What causes driver corruption?

[Shrek]

If you are very, VERY unlucky a cosmic ray can cause a bit flip and corrupt a file, as it passes through your computer

How so? someone told me recently that there is one Cosmic ray per second per square inch at ground level.

Seems he was right on the mark
Cosmic ray - Wikipedia

 

[Bill_Bright]

How so?

Calenhad is right. Do you consider one per second per square inch a lot? I don't. Yes, it is a huge number but these are tiny particles.

Even considering there are billions of transiter gates per square inch in today's chips, there is still a lot of empty space between them a chip for that cosmic particle to slip by without hitting anything.

It is important to remember these are sub-atomic particles. That means they are pretty darn small. And just because an electron or proton manages to hit a gate, that does not mean it will cause it to "flip". It could just pass on through continuing on it merry way.

Seriously, how often do you see file corruption at all? Not very often. But even among those events, how many corrupt files would you suspect were actually caused by a particle in a cosmic ray hitting a gate and causing it to flip?

How many atoms would such a particle have to slip by before it actually hit a gate? It would come all the way across outer space, then through 100s of miles of atmosphere, the roof (which is thick shingles, tar paper, and plywood) of the home, ceiling of the room, the computer case, then the outer shell of the memory chip (and maybe a heatshield too).

To me, it is much MUCH more likely file corruption would be due to a power anomaly, an impurity in materials used to create the memory device, a physical flaw or damaged caused during manufacture, malware, ESD (perhaps from user abuse/mishandling) - and countless other things.

 

[Shrek]

I don't normally reply to you, but YES cosmic rays do cause errors, that is why we have ECC

What Is ECC RAM and Can You Use It for Gaming? (makeuseof.com)

"
For example, in 2003, a candidate in the Belgian elections received more votes than was possible. This error was only detected because of how the Belgian preferential voting system works. When the election officials manually recounted the ballots, they discovered one candidate had an extra 4,096 votes.

They found no errors after rechecking the entire system for any security breaches or foul play. However, the only reason they could come up with was that a cosmic ray hit one of the computers and caused a bit flip at the 2^13th transistor, thus adding an extra 4,096 votes to the total tally.
"


"You are entitled to your own opinion, but not your own facts. Do your homework. Google it. Verify your facts. Then post."

Hey, maybe you can get someone to shut-down this thread (as you have asked of me in the past) or send me another PM and then shut it down, so you don't allow a reply; that given, I still respect your opinion.... and... I could indeed be wrong about Cosmic rays...

Apologies to all (Bill included): just letting off a bit of steam; but it needed to be said.

 

[Tetras]

I don't normally reply to you, but YES cosmic rays do cause errors, that is why we have ECC

The studies I've seen are rather inconsistent (perhaps their age has a lot to do with that), but generally speaking, environmental errors appear to be less common than errors due to faults, but this seems to change dramatically with altitude (to the point that if you live at a high altitude, it is worth considering).

ECC is often chosen because you know when it is broken, more than because it can protect from an error, since invisibly corrupted has the potential to be so catastrophic. In the example you gave, this is quite rare to be so critically important, there's usually multiple layers of redundancy. They are used sometimes even on every day things (like checksums) and e.g. there are algorithms that can alert you to inconsistencies by tracking the data used at each point and by checking past patterns.

tldr: An error every now and then tends to be "meh" to the idea of "WOAH, the whole frickin' drive is corrupted!".

 

[Shrek]

Fair enough, but we are now down to 5nm, which is about 10 atoms across so it might become an issue in the future if not already; and the tracks left by cosmic particles are macroscopic in size.

I know, I know, it's not really 5nm, that's just marketing, so let's say 20nm and so 40 atoms across.
How Big is a Nanometer? (forbes.com)

 

[Tetras]

Fair enough, but we are now down to 2nm, which is about 20 atoms across so it might become an issue in the future if not already.

It's only one study so idk if anyone should take it super seriously, but in the google study they suggested (in the conclusions) that newer generations and higher densities weren't actually worse (in the sense of how many errors they were recording), perhaps due to improvements in manufacture. But, yeah, there's no guarantee that would hold true in the future, assuming it is correct now.

 

[Shrek]

Prove me wrong and I'd be very happy... all my machines are without ECC

One bit error in an image, who cares; one bit wrong in my bank account...

 

[Tetras]

Prove me wrong and I'd be very happy... all my machines are without ECC

The main thing I'd suggest to make you happier about that is that they said errors are correlated to utilisation (Ian said a similar thing, that in consumer systems they'll usually happen somewhere that is harmless, or if not, it should be recoverable). I doubt you're hammering your memory 24/7. The other is that when memory shows an error it is some order of magnitude more likely to show another one and the next and the next. So, if it isn't showing faults now, you're probably good, or at least, you're a lot better off than if it was. The creaking gate theory definitely doesn't seem to be true with memory errors.

