Microsoft: Don't overclock Windows 8 unless you like our new BSOD

Microsoft's Blue Screen of Death is the curse of Windows. Not just its appearance, mid way through some serious game play or spreadsheeting, but the messages themselves - digital monologues on the existence of a problem, its possible causes and how you can fix it. Yeah, right. Just shut down and re-start like everybody else. …

[Anonymous Coward]

Sooo...

...the bit where it says overclocking make your machine unstable turn out to be true after all. And major OEM who often right their own driver sets, or at least test them 1st before release are more stable than a minor one, who is likely to slap in different bits and bobs as they arrive.

Wow, who'd of thunked it.

However it's still a risk vs reward option. Pay more for less performance, or take the risk of it crashing. Your call.

[Al Jones]

Re: Sooo...

You won't be paying much more, if anything at all, unless you are building a monster system, or you want a really nice case. For a middle of the road system, Dell and HP can buy the parts and put them together cheaper than you can.

(Having said that, HP wants $120 to go from 4GB of RAM to 8GB. When I can get 2x4GB Crucial DDR3 for $42, someone is being robbed. And the price hike to choose a beefier processor is substantially greater than the retail difference too).

[Nater]

Re: Sooo...

Probably has more to do with people not knowing what they're doing and all the new fangled "overclock it for me" utilities. I've never used one, but I'd guess such utilities are heavy handed with the vcore. I've been running a Q6600 overclocked to 3.6GHz for the last five years without issues. The only BSODs were due to RAM issues, which as we all know, are all too common these days. In those five years I've probably been through three sets of DIMMs because one or both in a set failed. The memory, for what it's worth, is running at stock speeds and timings. I don't see a reason to mess with it.

Time to replace the old beast, though. I'd have never thought I'd have used it so long back when I built it.

[JEDIDIAH]

Re: Sooo...

Nonsense.

This is just an attempt to discourage consumers from buying from companies that aren't under Microsoft's thumb. White box vendor means someone that's not afraid of losing bulk discounts on Windows licenses.

Overclocking is an entirely separate issue.

[kororas]

This is what Prime95 and that ilk is for.

[Parax]

Prime95 for CPU but also Memtest for ram stability and both for at least 24hrs.

Most clockers I know, know what they are doing..

But that said most BSOD are from bad memory or bad settings (ie clocked badly).

I have never found a BSOD'ing machine that has not thrown a million errors in memtest.

[Danny 14]

too many variables

Depends on the components! Dodgy rubbish mobo and psu + overclock = more crashes. Unlocking 2 extra cores on a dual core = more crashes. Mild overclock on a known good overclocking component with good PSU and mobo = probably ok.

[mafoo]

RAM Timings

I've found that the biggest cause of crashes on non OEM PCs is the RAM timings not being set in the bios correctly. The bios often resets them when you upgrade it and then phantom crashes start occurring. Its not something the average user would think about.

[Anonymous Coward 15]

Re: RAM Timings

Shouldn't those come from the SPD on the module and be known good?

[Anonymous Coward]

Re: RAM Timings

The SPD usually has default timings but overclockers like to push the envelope with more aggressive settings.

In fact, most high end memory is by definition "overclocked" and its SPD settings will in certain cases force it to perform at lower settings than rated.

[Wibble]

So, to summarise...

Errm, so they're saying don't use Windows?

[Voland's right hand]

Re: So, to summarise...

No.

The summary is - always use CPU frequency scaling and have a cooling system which has feedback control of some sorts - either 4 pin fans or 3 pin fans with integrated thermal sensors and rev control. Make sure you do not have hot pockets, etc too.

This would explain the rather strange laptop stats. The average laptop cooling system sucks bricks sidewize through a thin straw compared to a desktop. However, all of it is controlled by the OS (via acpi or whatever other interface is available) and cranked up to match the heat output. In addition to that air is taken from outside and dumped to the outside. There is no internal recirculation.

Out of all "other" reasons this is the most likely reason for "white box sucketh" results too. Most whitebox manufacturers do not have the resources to spend on analyzing and fixing airflow in their systems so they end up with hotspots here and there. Otherwise the parts which they use are not that different from "big labels".

By the way - the summary is totally valid for Linux too. If you want it stable - ensure that your cooling system operates properly and is matched by appropriate controls in the OS - lmsensors, fan control or the odd script which starts limiting the CPU frequency if the temperature crosses a particular threshold.

[Anonymous Coward 15]

Re: Make sure you do not have hot pockets

No, have ramen noodles instead.

[Blitterbug]

Re: So, to summarise...

No indeed. And also, the spurious 'Windows 8' in the article headline is misleading and rather naughty, as these are generic rules across the entire installed OS userbase.

Posted from a happily O/Cd i5 Win 7 lashup.

[h4rm0ny]

Re: So, to summarise...

"Errm, so they're saying don't use Windows?"

Yes. Well done. That is exactly the message you should take from this. Because when my CPU overheats or my RAM can't keep up with the memory timings I have set for it, it doesn't phase my Linux box in the slightest. Non-MS operating systems don't actually need reliable hardware to run. In fact, the processor is really only there for looks with them.

[Fatman]

Re: "Errm, so they're saying don't use Windows?"

It looks like I was not the only one who took:

According to Microsoft, the longer the TACT - the longer a PC runs continuously, without shutting down - the more likely it is to experience its first crash thanks to a CPU failure.

to imply that Windows systems have sever uptime limits compared to Linux systems.

I always knew that WindlowZE was a shitty O/S, and this now convinces me of that observation.

[Oninoshiko]

Re: "Errm, so they're saying don't use Windows?"

"It looks like I was not the only one who took:

According to Microsoft, the longer the TACT - the longer a PC runs continuously, without shutting down - the more likely it is to experience its first crash thanks to a CPU failure.

to imply that Windows systems have sever uptime limits compared to Linux systems.

I always knew that WindlowZE was a shitty O/S, and this now convinces me of that observation"

No, you are not the only illiterate moron here.

[Anonymous Coward]

Re: So, to summarise...

Or the other way I can look at it is this another peice of propaganda from the church of MS.

[John G Imrie]

Micosoft say

By from those OEM's who pay us for bulk licences or face the consequences.

Or am I reading this wrong.

