Prime95/IntelBurnTest: how much is enough?

[rga]

I have an i2600k running at 4.5GHz at 1.36V with hyper threading disabled since I don't really run anything that takes advantage of it, and the chip seems to run cooler (I know, I know. I could've saved $100 and gone with a 2500k).

It just hit the 12 hour Prime95 blend test mark, and it ran 100 times through IntelBurnTest without any output disparities and without crashing.

I've been a long time lurker, and I've been reading around the web somewhat. I see a lot of people claiming that their chip is stable after passing three hours of Prime95. I've always thought 24 hours was the default benchmark.

I understand that if I want 100% stable I should run it at stock because that's what it was designed for. But is three hours of Prime95 enough to call stable? Is 24 hours too much?

By the way, I'm using a CoolerMaster Hyper 212+ with the push/pull fan setup. Prime95 doesn't seem to push temps past 65C, but would crash within 10 hours with lower voltages. IntelBurnTest, on the other hand, has pushed my chip up as high as 76C, but has passed 100 tests with voltage as low 1.335V.

 

[Concillian]

Personally, I find my limit on IBT@ 8-10 hours (overnight), then turn down clocks about 5% from there. That way, next summer when I haven't gotten around to cleaning out my case and I have some dust fuzzies in there, my computer to work flawlessly, but every so slightly slower than it could be working if I had air conditioning and cleaned out my case twice a month.

If it passes IBT, and you turn clocks down 5% (which is like 200 MHz, so quite a bit), then guaranteed it will pass Prime forever.

I'm sure most of the posters here are less conservative than I am. I just don't see any value in a potentially unstable system.

I have had OCs that were 24 hour prime stable and certain games would consistently and repeatably cause crashes. IMO Prime is not enough. IBT uncovers issues quicker. No matter what you use though, it's much easier to just back down a couple hundred MHz than to test every possible condition.

As you are hitting in your case, I usually hit my max comfortable temperature in IBT before instability. 76C is on the high side. Usually I stay 25-30C from tjmax (98C for 2500k / 2600k). 70C is a comfortable limit. In your case, I'd back down to 4.3 GHz and call it a day.

 

[sangyup81]

I've had errors in Prime Blend between 24-48 hours. Unfortunately, the memory controllers built in to CPUs make it a pain to test stability

 

[tweakboy]

I have an i2600k running at 4.5GHz at 1.36V with hyper threading disabled since I don't really run anything that takes advantage of it, and the chip seems to run cooler (I know, I know. I could've saved $100 and gone with a 2500k).

It just hit the 12 hour Prime95 blend test mark, and it ran 100 times through IntelBurnTest without any output disparities and without crashing.

I've been a long time lurker, and I've been reading around the web somewhat. I see a lot of people claiming that their chip is stable after passing three hours of Prime95. I've always thought 24 hours was the default benchmark.

I understand that if I want 100% stable I should run it at stock because that's what it was designed for. But is three hours of Prime95 enough to call stable? Is 24 hours too much?

By the way, I'm using a CoolerMaster Hyper 212+ with the push/pull fan setup. Prime95 doesn't seem to push temps past 65C, but would crash within 10 hours with lower voltages. IntelBurnTest, on the other hand, has pushed my chip up as high as 76C, but has passed 100 tests with voltage as low 1.335V.

Your never going to game for 12 hours or 24 hours straight! So its pointless to run this. As long as games run for 30 min to hour, your OC is stable.

You use 60 percent of CPU power,, this is just a benchie or stability tool,, no need to run it for more then 2 hours.... gl Intel says 80c is danger zone. As long as you in 60's and 70's on load you are safe. The chips are meant to get hot. gl

 

[4ghz]

My i5 2500k run would prime stable for over 3 hours then bsod sometimes when booting into Windows. Personally I just run Prime for about 3-4 hrs now then 1 hr of Intel burn test. Then use it normally for a week. No locks or bsod then I consider that stable enough for me.

