Overclock or Not? Silent corruption/testing

Well, I can share you my experience with overclocking.

I've been overclocking since a few years, I always wanted to squeeze every bit of performance out of my machine. I tested them rigorously for stability to make sure I it was really stable. (48 hours of prime and other stress test plus 3Dmark).

Usually, the motherboard is the first piece of hardware to go. The last 3 motherboards I used (all 2 years +) started to act weird after 2 years or so. The first one started to have issues with the mosfets because of running high voltages for an extended period (~1.5 years). The second started to have issues with random reboots of which I think was related to the chipset burning up. The north-bridge got extremely hot, even though it was still "within" specs. The third one same story, although I haven't figured out whether it is a software isssue, or hardware related.

My current board holds up pretty well actually. I oc'ed my e6600 to 3.4 for more than a year, and currently I'm running my e8200 @ 4Ghz. I have to admit though, the moment I'll graduate and find a job, i'm going to replace my computer. I basically loose confidence in my computers do to critical work after 2 to 3 years because of overclocking. The CPU, RAM, GFX etc are all fine, but it is always the motherboard that's acting up.

Anyways, that's my experience with overclocking.

My recommendations is to, overclock the heck out of it, test it rigously for more than 24 hours. Run multiple prime tools, burn in tools and gfx tools. Just stress the heck out of it. Keep an eye on the temps, and if everything is stable, back of your overclock a bit, say 10% so you won't kill your system within a few years.

Back in the days when I was hardcore overclocking, I couldn't care less about the longivity of my computer, because I upgraded my components constantly. I kept my the old ones around though, in my gf's computer of my parents, so I now know what overclocking will do to what components. Now I don't care about overclocking anymore, but do want a stable and fast system that lasts at least a few years so..overclock wisely folks!

Moppie wrote in post #7376329
Over passionate disucssion on the vitures and problems of overclocking really belong on the dedicated over clocking forums.


Running ANY piece of technology beyond it's manufacturer rated specifications increases its chance of failure. It does not matter if it is a bridge, a tyre or a processor. In all cases, they either work OR they don't.


Overclocking, if done properly, can yield some impressive performance imporvments.
But, the catch here "if done properly". If not done properly, with care, the correct parts, and proper testing, then there is a very high chance you will end up with unstable machine.
Even if done properly, there is still a small chance things could fail, or the life exptancy of the machine could be compromised.
It might be possible to build a very stable quad core i7 or Q6600 running at 3.8 or more ghz, but there is absolutly no data available to say that same machine will still be stable in 2 years, because neither of those chips have been availbe for over clocking for more than 2 years.
Long term effects of over clocking are not well tested, by the very nature of the over clocking comunitiy, most members always strive for the best, the fastest and the latest and greatest, so do not hold onto complete systems long enough for a wide enough range or useable data to be gathered.


I would suggest anyone looking at overclocking thier photography work station spend a lot of time, and do a lot of resurch into the subject. There is "free" performance to be had, but there is also the increased risk of system failure.

[edit]

last note:

Heat is the enemy of every electrical equipment, but more so in computers. Cooling is paramount, so invest in a good cpu cooler (use the big ones, big fans=lower rpm=quit) and a good case with preferably 120mm case fan mounts. Small fans make a higher pitched sound, so avoid those. Also try to put a fan on your hard drives, keep them under 50 degrees centigrade. I've seen many hdd dying because of the lack of cooling.

The key to overclocking and having a stable system while not destroying hardware is proper cooling and keeping the lowest possible voltages. It is very easy to throw a huge heatsink on the processor and crank the voltages to achieve a stable overclock, but that will stress all the power supply components and lead to possible pre-mature failure. The best way is to slowly raise the voltages till you can get it prime-stable then increase the voltage 1 or 2 steps.

Prime-stable means running prime95 for many hours without recieving an error. Prime95 is designed to produce an error on the slightest miscalaulation, I believe even if it is only 1 bit off. It would not be uncommon to spend 3 days testing stability, 24hrs for each of primes stress test modes: Small FFTs, In-Place FFTs, and Blend. If all processor cores are not maxed at 100% usage then run 2 copies of prime.

Proper temperatures are also very important. Just keeping them at or a few degrees below spec is not enough, try for as low as possible. Heatsinks are cheap, get a good one for your processor and use a silver based heatsink compound. You should also get better heatsinks with fans on your northbridge and southbridge. Heatsinks on the processor power mosfets that are around the processor socket are not a bad idea either and they run only a few dollars.