[Ian McNee]

"Quality" OEMs

And as anyone who fixes PCs knows, we *never* have nightmares with shed-loads of Acer laptops and Dell desktops built to the headline price set by the marketing department to sell shed-loads of boxes to unwary buyers. As if.

</sarky-fucker>

[Anonymous Coward]

Re: "Quality" OEMs

The word "quality" is likewise not something I would associate with my 5 1/2 year old HP work laptop's HW.

Note: I do have a replacement laptop (Dell) but my employer's admins seem incapable of imaging it correctly, and it takes me about 1.5 days to re-install all my open source utilities each time, making me more tolerant of the old HP pos.

I agree, MS is publishing interesting stats here, so let's not crap on them for doing so, "quality OEMs" aside.

What intrigued me was the increased likelihood of a first crash after long TACT and increased subsequent crashes probabilities after the first.

- heat buildup may be driving this and turning a computer off can't hurt.

- once you've had your first BSOD you may either have damaged your CPU/RAM. Or you have been issued a lemon in the first place and it will continue in that direction. Or your overclocking is incompetent and will remain so until fixed. In any case => more BSODs to be expected compared to a BSOD virgin.

[streaky]

Re: Micosoft say

You're reading it wrong, but sort of not.

Also it's just overclocked pcs crash more than not overclocked pcs, which is like saying people without ebola bleed less than people with ebola.

10/10 for stating the obvious and all that.

[Tom 15]

Errm...

What's this got to do with Windows 8? It's a study into all existing Windows PCs. And yes, most BSODs are caused by driver problems so naturally you'll get less problems with mass marketed, expensive, branded products than cheap own-brand stuff.

[Anonymous Coward]

Re: Errm...

Or poor error trapping in the OS

[streaky]

Re: Errm...

"Or poor error trapping in the OS"

You need to go back to writing kernel school for that one I'm afraid. Same thing that causes a BSOD in windows causes a kernel panic in Linux and that stupid error screen in OSX.

These are uncatchable errors and not normally the fault of windows (or linux or the osx kernel).

[Anonymous Coward]

Re: Errm...

Back to programming school

Things crash because they are not error trapped !

[GrumpyOldBloke]

Re: Errm...

May not be that simple. Remember the tilt bits and other DRM junk built into vista and above. What microsoft may be saying is that there are now so many checks, balances encryptions, decryptions and obfuscations built into their OS to keep everyone but the purchaser happy that it is now barely stable in the real world. One flipped bit somewhere and the whole thing comes crashing down.

[Fatman]

Re: the whole thing comes crashing down.

WindblowZe kind of reminds me of a tall building that has only 4 supports, one in each corner.

Pity the fool in a bulldozer that should accidentally knock down one of those supports.

Every tine I watch a building demolition, I am reminded of the 'shitty house of cards WindblowZE is'.

[Nigel 11]

OT - skyscrapers

You might be surprised to know that most tall buildings have only one support, right in the middle. Everything else is cantilevered off this core. These days, the core has to designed proof against airliners colliding with it and large (I don't know how large) explosions.

[BenR]

Re: OT - skyscrapers

http://en.wikipedia.org/wiki/Ronan_Point

Icon - well, because.

[Anonymous Coward]

Crashes just like the Surface ?

Hmm anyone recall Steve Sinofsky's mid-demo crash of the Surface.

[tekgun]

If the hardware is stable then what difference does the OS make? or are they just stating the obvious, that unstable hardware will crash eventually.

[TeeCee]

Yup, I think you have it.

<--- Straight out of this school.

[Ilgaz]

I know their problem

On Mac, if you see the more stylish kernel crash, you immediately suspect memory. On Linux, you -again- suspect memory or your freak hacks while building kernel.

On win, you suspect nothing, you directly blame Microsoft since their image is horrible, majority of users are not advanced etc.

[Dana W]

Re: I know their problem

I agree, the only Mac Kernel panics I have seen in years have been from iffy ram. Macs are reliable, but they are fussy fucks when it comes to ram.

[Anonymous Coward]

What?

Don't overclock? Don't build it yourself? One word: Microsoft can go screw themselves.

[Arctic fox]

Re: "Don't overclock? Don't build it yourself?"

No, that is not the take home message for those of us who like to build ourselves and "fettle" a bit. The (rather obvious) take home message is that if you do, use decent components. If you do not, or you buy a box from GenericKrapTek PLC you are likely to end up in bother - thought we all knew that didn't we, hmm?

[n4blue]

Bad science?

"Once a PC crashes, its crash probability rate goes up by a factor of 100 and for a second and third crash."

This implies cause, doesn't it? Unless the first crash can be artificially engineered somehow for a control group, I don't know how you can support this.

[xyz]

Re: Bad science?

So once a machine knows how to crash, it remembers that look on your face and keeps doing it for fun?

Who'd have thought it... bad machine!

[No, I will not fix your computer]

Re: Bad science?

You're 100% right, it does imply some kind of cause, but it's not at all it's an after the fact corelation, if a machine is flaky then it will probably crash, you don't know which will crash until it does, once you know which one is likely to crash then it's not more likely than before (as the implication) it's just been identified as one of the more likely, as they say, there's lies, damn lies and statistics.

[h4rm0ny]

Re: Bad science?

"This implies cause, doesn't it? Unless the first crash can be artificially engineered somehow for a control group, I don't know how you can support this."

It's just a slightly ambiguous translation from the maths. What they should properly say if they want to be clear is that for a bunch of the same machine, with a chance of crashing in a given time span of X, if one of those machines is known to have crashed previously, the chance of that machine crashing in the given time span is actually 100X. The the probability of recording a crash later has gone up by a factor of a hundred for that machine.

I.e. they are not to say that a previous crash makes it more likely that the machine will crash, but that the probability of a machine that has previously done so crashing, is higher than that of a machine that has not previously done so.

[Anonymous Coward]

So - if you buy an overpriced PC that's been "built" by an expert outfit such as D(h)ell it will help stop the robot rebellion. Sounds like it could be worth it.

*It is a little known fact that the hosepipe ban was engineered by the robots, taking our first line of defence. If only they had been running on a robust platform built by D(h)ell

[AndrueC]

Nothing changes does it? I remember IBM saying much the same thing about OS/2. One comment was "If the motherboard says 20ns RAM latency that's what the OS assumes it is."