 

[cantholdanymore]

Sticky:
http://forums.anandtech.com/showthread.php?t=2195063

 

[BTRY B 529th FA BN]

Depends what you do on your machine. I would not trust a high overclock while running data sensitive material..

 

[BonzaiDuck]

There are some four settings for stressing under IBT -- from "standard" to "Maximum." There is a consensus among other colleagues here that 30 to 40 iterations should nail it. That's probably a couple hours-worth.

If I remember -- [correct me] -- an earlier Sandy chipset doesn't use the on-chip Intel graphics. (P67?) For a Z68-based or later, you'd want to be sure you update either IBT or LinX with the binaries that support stressing that part of the CPU -- it makes a big difference in thermal output.

I run Prime95 -- all three tests -- but focus on the Large-FFT for the longest test. But I also run IBT or LinX for maybe triple what their promoters say is sufficient. Since I validated the voltage setting for both 4.62 and 4.64Ghz (with a tweak to the bCLK thrown in to the mix), I reset the bCLK to its stock 100Mhz and run at 4.6Ghz with the same voltage -- consistent with some other advice on this thread.

Without IBT or LinX, I'd be inclined to run PRIME95 Large-FFT for maybe 15 hours -- if the other two tests passed after 4 or 5 hours. Also -- if you're not tweaking the bCLK, you'd think that running the memory at its spec voltage and speed wouldn't require a really long Blend test, if it passed HCI-Memtest(64) or a similar memory testing program for "thorough." "Thorough" in HCI-Memtest is about 7 hours in "DOS" mode from the self-booting CD you can make from the ISO file of the program.

But the point has been made countless times. The sort of games I play mostly load the CPU to a maximum 50%. Normal usage will seldom -- if ever -- encounter the sort of stress you apply with IBT or PRIME. You want to obtain validation of reliable stability in a reasonable amount of time, just to be sure, though.

My experience with an NH-D14 cooler in a pressurized case: I reach my limit for "safe VCORE" voltage before I reach a thermal limit of 72.6C "TCASE." And these are my self-imposed limits.

 

[Zap]

Disclaimer: I no longer aim for maximum hardcore overclock. These days I go for a reasonable and easy overclock using minimal voltages.

I usually run Prime95 for around 15-30 minutes. The reason is that if it is massively unstable due to high clock speed or insufficient voltages it will error pretty quickly. Also, temperatures jump up in the first minute and then almost plateau, but still climbs slowly for a while until it truly levels off. 15-30 minutes should be sufficient for that.

As I said, this is not for maximum hardcore overclocking.

 

[BonzaiDuck]

Disclaimer: I no longer aim for maximum hardcore overclock. These days I go for a reasonable and easy overclock using minimal voltages.

I usually run Prime95 for around 15-30 minutes. The reason is that if it is massively unstable due to high clock speed or insufficient voltages it will error pretty quickly. Also, temperatures jump up in the first minute and then almost plateau, but still climbs slowly for a while until it truly levels off. 15-30 minutes should be sufficient for that.

As I said, this is not for maximum hardcore overclocking.

I pretty much agree on both points. If you can boot into Windows with the OC settings, you would want to find where PRIME95 fails in 10 or 20 minutes. Not too much adjustment from that point would make it stable for hours. My preference, however, is to then validate rock-solid stability over a greater length of time or through a program like IBT or LinX.

OF course, preferences and views vary, but Zap and I choose the same OC'ing objectives . . .

 

[Hulk]

I'm curious as to why the OP paid $100 more for a hyperthreaded CPU just to turn it off?

 

[BonzaiDuck]

I'm curious as to why the OP paid $100 more for a hyperthreaded CPU just to turn it off?

Which of the SB CPUs does not have HT? For my part, I chose to do all the over-clocking with HT "Enabled." But truth be told by many here, it's not that much of an advantage. I might also choose to disable it and try for a slightly higher over-clock speed.