 

[Shrek]

I know the squeaky wheel theory, but creaking gate is new to me... had to look it up.

I also like the cynical theory that ECC is to cover up bad RAM; a little paranoia goes a long way (I mean this in the positive sense)

I don't believe it, but still feel it important that we question motives.

 

[Tetras]

I know the squeaky wheel theory, but creaking gate is new to me...

I also like the cynical theory that ECC is to cover up bad RAM; a little paranoia goes a long way (I mean this in the positive sense)

I don't believe it, but still feel it important that we question motives.

I should have called it a saying instead of a theory, my bad.

Why thank you I should change my avatar to a pic of DDR5 with a red circle and a line through it.

With the creaking gate thing not being true, my take was how if you start getting zippers reporting corruption, games don't load properly, saves corrupted, that kind of thing, it is definitely time to investigate, cos it doesn't mean your memory will last longer. I do like the creaking gate saying in some computer things though, quirks just show you it is alive.

 

[Count von Schwalbe]

Another thing that could cause corruption - dependencies.

Ex.1 - your driver is updated but an expected change in a file it is dependent on does not occur (or vise versa)

Ex.2 - (this one happened to me) two drivers trying to use the same files/libraries and requiring different versions (a R7 450 OEM turned into a mobile GPU, according to the driver)

Ex.3 - user or other program messing with or moving files the driver is dependent on.

 

[Tetras]

Another thing that could cause corruption - dependencies.

Ex.1 - your driver is updated but an expected change in a file it is dependent on does not occur (or vise versa)

Ex.2 - (this one happened to me) two drivers trying to use the same files/libraries and requiring different versions (a R7 450 OEM turned into a mobile GPU, according to the driver)

Ex.3 - user or other program messing with or moving files the driver is dependent on.

Kind of the same / related, installing things (like drivers and patches) in the 'wrong' or an unexpected order.

 

[Zareek]

My PC (see specs) originally had 16GB DDR3 as a 4x4 arrangement. For years it was intermittently unstable, ie BSOD and other niggly errors. Really intermittent, so much so that troubleshooting just couldn't pinpoint it, including removing the memory modules. It often looked like software was doing it, but I didn't think so.

A couple of years ago, the instability started to get noticeably worse where it would BSOD every few minutes which allowed me to finally track it down to one of the memory modules with 100% certainty. I did notice that with 4 modules installed, they were all running hot, with the one closest to the CPU very hot, even though the PC isn't overclocked. I suspect that this heat was enough to make the module unstable and eventually to damage it, or maybe it was slightly faulty from new, I'll never know. I don't know if that faulty module was actually in that slot though.

I then replaced all the modules with new 2x8 Corsair modules of the same type and model range and put them in slots 1 & 3 so that the closest slot isn't occupied and the PC has run fine ever since - a blessed relief after all that time not running quite right. The two modules run a lot cooler, too. I've actually got another 2x8 modules used bought from a friend as part of a spare mobo / CPU / RAM / cooler bundle. I tried them and while it was great seeing 32GB in this old Sandy Bridge rig, the modules ran hot again, so I've removed them. I don't need 32GB and it's just not worth the headache. I'd have to add an extra fan there to fix the heat problem and that's not good for a PC built to be very quiet.

Yeah, bad RAM can be hard to diagnose, especially when it is something borderline. Memtest86 & Memtest86+ are handy tools for that but they can't simulate every condition that causes memory errors. I have personally seen random memory errors that don't show in testing but show up randomly and a different set of DIMMs fixes the issue. I don't have much fun like that anymore, that used to be work. Now everything I do is software related. It's not as much fun as hardware troubleshooting and repairs but it pays much better.

 

[qubit]

Yeah, bad RAM can be hard to diagnose, especially when it is something borderline. Memtest86 & Memtest86+ are handy tools for that but they can't simulate every condition that causes memory errors. I have personally seen random memory errors that don't show in testing but show up randomly and a different set of DIMMs fixes the issue. I don't have much fun like that anymore, that used to be work. Now everything I do is software related. It's not as much fun as hardware troubleshooting and repairs but it pays much better.

Oh yeah, it passed every sodding stress test going, including memtest. Weirdly, playing games made the PC more stable while just sitting on the desktop would blue screen like a bitch.

 

[Rock N Roll Rebel]

this video might help with driver problems