[JDX]

overclocked CPUs are more likely to make a Windows PC crash

Um, no s**t Sherlock. Running a CPU faster than designed raises the possibility of problems? Surely ANY computer running ANY OS is more likely to crash in such cases.

[Peter Gathercole]

Re: overclocked CPUs are more likely to make a Windows PC crash

Well, not strictly true.

Most CPUs are designed to run at a certain speed. When a particular member of a chip family is first spun, chances are only a small percentage of the silicon will run reliably at the full design speed, but many more will run at a fraction of that speed. So they are marked with the slower speed, and sold as slower chips. But they were still designed to run that the higher speed.

Manufacturers put pretty much every CPU through some testing, starting at lower speeds and increasing it until the chip fails to execute something correctly. They then stamp the chip with the last speed that worked sucessfully, and then move on.

What overclockers do is that they reason that when a chip runs above it's tested and rated speed, the cause of failure is probably due to heat, so they put a better heatsink on the chip, and then ramp the speed up above the rated value until it fails, and run it at the highest speed that it functioned correctly. The better the cooling, the higher the clock speed you can run it at (that is why some HPCs have direct water cooling of the CPUs, and why people like Amari [I believe] used to sell an actively refrigerated PC at one time).

Unfortunately, another aspect of heat damage is that it can be cumulative. This is, I believe, what Microsoft are trying to say. This aspect has a name, and it's called 'cooking' the CPU. Once you've cooked it, the chances if it running reliably at the same clock speed (or even at it's rated speed) is seriously reduced.

The most obvious case of this I saw was Throughbred AMD Athlon XP2600s (that was the highest speed Throughbred cores with 133MHz FSB, faster Athlon XPs were Barton cores with an FSB of 166MHz). These were actually clocked with a multiplier at something like 2.06GHz, but over time, even if you did not overclock them, they stopped performing at their rated speed. You had to gradually step down the speed to keep the PC stable. Replace the CPU, back up to full speed, at least for a few months. I went through three or four before I realised what was going on, and this happened even with overspec'd heatsinks and fans.

[Paul Smith]

damn lies

Wow, I had forgotten just how gullable El Reg reporting could be on occasion. Somebody did a data trawl and wanted to get paid for for it, so they came up with messages that support their sponsors beliefs (or wishes): "beware the overclocker!" (for he shall not upgrade to your latest overpriced kit) and "beware the non-mainstream PC" (for it shall not make us profit).

Yet , if you look at the same figures from a slightly more critical perspective, you get:

1) "If your PC has a problem that will show up with time, then give it time and it will show up!"

2) "If you do something that makes your PC crash, then if you keep doing it, it will keep crashing!"

But I don't suppose anyone will pay them for stateing the blindingly obvious...

[No, I will not fix your computer]

Re: damn lies

My two main PC's are both overclocked (both Dell; dual quad xeon 490 2.66Ghz -> 3.3Ghz, dual quad xeon T7400 2.5Ghz -> 3Ghz), the T7400 has always been stable (can't recall any crashes) and the 490 crashed a few times until I put the volts up and now is rock solid (which is the opposite of the "if it crashes, expect more" in the article).

The chip manufacturers trouble is that they produce really good chips, the Intel chips have been always ripe for stable overclocking sometimes to a really significant extent (and more so as the yields go up), they often underclock a higher spec chip to make sure they have products across the range.

[tekgun]

It doesn't even mention Windows 8 in the report.

[Roger Greenwood]

Smoke and mirrors . . .

. . to disguise the fact that we all know the real probability of any crash increases exponentially with the importance of the work being done, the proximity of the deadline, and the lack of a backup on another disk.

[Ilgaz]

They will hate you anyway so tell about cheapo RAM next

Honesty from Microsoft risking flames of overclockers.

Why not tell the side effects of using horrible, untested RAM just because it is cheapest?

ECC isn't that luxury anymore too and doesn't need "server/ workstation" class mainboards.

OK, enjoy your bsod andb saved (?) money, just don't call people stupid when they opt in for better reliability instead of 500 fps.

[Gordon 10]

Re: They will hate you anyway so tell about cheapo RAM next

You dont really get overclockers do you?

Most overclockers :

a) Know and accept the risks they run.

b) Spend hours on testing (prime95 and memtest etc) until a stable configuration is reached.

c) backup backup backup.

Indeed lots of us do it just for the lulz - not particularly for some 500 fps magic number. Even for those of use that have some specific performance goals - we still are happy with a trade off between performance and stability.

Lets face it how often have we all seen a standard windows machine crash - pretty damn often.

* Overclocking since the Celery 300a and proud of it. I personally have overclocked specifically to up the performance of my handbrake mkv conversions - and it works quite nicely thanks.

[No, I will not fix your computer]

Re: They will hate you anyway so tell about cheapo RAM next

Thumbs up from me, overclocking is often more than just "better value" or some "free CPU", it's just because you can, often not getting any practical benefit (although video compression/conversion is usually a good reason; consistent, predicatable workload etc.).

I also had a 300a (@450), but my first overclock was a 486Dx 50 @66mhz, I actually has to put a fan on the heatsink! I also had (have somehere?) a BP6 dual celeron 2x466 @600Mhz had to run NT to use the second proc (had to reinstall to get the right HAL), then XP came out and it was lovely, then the celly 667 @1Ghz.... sigh....

[banjomike]

one chance in 2.4 for a crash ????

That is insane. If your PC crashes THAT often then you have more problems than just an overclocked chip! Either that, or Mcrosoft are not overclocking in a sensible way.

[Andus McCoatover]

Overclocking mularkey?

Sorry, don't get it.

I buy a CPU, rated at a certain speed, but moreover has been tested through its design specification - samples at the extreme envelope of reliability.

OK, now because the girlies on the downloaded porno aren't shagging fast enough for my personal enjoyment, and I up the clock outside the manufacturers' carefully designed parameters, why should I get angry when I suddenly get the BSOD on the 'vinegar strokes'?

[Gordon 10]

Re: Overclocking mularkey?

again - you just dont get it. No overclocker I have ever met gets "angry" because their machine crashes more often. They accept thats part of the risk - and minimise it with extensive testing.

