Data corruption from overclocking?

[Jumpem]

Is it enough of a threat to remain at stock speeds? Or to keep important files on a different PC.

 

[BTRY B 529th FA BN]

keep imp files on different pc

 

[Markfw]

I already replied in another thread, but that wasn;t the subject of the thread, so I will post here also.

Key word here....Stable, then NO, no corruption.
Second: Backups, all hardware can die and corrupt things, even not overclocked. I have Tape, Hard disk, and DVD backups. Triple backups for the most important data/pictures/video.

I don't even play FPS games, I just crunch for F@H....

 

[Jumpem]

Originally posted by: BTRY B 529th FA BN
keep imp files on different pc

I only have this main PC aisde from a cheap laptop.

I back up onto multiple hard drives already. The porblem is that if you corrupt files unkowingly and then backup over your previous backups. Now your backup is in the same bad state.

Looking at Anands 5870 scaling review, the difference between a stock i7 860 and one at 4.2GHz is only 5 fps or so.

 

[Idontcare]

Originally posted by: Jumpem
I back up onto multiple hard drives already. The porblem is that if you corrupt files unkowingly and then backup over your previous backups. Now your backup is in the same bad state.

There is never a shortage of threat vectors to your data. Data corruption is but one of them, yes the likelihood of it occurring can become elevated as your processor becomes hotter and overclocked.

But that doesn't mean the likelihood becomes near unity, nor have you done much to reduce the likelihood of losing your data simply by refusing to OC your cpu.

Fire, theft, lightening, acts of god, etc. In addition to that you could have any number of "walking wounded" hardware components that are silently causing data corruption - that one memory stick that suddenly (unknown to you) has an IC that is failing even at stock voltage and clocks, or that spindle-drive that is about to give you some clicks of death.

There is no such thing as absolute data security...all you can do is know what to be afraid of and implement counter-measures that reduce the probability of those events occurring.

For example I keep my system backup drive in my car, that way anytime I am traveling so too is my backup. Perchance my car is stolen I still have my originals in my house. Perchance my house is burglarized I still have my backup in the car. But if my house gets burglarized and near simultaneously an asteroid lands on my car then I'm screwed, and I won't have my data anymore as well.

I do not worry about silent data corruption, I make sure my system is small FFT prime95 stable for 24hrs, same as for large FFT and again same for Memtest86+. That's about the extent of my counter-measures against silent data corruption. More rigorous counter-measures can be implemented, but I haven't gone to the effort to pursue them.

 

[Jumpem]

Originally posted by: Idontcare
There is never a shortage of threat vectors to your data. Data corruption is but one of them, yes the likelihood of it occurring can become elevated as your processor becomes hotter and overclocked.

But that doesn't mean the likelihood becomes near unity, nor have you done much to reduce the likelihood of losing your data simply by refusing to OC your cpu.

Fire, theft, lightening, acts of god, etc. In addition to that you could have any number of "walking wounded" hardware components that are silently causing data corruption - that one memory stick that suddenly (unknown to you) has an IC that is failing even at stock voltage and clocks, or that spindle-drive that is about to give you some clicks of death.

There is no such thing as absolute data security...all you can do is know what to be afraid of and implement counter-measures that reduce the probability of those events occurring.

For example I keep my system backup drive in my car, that way anytime I am traveling so too is my backup. Perchance my car is stolen I still have my originals in my house. Perchance my house is burglarized I still have my backup in the car. But if my house gets burglarized and near simultaneously an asteroid lands on my car then I'm screwed, and I won't have my data anymore as well.

I do not worry about silent data corruption, I make sure my system is small FFT prime95 stable for 24hrs, same as for large FFT and again same for Memtest86+. That's about the extent of my counter-measures against silent data corruption. More rigorous counter-measures can be implemented, but I haven't gone to the effort to pursue them.

Indeed. I keep it duplicated on four drives. Two in my PC, and two that I connect externally only once a week when I sync to them. I'd like to get a safe deposit box to keep one of them in.

I use Beyond Compare to keep the four drives in sync. Perhaps I should start doind a binary comparison, in addition to file size and timestamp.

 

[Zenoth]

It's entirely possible, it happened to me recently, bad over-clock lead to corruption to certain OS files, some programs refused to either launch or uninstall just after the first reboot after I set the over-clock settings, while before that reboot everything worked fine. I had to re-install the OS entirely, never over-clocked again ever since, and everything works fine as before again.