You should also show some restraint when overclocking. I have my current 2.4ghz q6600 running at 3.0ghz even though I have tested it to be stable at a considerably higher speed. 3.0 is just fine for me right now and it is keeping my temps and voltages even lower. I ran my previous 2.8 P4 at 3.4 for many years with no issues on the motherboard or processor.

Also make sure to clean out your system of dust and other buildup avery couple of months.

Those are the tips that have helped me over the years, hope they help someone else.


Mike

You call 84 degrees stable? Are you out of your f***ing mind?! You should really read up on the Nehalem limits because I've got a funny feeling that you have no regard for the specifications put in place to maintain consistent operation of the CPU. While TJMax on the i7 is still 100 degrees the CPU suffers degredation above 85 degrees as the added heat actually causes an adverse performance effect.

Also, prime95 means nothing in the i7 realm... run LinX 0.5.6 and see if you can get past 8 runs using all available memory. If you're pushing 84 degrees on Prime95 @ 3.8 then I'd be willing to bet that you'll see 90+ on LinX and will most likely fry something in the process. Again, this is an entirely new architecture which has very different behaviours from previous generations and as such needs to be tested differently to assure 100% stability.

For those interested in overclocking as mike105105 said keep it simple and maintain proper temps. I've got my i7 920 dialed into 3.6ghz and my load temps never go above 62 degrees running LinX and I see about 54-56 degrees in video encoding/rendering.

Things to keep in mind if you're juicing up your i7.

- CPU VTT needs to be within .5v of you DIMM voltage.
- DIMM voltage cannot exceed 1.65v according to Intel spec, ignore this and you risk killing your memory controller.
- be VERY aware of your NB and VREG temps, a lot of people just look at the CPU temp and forget that if your VREG and NB reach extremes you are going to be in a whole new world of problems.

For those with RAID setups, do not perform overclocking while your RAID array is active as you risk damage/degradation to the disks or controller. Drop the OS on a single drive, work out the gremlins and then when you're happy form the RAID array and start a fresh install.

smcclelland wrote in post #7396550
You call 84 degrees stable? Are you out of your f***ing mind?! You should really read up on the Nehalem limits because I've got a funny feeling that you have no regard for the specifications put in place to maintain consistent operation of the CPU. While TJMax on the i7 is still 100 degrees the CPU suffers degredation above 85 degrees as the added heat actually causes an adverse performance effect.

Also, prime95 means nothing in the i7 realm... run LinX 0.5.6 and see if you can get past 8 runs using all available memory. If you're pushing 84 degrees on Prime95 @ 3.8 then I'd be willing to bet that you'll see 90+ on LinX and will most likely fry something in the process. Again, this is an entirely new architecture which has very different behaviours from previous generations and as such needs to be tested differently to assure 100% stability.

For those interested in overclocking as mike105105 said keep it simple and maintain proper temps. I've got my i7 920 dialed into 3.6ghz and my load temps never go above 62 degrees running LinX and I see about 54-56 degrees in video encoding/rendering.

Things to keep in mind if you're juicing up your i7.

- CPU VTT needs to be within .5v of you DIMM voltage.
- DIMM voltage cannot exceed 1.65v according to Intel spec, ignore this and you risk killing your memory controller.
- be VERY aware of your NB and VREG temps, a lot of people just look at the CPU temp and forget that if your VREG and NB reach extremes you are going to be in a whole new world of problems.

For those with RAID setups, do not perform overclocking while your RAID array is active as you risk damage/degradation to the disks or controller. Drop the OS on a single drive, work out the gremlins and then when you're happy form the RAID array and start a fresh install.

How about 20 passes of linx with all memory over 9 hours WHILE running prime95 at the same time? (I posted screen shots) Temps hit 85 which is fine at the high end for this CPU. Just running linx alone they do not get anywhere near that high.

So I think you are fear mongering with the temp thing. I have read everything I can about the CPU and can not find any reference to your Max of 85. Its 100C according to Intel. For what it is worth my PC idles the CPU right around 36 or 37C.(hottest core)

When playing games or doing image processing ( large batch processes that take a while) it will get into the mid 60's to 70C

The other stuff about voltages is correct and useful, so is the bit about RAID. I think raid on a O/C'd system is asking for trouble.

I only see burn test and prime95... no linx in that first screen shot that I can see of.

TJMax is 100 degrees which means you're at the processors thermal max. 85 degrees is the hottest recommended operating temperatures for any period of time due to the fact you are working with an IMC. If you want to run 85 degrees on your chip that's cool, but lots of people will tell you to keep it well below 80 degrees for optimal operating.