[Andus McCoatover]

Re: Overclocking mularkey?

Gordon, dear hart..It was a laugh. Humour. (maybe a poor example, but...)

'Extensive testing' ? Yep, with the same facilities that Intel/AMD etc. have at their fingertips?? Righty-o.

Please, get a life. Or a girlfiend/boyfriend (whatever floats your particular boat). They are not necessarily mutually exclusive.

(Partner vs. life, I mean, although I have to admit having a boyfriend and a girlfriend simultaneously might put me totally off my lunch)

(Palm, meet forehead...)

[yeahyeahno]

I see a problem with the analysis

The Windows computer that tend to run longest without reboot are in large corporates tending to do one task. Corporate machines tend to be from top manufacturers.

I content that the low BSOD on the above machines is related to how they are used, single task, no games, no stress on the machine, no stress on the OS.

Most white box, overclocked machines tend to be used by home owners and game players. These machines run any number of tasks, some of these tasks (like some games) are like an entire additional OS overlaying the Windows OS.

This stresses the hardware and the OS, hence the more frequent BSOD.

[J__M__M]

Re: I see a problem with the analysis

I'm glad I'm not the only one thinking a certain variable was missing. I usually name the variable "user".

Did we really need to crunch numbers to know the guy with the minus-a-side-panel home built crashes more often than his Mom with her Dell laptop?

Crashes juuuuuuuussst might have something to do with whoever's driving...

[Fatman]

Re: ...variable was missing. I usually name the variable "user".

NO, in the WindblowZE world, they are called (l)user.

"Users", are found in *nix land; and "sheeple" are found in Mac country.

[noboard]

Re: I see a problem with the analysis

Exactly what I was thinking. "Machines with a higher chance of having crappy software installed are more likely to crash" should be the title of the report.

[Anonymous Coward]

Well isn't it lucky you have to reboot every patch Tuesday, win8 server will be the pile of shite that server 2008 is. The workstation I'm working on has 116 days up and the server I'm ssh'ed in to has 469 days, Linux of course, windoze just doesn't cut it.

Anybody after a windoze dedicated server, remember, it's impossible to have 3 nines uptime on a single windows box as they take longer than 5 minutes to patch and reboot every month!!!

[the-it-slayer]

Hold on...

Course you can get the three 999s, as long as you cluster your services; no-one would ever know. Would be stupid to rely on a single server for important services. Even if you're on linux.

[JDX]

Re: Hold on...

Don't be daft, he thinks he can entrust all his data to a single 486 desktop running Linux - no need to backup either. After all, it's Linux.

[Anonymous Coward]

Re: Hold on...

Stupid assumptions.... I look after both OS, Linux gives me an order of magnitude less hassle.

I installed a replacement windows DC edition blade this weekend, installing SP1 was over 30 minutes. Click install, wait, reboot, wait while installing 1 of 3 finishes to continue the reboot & restart, wait while 2 of 3 installing then it shuts down and restarts again , now wait again installing 3 of 3 finally 100% done, but wait another 5 mins, ta-da the desktop. Oh the joy, now the next batch of updates.... I think it was 5 reboots, if this was a live server that's an awful lot of down time. Contrast Red Hat Enterprise, yum update and reboot over in a few minutes.

[Anonymous Coward]

Re: Hold on...

Is that what you think? I think I can read, but maybe I can't.

[Anonymous Coward]

Re: Hold on...

The post says _single box_, there are thousands of single dedicated sever hosting boxes online, none with 99.999% uptime. You need enterprise version and shared storage for a cluster.

[h3]

Re: Hold on...

You can have a single box any of the proper UNIX big iron (Or a mainframe) with 99.999% uptime with IBM or Sun / Fujitsu

[JEDIDIAH]

Re: Hold on...

If you don't think the overpriced Unix machines don't have their own single points of failure, you've simply not had enough experience in this industry.

[ukgnome]

Surely it's more about keeping within the engineering tolerances than thrashing the life out of your chips.

[Gordon 10]

Engineering tolerances <> Marketing "tolerances" Generally where Intel are concerned the same physical chip is used at multiple marketing price points coz their yields are generally so good. So those of us who OC like the idea of getting something for nothing.

Additionally I guarantee you that OC'ers will

a) Know their TDP rating off the top of thier head.

b) Be running temp monitoring utilities such as speed fan over and above what the bios provides.

All in all an OC'er will have an idea their PC is going tits up long before the average joe.

[jockmcthingiemibobb]

Sure...the Acer and Compaq $h!te sourced from the local high street chain store is far more reliable than the customised boxes my local PC shop build that have never once blue screened in 5 years? Give me a break.

Be far more interesting to see an AMD vs Intel BSOD report

[blcollier]

Bullpat

Anyone who knows what they are doing would not leave an overclocked system in an unstable state. Frankly anyone who doesn't know what they're doing should leave overclocking well alone. Or at least do some research first and be prepared to be laughed at (and potentially out of pocket) if you break something spectacularly because of your own incompetence.

My PC runs a 2.5GHz E5200 Dual Core chip. I've overclocked that to 3.75GHz using generic El-Cheapo-Brando RAM (though I did get a motherboard that allows me to change the divider value, so that I'm not pushing the RAM past it's stated 800MHz speed). That's a 1.25GHz speed boost for absolutely no cost whatsoever. Clock speed isn't the limiting factor that it once was, but an extra GHz-and-a-quarter for free is not to be sniffed at. It's 100% stable and has never missed a beat. I've never had a BSOD in this stable configuration - ever. The only time I did was because I puished it too far and the RAM was being stressed; so I took the speed back down and, wow look at that - stable again.

Overclocking does not lead to BSODs - if you do it properly. If you don't do it properly then it's your own fault for d**king around with something that you probably shouldn't have been d**king around with.

[JDX]

Re: Bullpat

You're talking nonsense. Problems due to overclocking are statistical in nature... Intel clock it at 2.5Ghz because that means it statistically gives a very low failure rate, at 3.75 they are not satisfied with reliability. The fact your chip runs at this speed just means you got lucky, or that it hasn't broken yet.

[Gordon 10]

Re: Bullpat @JDX

Rubbish.