 

[VirtualLarry]

Always do your overclocking/testing on a "test install" of your OS. After you get it stable, do a complete format and re-install.

 

[Idontcare]

Originally posted by: VirtualLarry
Always do your overclocking/testing on a "test install" of your OS. After you get it stable, do a complete format and re-install.

Heh, I remember long ago before we had canned stress test applications like Prime95 that the conventional wisdom was to be sure and install your OS with your rig at stock because if your rig was OC'ed and on the edge of being unstable then your OS would suffer silent data corruption and it would be fubar'ed forever (until you reinstalled it).

Whenever that was, Win95 era IIRC, successfully completing a clean install with an OC'ed rig was considered to be a stress test in and of itself.

Amazing how far we've come in the OC'ing community in terms of the ease of being able to OC a rig (hardware level, soft-bios and all that) and the ease with which we can execute stress tests. Moore's law in its own way is at play there too.

 

[Ben90]

Originally posted by: Idontcare

Originally posted by: Jumpem
I back up onto multiple hard drives already. The porblem is that if you corrupt files unkowingly and then backup over your previous backups. Now your backup is in the same bad state.

There is never a shortage of threat vectors to your data. Data corruption is but one of them, yes the likelihood of it occurring can become elevated as your processor becomes hotter and overclocked.

But that doesn't mean the likelihood becomes near unity, nor have you done much to reduce the likelihood of losing your data simply by refusing to OC your cpu.

Fire, theft, lightening, acts of god, etc. In addition to that you could have any number of "walking wounded" hardware components that are silently causing data corruption - that one memory stick that suddenly (unknown to you) has an IC that is failing even at stock voltage and clocks, or that spindle-drive that is about to give you some clicks of death.

There is no such thing as absolute data security...all you can do is know what to be afraid of and implement counter-measures that reduce the probability of those events occurring.

For example I keep my system backup drive in my car, that way anytime I am traveling so too is my backup. Perchance my car is stolen I still have my originals in my house. Perchance my house is burglarized I still have my backup in the car. But if my house gets burglarized and near simultaneously an asteroid lands on my car then I'm screwed, and I won't have my data anymore as well.

I do not worry about silent data corruption, I make sure my system is small FFT prime95 stable for 24hrs, same as for large FFT and again same for Memtest86+. That's about the extent of my counter-measures against silent data corruption. More rigorous counter-measures can be implemented, but I haven't gone to the effort to pursue them.

/agreed

I really think if your prime stable for long enough then the threat really isnt elevated that much. ive seen quite a handful of unstable ram at stock freq/timings/volts though in which case one could argue that an educated overclocker might be more protected against data corruption than someone who has never stress tested their system.

My x64 6000+ was <almost> unstable at stock clocks, but that was because the mobo couldnt even give the stock voltage to the chip... Thats a good case of how if my cpu was 1% crappier mixed with my horrible budget mobo; could really cause a lot of havoc on someone without their knowledge, and thinking they are safe because everything is set to auto

 

[Markfw]

Yes, I have had machines that were not stable at stock, so testing is always required.

 

[TantrumusMaximus]

Corruption is my #1 enemy when I try to get further ... I have the worst luck with corrupt drivers / os files etc after bad OCs.

 

[aigomorla]

correct me if im wrong, but doesnt ECC Ram fix this?

Thats why server memory has always used ECC even tho its latency was crap?

My Sammy uses ECC, never had i had a corruption on her yet.
And she's been on 24/7 for 2 yrs now.

 

[VirtualLarry]

Originally posted by: aigomorla
correct me if im wrong, but doesnt ECC Ram fix this?

Not totally. There can be data-corruption from the chipset, etc., that doesn't involve the RAM.


Interesting, from what I was reading in the video forum, the new Radeon 5xxx series cards use some sort of error-correcting protocol for the DRAM, so that when you overclock the RAM, and it gets unstable, instead of visibly artifacting, your performance instead slows down. Kind of like overclocking a CPU and having it overheat and thermal throttle.

 

[aigomorla]

Originally posted by: VirtualLarry

Originally posted by: aigomorla
correct me if im wrong, but doesnt ECC Ram fix this?

Not totally. There can be data-corruption from the chipset, etc., that doesn't involve the RAM.

LOL sorry if i sound like a noob larry...