 

[Plimogz]

I have an i2600k running at 4.5GHz at 1.36V[...]

It just hit the 12 hour Prime95 blend test mark, and it ran 100 times through IntelBurnTest without any output disparities and without crashing.

[...]I see a lot of people claiming that their chip is stable after passing three hours of Prime95. I've always thought 24 hours was the default benchmark.

I understand that if I want 100% stable I should run it at stock because that's what it was designed for. But is three hours of Prime95 enough to call stable? Is 24 hours too much?

By the way, I'm using a CoolerMaster Hyper 212+ with the push/pull fan setup. Prime95 doesn't seem to push temps past 65C, but would crash within 10 hours with lower voltages. IntelBurnTest, on the other hand, has pushed my chip up as high as 76C, but has passed 100 tests with voltage as low 1.335V.

That's plenty stable! I'll go as far as to wager you a :beer: that if you download and install BOINC and have it crunch Correlizer for a whole week, it won't return a single invalid erroneous result during that time

And if ever it does, well you'll know it from the log, so you'll be able to tweak that OC of yours. Winning!

 

[Hulk]

Which of the SB CPUs does not have HT? For my part, I chose to do all the over-clocking with HT "Enabled." But truth be told by many here, it's not that much of an advantage. I might also choose to disable it and try for a slightly higher over-clock speed.

Actually only the 2600/2700's (k's included) are hyperthreaded. All the rest of the SB's don't have HT. Just wondering why you'd spend the extra $100 for HT if you're going to disable it. I wanted HT but was too cheap to spring for it.

 

[toyota]

Actually only the 2600/2700's (k's included) are hyperthreaded. All the rest of the SB's don't have HT. Just wondering why you'd spend the extra $100 for HT if you're going to disable it. I wanted HT but was too cheap to spring for it.

the SB i3 and i7 cpus have HT.

 

[Seero]

If 100% stability can be acquired simply by running those, then there will be no need for QA. You can assume the system is stable by default unless you OC. If your system is not stable, then those programs became handy. Again, your system can survive through these programs for days yet still unstable. If the system fails during these tests, than thesystem is not stable. In logical terms, A implies B doesn't mean B implies A.

For overclockers, they need to know if coolings is sufficient, and these programs can ramp up temps quickly. Usually speaking, 1-2 hours of prime will bring case temp to its terminal temp, but room temps may be impacted. Days of burning may heat up not only the room, but the entire house, which represents ambient temp. Only you knows if that is important or not.

I won't run those programs unless there is a problem, it is a waste of time and electricity. If heat is not a problem in summer without air conditioning, than heat won't be an issue in winter.

 

[Rvenger]

I am with Zap, I run Prime 95 and IBT for about 15 - 30 mins. Then after that I start using the computer and run my everyday tasks etc. If its unstable, its bound to show up playing games like BF3 or just surfing around sooner or later. Once I get a BSOD, I read up on the code and scale back on the overclock or increase voltage.

 

[thewhat]

For testing the CPU itself (and CPU voltage), I find LinX/IBT better than Prime. This superiority was most evident with Core2 CPUs, it's less so with i5/i7, but it still finds errors a bit faster in my experience.


But there's a different issue I encountered. I would test with Linx and Prime, both would indicate stability. However, I would then just start some program or resize a window and I'd get a BSOD/crash/freeze.
To fix this I raised the voltage of these: QPI/Vtt, PCH or CPU PLL (don't know exactly which one was decisive).
FYI, the RAM wasn't overclocked.

So besides posting this as a warning, I'd also like to ask which program to test that kind of stability with? Do you have an idea as to which component caused those crashes?

 

[Puppies04]

I have an i2600k running at 4.5GHz at 1.36V with hyper threading disabled since I don't really run anything that takes advantage of it, and the chip seems to run cooler (I know, I know. I could've saved $100 and gone with a 2500k).

It just hit the 12 hour Prime95 blend test mark, and it ran 100 times through IntelBurnTest without any output disparities and without crashing.