Since intel dont publish their reliability test stats you dont know whether the 2.5 Ghz is a marketing limit or an engineering limit. History has shown that with the low-mid range limit its generally the latter, with Intel having better yields than demand at that particular frequency & price point so they downsell better chips.

You also need to remember that the price of cpu's has very little to do with the cost of manufacturing and testing at a base level and is all about charging what the market will bear at a particular price point.

[Nigel 11]

Re: Bullpat @JDX

I don't believe that Intel or any other CPU manufacturer would knowingly ship CPUs where getting the right result from any particular operation was be design and testing only probable rather than certain.

Of course, there's a thermodynamically large set of states and they cannot test all of them. They do, however, have access to the CPU simulator, and the ability to probe the actual signals at the surface of the die to validate and calibrate it. They therefore know what are the speed-limiting transitions, and can design their tests to exercise these in particular. If they don't sell a faster version of a particular die, it is fairly likely that they *know* that for this chip and at that speed, at the maximum operating temperature and worst in-spec chip power supply, there is at least one instruction sequence that is very likely to fail.

Overclocking a game is one thing. Overclocking a financial, scientific or engineering model is quite another. Don't. It's more important that the results are right and the system reliable, than getting an extra few percent of speed.

[MJI]

Overclocking

I had one at work once, very rapid but kept going wrong.

Was set up as 16MHz but was a 12MHz chip (286) was slow when changed to match chip afterwards

[phlashbios]

Someone better tell Intel...

Intel currently (and have for some time) sell CPU's with and without a 'K' suffix on them. The 'K' suffix denotes a CPU that is not multiplier locked.For example a Core i5 Ivybridge 3770K 3.5Ghz is an unlocked CPU where as Core i5 Ivybridge 3450 is locked. They deliberately market the 'K' series processors at the overclocking market, and you can reasonably expect the example 3770K CPU I have mentioned here, to overclock to 4.7Ghz in a system that is built to do so (specialist CPU cooler, good quality motherboard and other components)

Intel clearly wouldn't market overclockable CPU's if a) there wasn't a market for them and b) they collapsed in a heap when you tried to overclock them. I am not sure what the point of the authors article is, but I have been happily running my old 2.4Ghz E6600 at 3.4Ghz for the last 4 years with no issues whatsoever, having ensured I provided it with an appropriate cooler and good quality system components to compliment it.

You can get BSOD's on standard PC's with no problem at all if the components are poor quality, but there is no reason to expect overclocked PC's to be any less stable if quality components are used. Just ask Chillblast, Overclockers, Cyberpower, Yoyotech, PCspecialist, Palicomp, Scan, Wired2Fire and all the other specialist PC companies like them that build PC's to order and overclock them.

[Ilgaz]

Re: Someone better tell Intel...

When I first switched from Amiga to PC, evil PC shop sold me a Pentium 75 overclocked to 120 mhz. It took a week to figure the amazing amount of restarted msdos sessions, games dropping to command etc.

These guys really know the PC jungle, tricks stupid shops do etc.

They don't talk about real overclockers. They warn against thieves. They know who will customer blame when they see bsod: them.

[Anonymous Coward]

Good overclocking or bad overclocking?

I've been building and overclocking PC since clocking my bugged Pentium 60 to 66MHz. If done right an overclocked system need not be any less reliable than a stock system. Of course there are many ways to do it wrong - but that's not the same as saying overclocking make unreliable systems. Only bad overclocking does that.

[Tom 35]

Buy your computer from Dell, HP...

It will come with Windows 8 and no off switch for secure boot.

Next they will tell us you get more BSOD if you don't have a touch screen.

[Sarah Davis]

Dog Toffee !!

surely if you have an unstable overclock any OS run on it will be effected,..

unless MS have built some kind of code into W8 that will cause BSOD on O/C'd systems (which i think could be illegal) then this article is bunkum !

"Once a PC crashes, its crash probability rate goes up by a factor of 100 and for a second and third crash." - well D'uhhhhh !.. Not if you fix the problem, in which case the probability drops by an equal factor

Unless W8 is unstable, it should be more or less as effected by overclocking as any other OS

[tre4bax]

Don't I remember reading somewhere that DELLs success was on the back of overclocking 286 chips and thus delivering the fastest device of this type available at that point. Of course nobody had invented the BSOD at this point.

[Jamie Kitson]

Guesses on vendors A and B?

I'm right in thinking this is Intel and AMD, right? Or are they talking computer manufacturers?

[jason 7]

FFS! Its got nothing to do with overclocking, ram etc.

It's more likely to do with the $5 firebox 300W PSUs they install in the crappy white box machines.

It's all in the PSU.

Never spend less than £30 is my motto.

I average around £40-£50 in my machines. Never a BSOD to be seen. Thing of the past.

[EvilGav 1]

Re: FFS! Its got nothing to do with overclocking, ram etc.

Congratulations!!

I was reading down the comments to see if anyone mentioned this. The number of issues that are caused by a faulty PSU as opposed to any other component is astonishing.

Never, ever, ever scrimp on the PSU - if it cant hit within ~5% of the voltage it's meant to be putting out, then it's no use for OCing.

As for the big box manufacturers being more reliable - oh really? My office has exclusively HP and Compaq machines, at one point one of them were building machines with bad PSUs, any desktop problem resulted in a new PSU as they were known as the most likely cause of problems - it got to the point that the engineers carried lots of them around all the time.

Conversely I build and overclock my own machines (and those of friends). My current main machine ran for around 4 years before I saw the first BSOD (driver conflict) and has had a maximum up time of around 90 days before I needed to restart it.

[blcollier]

Re: FFS! Its got nothing to do with overclocking, ram etc.

If a PSU can't hit 5% of the specified voltages then it doesn't match ATX spec and shouldn't be anywhere near a PC.

Not that the rest of your post was inaccurate though... There are big differences between a 500W PSU for £15 and a 500W PSU for £45. For starters, the £45 PSU is less likely to explode when you actually try and get 500W out of it (and I do mean literally explode - it ain't a pretty sight).

[mark l 2]

Is it just me that read between the lines of this MS press release as just a way of saying pretty please buy some Windows 8 pcs from the big OEMs as without them supporting Windows 8 and forcing it down everyones throat whether they want it or not we wont be able to claim 5 million Windows 8 devices sold on first week of release. Where as a small box manufacturer will probably still be able to put Windows 7 or even linux on it.