Trust me, if I was fear mongering I'd have a lot less of an educated opinion

smcclelland wrote in post #7397290
I only see burn test and prime95... no linx in that first screen shot that I can see of.

TJMax is 100 degrees which means you're at the processors thermal max. 85 degrees is the hottest recommended operating temperatures for any period of time due to the fact you are working with an IMC. If you want to run 85 degrees on your chip that's cool, but lots of people will tell you to keep it well below 80 degrees for optimal operating.

Trust me, if I was fear mongering I'd have a lot less of an educated opinion

Burn test is a front end for linx no? As is OCCT? Correct me if I am wrong!

Ok did some more research Linx is also a front end for linpac.

INTEL BURN TEST
OCCT Linpac
Linx

are all front ends for Linpac.

So yes what you saw in the first screen shot was 9 hours of Linpac and prime 95 at the same time. Returning perfect error free results.
This at 100 percent stress on the CPU. Your right that is a intense test. But it passed with out issue. Now to me this means my PC should deal
with day to day operations with ease. Thats the point of this test right?

I am being serious here.... am I missing something? I will always stop to learn from somebody that is in the know.

Keep in mind, if it's winter-time now where you are living (northern hemisphere), and you don't have air conditioning for the summer, the ambient temp difference is going to add up to the 84 degrees..maybe that's a good enough reason to cut your overclock a bit...

better save than sorry, right?

So the problem is that IBT does not account for virtual cores (i.e. HT) so the results are essentially a false positive and the other main issue is it's not really a solid test for stability. The same goes for OCCT Linpac, it does not use all 8 cores (4 physical + 4 virtual via HT) so if you take a look at your CPU charts running just Linpac you will see that you're hardly even pushing the processor.

LinX does what is essentially known as a suicide run... it will absolutely bombard your memory/physical + virtual cores until the cows come home or your computer explodes in a ball of fire. 0.5.6 was written with the intention of suicide runs on the Nehalem platform and when you get to roughly the 8th pass you will know whether or not you are completely stable (don't try to get to 8 just for the sake of doing it, if you see 80 degrees or more you need to bail and take another look at the OC).

Again, running Linpac + Prime95 at the same time doesn't really show anything given that CPU allocation will be put towards Prime95 and what's left reserved for IBT/Linpac. Theoretically you are running the two in parallel but it's just hashing out the processes based on demand of CPU consumption.

Prime95 is pretty relaxed, I've run it for 24 hours stable on an old Q9650 box I had and then gone to encode a movie and my machine has BSOD'd so take the results with a grain of salt. These tests are all great to show "intense" situations but remember that these are all synthetic benchmarks and actually have little to no bearing at all on everyday usage. What the OC stability really boils down to is can you do everything you purchased the machine for day in and day out 24/7 without errors.

Again, to each their own but I would not recommend running a Nehalem processor at that level of heat for any periods of time unless you cherish the idea of dropping money on processors and boards frequently or you are intending on dousing that bad boy in LN2 and calling it a day

OK serious question here, I need to ask you a question. This is not to try to insult you, but needed because what you are saying goes beyond conventional Wisdom. ( net wisdom) What is your back ground? Where do you get your information? You sound like a authority on this, I need to make sure you actually are, before I launch myself down this road again.

smcclelland wrote in post #7398269
So the problem is that IBT does not account for virtual cores (i.e. HT) so the results are essentially a false positive and the other main issue is it's not really a solid test for stability. The same goes for OCCT Linpac, it does not use all 8 cores (4 physical + 4 virtual via HT) so if you take a look at your CPU charts running just Linpac you will see that you're hardly even pushing the processor.

LinX does what is essentially known as a suicide run... it will absolutely bombard your memory/physical + virtual cores until the cows come home or your computer explodes in a ball of fire. 0.5.6 was written with the intention of suicide runs on the Nehalem platform and when you get to roughly the 8th pass you will know whether or not you are completely stable (don't try to get to 8 just for the sake of doing it, if you see 80 degrees or more you need to bail and take another look at the OC).

Again, running Linpac + Prime95 at the same time doesn't really show anything given that CPU allocation will be put towards Prime95 and what's left reserved for IBT/Linpac. Theoretically you are running the two in parallel but it's just hashing out the processes based on demand of CPU consumption.

Prime95 is pretty relaxed, I've run it for 24 hours stable on an old Q9650 box I had and then gone to encode a movie and my machine has BSOD'd so take the results with a grain of salt. These tests are all great to show "intense" situations but remember that these are all synthetic benchmarks and actually have little to no bearing at all on everyday usage. What the OC stability really boils down to is can you do everything you purchased the machine for day in and day out 24/7 without errors.