I've been a long time lurker, and I've been reading around the web somewhat. I see a lot of people claiming that their chip is stable after passing three hours of Prime95. I've always thought 24 hours was the default benchmark.

I understand that if I want 100% stable I should run it at stock because that's what it was designed for. But is three hours of Prime95 enough to call stable? Is 24 hours too much?

By the way, I'm using a CoolerMaster Hyper 212+ with the push/pull fan setup. Prime95 doesn't seem to push temps past 65C, but would crash within 10 hours with lower voltages. IntelBurnTest, on the other hand, has pushed my chip up as high as 76C, but has passed 100 tests with voltage as low 1.335V.

Run both tests for an hour and keep an eye on temps, if you have no problems then stop stressing and use your machine. If you manage to get a BSOD after you have done this then I would see the need for extended testing but otherwise you are just wasting your time and electricity.

I really don't get why people get so anal about extreme stress testing, the only time I bother is when I know there is a fault and I want to pinpoint it, I don't see many people leaving their cars on their driveways with a brick on the accelerator overnight to to "prove" that the engine doesn't have a fault....

BTW have you tried lowering your voltage slightly? 1.36 for 4.5ghz is acceptable but it is right at the top of what i would expect for a stable overclock.

 

[BonzaiDuck]

Actually only the 2600/2700's (k's included) are hyperthreaded. All the rest of the SB's don't have HT. Just wondering why you'd spend the extra $100 for HT if you're going to disable it. I wanted HT but was too cheap to spring for it.

HT wasn't what I was goin' after when I bought my "K" chip: I wanted the unlocked multiplier. HT was just an extra "bennie."

 

[Puppies04]

Actually only the 2600/2700's (k's included) are hyperthreaded. All the rest of the SB's don't have HT. Just wondering why you'd spend the extra $100 for HT if you're going to disable it. I wanted HT but was too cheap to spring for it.

While I know what you are trying to say you are wrong. All the I3 SB chips are hyperthreaded dual cores

 

[toyota]

HT wasn't what I was goin' after when I bought my "K" chip: I wanted the unlocked multiplier. HT was just an extra "bennie."

that makes very little sense. you could have just bought a 2500k and gotten nearly the same oc out of it.

 

[MacLeod1592]

While I know what you are trying to say you are wrong. All the I3 SB chips are hyperthreaded dual cores

Correct. Its the hyper threading that lets an i3-2100 outperform any AMD proc in gaming benchmarks.

I think stability is relative to the user. If all you do is game a few hours at a time and surf the internet the rest of the time, a 48 hour Prime95 run isnt necessary cause your rig is plenty stable for what you do with it. I run Prime95 for 4-6 hours. If its still running, its stable enough for me cause I never game more than a few hours and Prime95 will push my rig much further than any game ever will.

Thats just a lot of wear and tear on your motherboard so I dont want to run it any longer than it necessary.

 

[FAUguy]

I have an i2600k running at 4.5GHz at 1.36V with hyper threading disabled since I don't really run anything that takes advantage of it, and the chip seems to run cooler (I know, I know. I could've saved $100 and gone with a 2500k).

Are you using just a straight 1.360v or the voltage offset?
On my 2600K I have it set to 4500Mhz and a voltage offset of +0.055v that works just fine.
This produces core voltages around 1.304-1.312 and temps between 62-73C (depending on the core).
I have done Intel Burn Test with Custom 12000 MB and 100 runs with no errors.
I've been using this OC for 3 weeks now and haven't had any issues. I can push the system to 4.8Ghz but requires a voltage offset of +0.095 and the temps get 80-85C, so that's too much for comport.

 

[BonzaiDuck]

that makes very little sense. you could have just bought a 2500k and gotten nearly the same oc out of it.

Jeeez! OK -- I stand "corrected!" I've seen the benchies on both, so I know they're close. I chose to buy what was then "top-end," because it was within $50 of what I usually pay . . .