[richard 7]

Get off their Ivory tower

and stop tarring all small PC manufacturers with the same brush. Those of that care always make a point of using decent quality, reliable parts, hence why I'm one of the few small OEMs that issues a 3 year warranty as standard. I

How about you write software that doesnt actually suck and we will carry on building machines (that unlike the known brands) wont combust/die from known defects/have al manner of lockouts/are made by the cheapest bidder.

Lets see...

Dell - all manner of lockouts to standard parts, a whole slew of defective machines, unbeleiveably poor build quality and no wiggle room to upgrade.

HP - Sued nVidia for xx million then lie through their teeth about there being an issue with pretty much every machine they sold with that chip in it. Driver support for HPO is godawful and and lets not forget batteries

Acer - Veriton PCs, nuff said

Sony - Batteries, loads of lockouts to doing anything

I'll stop now....

[Anonymous Coward]

Cooling, not Overclocking

The most common cause of BSODs aside from genuine hardware faults I've seen comes from inadequate cooling, not overclocking alone and this applies to all brands, even well known ones, and especially laptops as they age.

The cooling fan solutions, thermal pastes you get might work just fine for a year or two from purchase (depending on environment, pet owners, smokers beware) but after that the cooling capacity ends up reduced greatly, you'll see chips stepping down in speed or if they can't step down any further you'll start to see system instability due to raised temperatures even if the automatic shutdown tipping point hasn't yet been reached.

Most computer users seem woefully unaware of this, yet it's a huge problem for the most current systems on the market, including the current generation of consoles which also have active cooling methods and are just as prone to those becoming clogged up or failing as anything else.

[Anonymous Coward]

It took me over a week to get last months "essential" Windows updates installed, with constant failures if I tried to install more than one of them at a time; and now my PC, which has never crashed in the 3 years since I built it, will not last more than a few hours without falling over.

It now insists on cleaning up the outlook folder at every start-up, even though I have NEVER had outlook installed.

M$ SUCK!

[Julian Taylor]

Stop not making sense

"According to Microsoft, the longer the TACT - the longer a PC runs continuously, without shutting down - the more likely it is to experience its first crash thanks to a CPU failure. And once it's crashed once, chances are increased it will keep crashing."

So the more reliable the machine, the higher the probability that it will recurrently crash once, heavens forbid, it finally does crash? Sorry, but isn't this some kind of advance MSFT technical support get-out clause? Can just see ...

"Well, I'm sorry but since your machine has now actually crashed ... yes, yes I KNOW it has had an uptime of almost 12 months without failure ...you have no option but to buy a new machine"

Followed up by:

"Can I recommend a surface tablet to replace your Dell Xen server?"

[philbo]

Maths fail?

>The probability that a machine with five days TACT would crash was one in 330 compared to one in 190 for a PC with an uptime of 30 days.

..and a few lines later..

>Microsoft found CPU subsystems of brand name PCs was one in 120 versus one in 93 for white boxes machines when tested for 30 days of continues running.

So if the probability for brand name PCs was 1:120, white boxes were 1:93 there must be some fantastically reliable neither-brand-name-nor-white-box PCs to get the overall figure up to 1:190

..or am I missing something?

[Anonymous Coward]

Aren't lappies underclocked in various ways... not just the processor itself, but also other slower components? Unless they are desktop replacement types that eat battery. Same goes for many smartphones... to keep battery running longer.

[jason 7]

Yes they are.

The often use lower power mobile versions of the CPUs and the GPUs (if installed) are often clocked far lower then their similar numbered desktop variant.

Sometimes they used to pull tricks like only allowing ram to work in single channel mode rather than dual channel etc. etc.

If it helps get another 20 mins or so out of the battery for say 10% less overall performance.....

[Gil Grissum]

If Windows 8 isn't overclock friendly, then don't overclock. It's not rocket science. As for it liking the top 20 better than PC builders, I built mine and it's rocked Vista and 7 with no probs, so it had better not have any with Windows 8, whenever I get round to upgrading to that.

[Anonymous Coward]

Ridiculous

The OS should obviously be designed to work no matter what the CPU or memory does. Computing is no longer a deterministic affair. Algorithms need to work out what the user wants, what the hardware can do, and fix up a way of achieving it. More of a "can do" attitude, please, Microsoft!

[Anonymous Coward]

So let me get this strait

Windows 8 as how much of a chance to crash after being up just 5 days? Nevermind the overclocking, that's one really crapy OS that it can't handle being up for a few days without it crashing with that much certainty. Two, not only will metro suck, but it looks like we're in for another ME/Vista release

[Custard Fridge]

Laptops more reliable than desktops?

I really, really cannot believe that laptops crash less often than desktops. I've looked after thousands of both over the past 18 years and the desktops always outlast the laptops for reliability / longevity. Surely, I cannot be alone in this?!

[Nigel 11]

Re: Laptops more reliable than desktops?

A laptop has a built-in UPS (the battery and charger) rather than crashing if the mains supply glitches. A laptop often has a slower CPU than a desktop, and a slower cooler hard drive. These may tip the balance, depending on what exactly is being measured.

The desktops last longer, but that's often because the laptop's RAM can't be upgraded enough to make it worth keeping, or because it's too expensive or too much hard work to replace its keyboard after something gets spilled into it. Laptop displays are also harder to fix (desktop: throw away the monitor and plug in another one). And of course, desktops don't get dropped onto a hard surface nearly so often.

[NumptyScrub]

Intel are in trouble then

Since any of their Core iX variants with Turbo Boost™ automatically overclock if they are still within the thermal envelope. I assume if you want a stable PC, and take this research at face value, then you need to specify "no Turbo Boost™ please" for the Intel CPU or get AMD?

[SplitBrain]

And as for a Unix Server

Best case in point here:

IBM RS/6000 running AIX 4.3.3, a good 9 years old, running Websphere MQ as a messaging hub for the whole of Europe for a very large Custodial bank.

We had this server running flat out, CPU's pegged at 100% for over 3 years, never missed a beat. Never missed a transcaction.