Again, to each their own but I would not recommend running a Nehalem processor at that level of heat for any periods of time unless you cherish the idea of dropping money on processors and boards frequently or you are intending on dousing that bad boy in LN2 and calling it a day

In the days prior to my photography habit, when my PCs were NOT being used to house important data,.. I used to overclock the snot out of my rigs.
I used to push them literally to death, I had 5 or six running at a time at full CPU load solving the sequencing of the human genome,. etc..

However, now my data is far too precious for that kind of risk, and I no longer have a use for the "render farm" so it's just the one editing PC.

RAID 0 has given way to redundant data security in RAID 1 mirroring, and I no longer push the timings beyond factory spec.

I can understand the desire to shave away at benchmark scores, I have known the call of cranking out work units as fast as possible, but none of these weeks help my photography, which is my priority,.

I'd advise against overclocking any PC that is not being used for "fun"

85 degrees seems pretty hot to me, i'd probably shut it down if it hit 70.

lens pirate wrote in post #7398516
OK serious question here, I need to ask you a question. This is not to try to insult you, but needed because what you are saying goes beyond conventional Wisdom. ( net wisdom) What is your back ground? Where do you get your information? You sound like a authority on this, I need to make sure you actually are, before I launch myself down this road again.

My background is primarily the tech sector, I've spent the last 7 years working in the CG/video games/commercial industry as a Technical Director/Technical Artist and spent 2 years prior to that as an intern with a networking/tech firm. I've been overclocking since the days of the dual socket celerons and PIII's and have blown more than my fair share of computer parts over that time frame I get my information from all over, it's basically my job description to stay informed about technology so I read lots of geeky white papers, look at specs, charts, graphs, read reviews, forums and then do my own research/testing.

I wouldn't call myself an "authority" on this, I'm just a code junky (programmer type) who takes a keen interest in the advances of threading and parallel processing.

Well in fairness, I have to admit that I went and got linx a alternative front end for linpac. I have no idea why but it DID place more load on the CPU than Combining prime95 and IBT.

In fact the temps got past my comfort zone and I stopped the first run at 85C when it showed no signs of slowing down.

I dropped my CPU speed to 3700 and dropped the voltage 1 notch. This time I let it run and it finished a run of LINX and two cores just touched 85C. No higher. It finished with out errors.

I had my case buttoned up tight with all the air filters in place. Tonight I will take the air filters off( good for 3C) and run all night to see how she goes.

So I need to thank you for finding a test that ran my machine even harder. I want every last bit of Stable safe speed but not one bit over that.

So in the end you did me a favor by introducing me to a better tool. For that I thank you. Even though you screwed me out of 100 MHZ LOL

Heh believe me, I would love to see people breach 4ghz on their i7 920's but I also don't want to see people disheartened when they fry their brand new PC Good to know that you grabbed LinX, it really is one of the most solid torture tests there is out there for the i7 chips.

One suggestion I would have is once you've gotten what you think is stable and are hitting those high temps start to back down your CPU VTT and your CPU VCore until you get to a pretty modest level. I run 170x20 @ 1.2375v and +100mV VTT with 1.15 QPI PLL and it's perfect for me at my speed/temps. The ultimate task is to run LinX through about 8 times (set the Times to 10 and let er go) and record any errors or temps. I urge you to check your NB and VREG temps as well, if your cores are at 85 degrees I'd be willing to bet your NB is close to 68 degrees and your VREG is probably pushing 71-76 degrees.

Keep it up

Well it passed 20 passes of LINX over night at 3600 ( 190x19) with a vcore of 1.375 top temp on the hottest core was 81.

It interesting. Its cooler but know more stable. My limit seems thermal.

Here is a question, running games, DPP and photoshop I never EVER get temps even coming close to approaching this. The temps at 3600, 3700, 3800, at CPU voltages up to 1.425 never climb above the mid 70's at 3800 and on average stay about 5 to 7C cooler at 3600. But the machine is faster at 3.8

Assuming that at 3800 the machine stays with in reasonable temp limts during use as defined by INTEL and remains stable and returns no calculation errors in testing..... why not run it this way?

So I guess the Question is assuming there is a envelope of performance that is OK for the PC as defined by official spec.
If you stay with in that spec at all times does it matter if you average 60, 70, 80, or even 90 percent of the Maximum allowable? Shouldn't your PC be able to run at 100 percent?

Once again LINX is really a tough test man... on my box at least 5c HOTTER than Prime 95 AND IBT running at once!!