But doesnt the chipset need to support ECC, so therefore it would fix the chipset corruption because the ram would report the values back as corrupted?

 

[Markfw]

Originally posted by: TantrumusMaximus
Corruption is my #1 enemy when I try to get further ... I have the worst luck with corrupt drivers / os files etc after bad OCs.

Install OS at stock. Then test stability@stock. If OK, proceed to OC slowly, checking stability. When done, normally there will be no corruption. If so, back it down to the last stable OC, and reinstall.

 

[VirtualLarry]

Originally posted by: aigomorla
LOL sorry if i sound like a noob larry...

But doesnt the chipset need to support ECC, so therefore it would fix the chipset corruption because the ram would report the values back as corrupted?

Ideally, every data-path, even internal ones, would be protected by a layer of ECC. But that isn't always the case.

Does anyone know if DMI supports ECC? I know QPI does, and can get by with a smaller number of lanes.

 

[BlahBlahYouToo]

Originally posted by: Idontcare

Originally posted by: Jumpem
I back up onto multiple hard drives already. The porblem is that if you corrupt files unkowingly and then backup over your previous backups. Now your backup is in the same bad state.

There is never a shortage of threat vectors to your data. Data corruption is but one of them, yes the likelihood of it occurring can become elevated as your processor becomes hotter and overclocked.

But that doesn't mean the likelihood becomes near unity, nor have you done much to reduce the likelihood of losing your data simply by refusing to OC your cpu.

Fire, theft, lightening, acts of god, etc. In addition to that you could have any number of "walking wounded" hardware components that are silently causing data corruption - that one memory stick that suddenly (unknown to you) has an IC that is failing even at stock voltage and clocks, or that spindle-drive that is about to give you some clicks of death.

There is no such thing as absolute data security...all you can do is know what to be afraid of and implement counter-measures that reduce the probability of those events occurring.

For example I keep my system backup drive in my car, that way anytime I am traveling so too is my backup. Perchance my car is stolen I still have my originals in my house. Perchance my house is burglarized I still have my backup in the car. But if my house gets burglarized and near simultaneously an asteroid lands on my car then I'm screwed, and I won't have my data anymore as well.

I do not worry about silent data corruption, I make sure my system is small FFT prime95 stable for 24hrs, same as for large FFT and again same for Memtest86+. That's about the extent of my counter-measures against silent data corruption. More rigorous counter-measures can be implemented, but I haven't gone to the effort to pursue them.

no live replication to a remote site?
weak...

 

[Nathelion]

This is the main reason I run server hardware at stock clocks. The stability is unparalleled. Since I installed Vista ~2 years ago I have had exactly one BSOD, and I'm pretty sure that was a driver issue. My hackbox that is also (unfortunately) running Vista seems to crash once a month.

 

[OVerLoRDI]

overclocking can cause data corruption, especially in the ram department.

That being said, if your files are mission critical I wouldn't do your work on an overclocked machine.

 

[Markfw]

Originally posted by: Nathelion
This is the main reason I run server hardware at stock clocks. The stability is unparalleled. Since I installed Vista ~2 years ago I have had exactly one BSOD, and I'm pretty sure that was a driver issue. My hackbox that is also (unfortunately) running Vista seems to crash once a month.

Well, my 12 boxes that are ALL overcloked run 24/7@100%load and I have not had a crash in years. I can;t remember the last time I did. That is what comes from proper setup and testing.

 

[TantrumusMaximus]

Originally posted by: Markfw900

Originally posted by: TantrumusMaximus
Corruption is my #1 enemy when I try to get further ... I have the worst luck with corrupt drivers / os files etc after bad OCs.

Install OS at stock. Then test stability@stock. If OK, proceed to OC slowly, checking stability. When done, normally there will be no corruption. If so, back it down to the last stable OC, and reinstall.

Always do. My issues usually are pushing the clocks and then get a BSOD that happens both in SafeMode and Normal Boots. Fine before the OC but then wham something just goes bad datawise.... and I say "why me?" LOL! Why can't I just leave it at 4GHz!

 

[MikeShunt]

isn?t this all down to the definition of "stable"

if your pc does a 48+ hour burn in test using the recognised utilities then surly its stable, overclocked or otherwise?

 

[eternalone]

I think its been said here before Never overclock on a working computer , always have a separate rig to do your important work on.