That's why Unix and RISC runs the world chaps, leave the BSOD'ing PC's to create your word docs and excel spreadheets......to be frank that's all they are good for.

Supposed Windows "Techies".....flame away.

[jason 7]

Re: And as for a Unix Server

Yep and I bet it runs through a decent built in PSU and also a lovely filtered UPS etc. etc.

It's really not difficult to build stable hardware running any OS as long as you spend a bit.

[Nigel 11]

Re: And as for a Unix Server

I've seen that sort of reliability from desktop PCs crunching numbers. No unscheduled downtime other than those caused by the electricity supply, up until the day that it was decided that a newer system would make better use of the electricity. They weren't even required to be quite that reliable, they just were!

Running Linux, of course. And I'm sure that your IBM's disk subsystem was taking a much greater pounding.

[jocaferro]

Re: And as for a Unix Server

"That's why Unix and RISC runs the world chaps,"

Do you mean the most powerful and fastest supercomputer in the whole world?

If this is the case, looking only at the top ten machines RISC is very well represented. However Linux equals to 100% so you "quite" wrong.

The whole 500:

- x86 dominates close to 90%;

- Linux dominates close to 93%.

Here you are "really" wrong.

They really know what they are doing and the simple "magic" formula is - the most reliable machines with the most reliable OS.

Windows?

2 machines!

(Please draw your own conclusions)

[Nigel 11]

Overclocked vs. Flat-out

An overclocked CPU is a CPU running outside its specification. It's been tested by the manufacturer (who knows the weakest spots w.r.t. timing) at a particular speed and may well have been found wanting at a higher speed. It's blindingly obvious that an overclocked CPU may not be working 100% correctly, and can only be recoemmended to someone who cares neither about correct results nor about reliability. A gamer, maybe.

Flat-out, on the other hand, should not reduce reliability. With modern CPUs there is a feedback loop to slow down the CPU when the chip temperature limit is reached. I work in an environment where desktop PCs are crunching numbers 24x7 most days of the year, and our desktop systems don't seem noticeably unreliable. By far the commonest failure is a PSU fail, followed by a hard disk fail. Failed CPUs are as rare as hen's teeth and failed MoBos only slightly commoner. In the old (Athlon) days when a CPU didn't slow down and could actually overheat until the heat crashed it, failed CPUs were also as rare as hen's teeth. I'd vacuum the heatsink, replace the fan, and the system would happily reboot and last as long as any other. Too high a temperature slows down the logic gates in the CPU, until it's the equivalent of a CPU that's overclocked one notch too far, and crashes.

Oh yes, and always run memtest overnight on a machine that's randomly unreliable. Low-incidence errors on RAM will do that. it's why servers (and serious engineering workstations) have ECC RAM. If memtest crashes rather than reporting errors, suspect your power supply first (you may or may not see the problem with a DVM).

[MaxxB1ade]

Elephants?

What exactly are these statistics referring to? 1 in 190 what? elephants?

It sounds more like Microsoft are trying to keep us buying from the big system manufacturers so they will continue pre-loading Windows on their machines.

[Don Mitchell]

ECC

There have been a number of reports within Microsoft on the cause of crashes. 50% are due to hardware memory errors, which is why servers use ECC ram (error-correcting coded). I like to build my own PCs, and definitely quality matters. Supermicro or Intel motherboards are a good idea too.

On a system with ECC and decent parts, like my last three machines, I have literally never seen a blue screen crash. My windows 7 machine runs for weeks, until some kind of undate requires me to reboot it.

[IGnatius T Foobar]

The real problem

The real problem is that your PC is far more likely to run like crap, if you run a crap operating system like Windows (any version, including Vista, Vista 7, or Vista 8).

[Nigel 11]

Re: The real problem

have you ever tried running Win 98 SE under VMware on state of the art hardware?

The OS was / is crap, but it sure boots impressively fast!

[Andy Fletcher]

I don't get

what has changed.

[Bradley Hardleigh-Hadderchance]

I won't be oc'ing any systems with win8

In fact only done a few with winxp.

I wouldn't oc a Dell or somesuch. They are pretty locked in the bios as you know. If you can reach it, then by all means give it a 'tug'. As for a machine you build yerself....

It took me a few weeks of reading up, loads of info on the net. Another couple of weeks to test and make sure everything is ok. Fairly easy to do if you have a bit of an ocd/autistic spectrum mindset. Most people would be better off paying someone else to do it for them (defeats the purpose), or buying a higher spec'd machine (defeats the purpose) or just leaving well alone.

I have a very stable 30 percent increase on my main workstation. No BSODs at all. It has been tested to buggery with Linx, S&M, Prime95, Memtest, Super Pi, Orthos and my all time fave - OCCT.

I'm of the school that says a 12 hour test is all you need and 24 hours is just too much. I may be wrong, but this is why I like tools like OCCT, I have passed over 12 hours of linpack and Prime95, only for errors to be picked up in 2-3 hours of this fine tool. OCCT has a new version out, btw, you should check it out.

I could have run my chip much more towards the edge, but the temp/power draw were not worth it. This is about as free a lunch as one can get in this life, so I accepted it happily after only paying about a month of my time to achieve it. I suppose if I had gone out to work for that time and be paid a decent wage, I could have bought two machines. But it wouldn't have been half as much fun.

My system trundles along nicely at stock temps, is humanly noticeably faster (10 percent is not noticeable by humans so is pointless, 20 percent is a grey area, 30 percent should be noticed by most), it doesn't crash, and apart from the few extra quid for the new fans and air-cooler, hasn't really cost me that much more than the price of the machine (the extra leccy being almost negligible). And the couple of other systems I did for other people, I have had no comebacks, but they are friends and knew the risk when I built them.

No, I wouldn't overclock a stock Dell, but there is no way in hell I wouldn't overclock a machine I built myself.

Just a nudge ;-)

[No, I will not fix your computer]

Re: I won't be oc'ing any systems with win8

Disagree with not overclocking Dells, bide your time and buy the right bit of kit and you can have some fun, there's no bios overclocking true, enough but my T7400 has a pair of E5420's BSel'd from 2.5Ghz to 3Ghz, no volt change (because the chip come from the same die/fab as the E5472's, apart from the FSB and the CPUID they are identical chips), OK so stick those chips on a Skulltrail and you'll be able to get more out of them (potentially a lot more), but hey. The 490 was a little trickier, got a pair of X5350's BSel from 2.66Ghz to 3.33Ghz, one worked perfectly with no volt change, the other needed a trickle extra, ran hotter than I wanted (ideally E5350's would have been better, but I got the X5350's for a silly cheap price), ended up with a very cheap, very fast, very stable non linear video processor.

