glass side is removed, I assume the glass wouldn't have any effect) it read up to 0.76μT.

When held against the outside of other side of the case (which has all-metal panels, including a metal panel between the motherboard and the PSU) which is closest to the PSU it shot up to 2.7μT - which surprised me because that's the still-shielded part of the case.

IMG_5950.jpg (194.03 KiB) Viewed 2601 times

IMG_5951.jpg (203.82 KiB) Viewed 2601 times

IMG_5953.jpg (210.29 KiB) Viewed 2601 times

So this makes me wonder if my PSU is the problem. It's an EVGA SuperNova 1200W PSU I originally got for my crypto mining rig a few years ago but I since moved it to my desktop - but it had been inside my desktop for at least a year before issues started appearing.

Unfortunately I don't have any spare PSUs lying around (at least not any rated high enough to drive my rig - I need a 700W or higher PSU, it seems).

[nuggify]

Because this issue is not due to electro-magnetic fields. I have the problem, and a much better EM field detector than you. Levels here are normal and not of concern. There are more than just EMF based power issues that can affect a PC. Your readings are not very high next to the PSU either.

[TooManyPixels]

Because this issue is not due to electro-magnetic fields. I have the problem, and a much better EM field detector than you. Levels here are normal and not of concern. There are more than just EMF based power issues that can affect a PC. Your readings are not very high next to the PSU either.

Yeah, I noticed that I can make the meter go-off just by waving a magnetized screwdriver over it - so much for that theory.

A friend of mine thinks it might be an issue with the VRMs on the motherboard.