[Nigel 11]

But how do you know?

How do you know that by overclocking your system, you haven't created conditions that cause, say 34387.00*1.01 to compute as 79231.48, that sum being the decimal representation of something that Intel knows is on the critical timing path of the FPU?

Next thing you know, all the grade 3 techs have been paid over twice their usual salary rather than the scheduled 1% pay rise, and the FD wants to see you NOW!

But even if you're just number-crunching a model that you know will unconditionally iterate to correctness (in the mathematical sense), you still can't be sure. Maybe the FP error was in the calculation of the residual error, and the iteration is terminated before the answer is right? Let's hope your Ph.D. doesn't depend on that result.

Or maybe it's not an FP error, it's in one of those rarely-used instructions that only OS kernels ever use (which may be where MS is coming from, though I have my doubts). Consequences: corrupt database? corrupt filesystem? deadlocked system? security compromise?

I won't overclock a CPU for work, period. (for fun, OK). Intel knows what are the timing-critical logic paths in their billion-transistor chip. I'm sure that if a significant fraction of the dies tested OK at 5% faster on the critical paths, Intel would sell them specified for running 5% faster. Fact: they don't. Becuase Intel knows, this chip doesn't work 100% reliably above that speed. Maybe you'll never crash into the invisible wall, but logically it must be there.

[Anonymous Coward]

nothing to do with Windows

Well, of course it makes sense that unstable computers are likely crash more than once. But also, multiple BSODs can occur while dialing in the maximum stable overclock. That might explain some of the statistics, depending on how they were gathered.

However, in the big picture, why overclock (Intel) processors at all these days? The processors that need the boost can't be overclocked and processors that don't need the boost can (for some extra money, of course, in both CPU and motherboard). The days of taking a 50 euro processor, combining it with a 70 euro motherboard, and overclocking it stably to faster than stock high end speeds are gone (farewell, E5200 and the rock-solid budget P43/P45 mobos). These days overclocking is tamed, just something to sell to "enthusiasts" to justify spending a minimum of 300 euros for a motherboard+CPU combo.

[Anonymous Coward]

Fuck me this is news!

Chips running right up against the limit crash!!!!

This does not change at all in Windoze 8, just looks different. Carry on.

[Will Godfrey]

In other news

Run your car with the revs in the red section and your engine might fail.

[Someone Else]

Methinks you got the message wrong

<i?"The message: pick your PC wisely and resist home-brew fiddling.</i>

Actually I think the message is: Just Say No to Windows 8!

(There...fixed it for ya!)

[CFWhitman]

It's a rather simplistic analysis of a rather complex set of information. "White box PCs are more likely to crash." That's quite possible, but it's also overly general. There could be certain white box makers that build PCs that crash all the time and others that crash less often than most large PC makers. When you lump all smaller companies together you are asking for misleading data.

[Waderider]

At Nigel11...

Nigel my friend, a little knowledge is a danger thing. This:

"You might be surprised to know that most tall buildings have only one support, right in the middle. Everything else is cantilevered off this core. "

....is nonsense! Your error comes from I suppose an understandable misunderstanding. Many tall buildings are built using a shear core (often housing lift shafts, utilities etc.) that deals with horizontal loadings such as wind. The steel or reinforced concrete frame coupled to this deals with the horizontal loads, so every column with a foundation (which is to say most columns) deals with dead and imposed vertical loads in the structure, as you would expect. Cantilevering the majority of the structure from a central core is certainly not the normal way of doing things and would at the very least make foundation design a challenge, and structural stability more difficult to achieve.

You must have watched something regarding structures on Discovery Channel that confused you.

[RAMChYLD]

Branded PCs

I have a few problems with them. Firstly, they're mostly Intel chip boxes. I am a AMD fanboy. That doesn't work.

Secondly, I want fancy features like SLI and hardware SPU in the form of an E-Mu equipped soundcard with it's own RAM. No branded PCs ship with that either.

Lastly, I don't like bloatware. Windows included, Will be ditching it for ReactOS the moment ReactOS goes gold (which sadly, will most likely not happen in my lifetime).

[Anonymous Coward]

Microsoft fud

So the rated speed of a cpu is where 99.99% run reliably indefinitely in an appropriately cooled environment, and overclocking pushes the performance up at the risk of increasing instability in a percentage of the total number of cpus based on the idea that 'most cpus can run a bit faster without raising instability unacceptably.

If 99.99% of the cpus could run reliably at the higher speed, they would be rated for the higher speed.

Overclockers knowingly risk the increased instability in return for the improved performance.

So much we knew. Microsoft seem to be saying that if you leave your machine running indefinitely and the OS hasn't crashed it within a number of days, then a subsequent crash is due to hardware, and overclocking increases hardware crashes. The Windows 8 interpretation by the author I can't follow

[Anonymous Coward]

The longer a PC runs continuously, without shutting down...

I see.

This assertion takes me back to the days when Windows was incapable of running more than a few hours without bombing out due to memory leaks and badly written code. Are Microsoft giving us hints about the expectations pc's running W8 (BTW are we expected to shorten that to WAIT?) are to suffering similar problems?

[jocaferro]

The real reason?

Dear Microsoft Windows user:

1. buy brand name Windows machines;

2. if you have problems with this kind of machines that is "only" due to the shit installed by default.

3. Don't worry be happy - we can clean that shit for the fantastic price of $99.

4. As you can see, buying a brand name machine is always a win-win solution.

Best regards.

[cswilson1976]

screw the oem or white box systems

I have built all of my own systems(other then laptops of course) and i seldom, if ever get bsd's.

I don't over clock them. I have more cooling then most. Then again I do tend to use good components.

current up time on my desktop is 14169509 seconds(163days) , and yes I do do updates, but tend to do the hotfixes. Rebooting takes forever for the media server on here to re-index. Last reboot was to upgrade raid array