Haswell-E impossible to get it prime95 stable with a reasonable OC?

[joeyjojo]

^ I'm sure the people who run it 24/7 searching for prime numbers would be interested to learn that.

 

[tyler_jrb]

I've not known a 5820k only overclockable to 4.1ghz. Most will do at least 4.4-4.5ghz. Unless you have a really poor chip.

If your wanting to use yours for general usage, gaming, compress, render etc. Just use real bench. It uses handbrake in its stress test.

Other than that using prime95 is pointless in your situation.

 

[Stealthploit]

Do very small changes each time... also some times the motherboard ur using has a huge impact

 

[ebilcake]

I've not known a 5820k only overclockable to 4.1ghz. Most will do at least 4.4-4.5ghz. Unless you have a really poor chip.

If your wanting to use yours for general usage, gaming, compress, render etc. Just use real bench. It uses handbrake in its stress test.

Other than that using prime95 is pointless in your situation.

I've tried two and both struggled just to run at 4.2GHz, they ran hot.

 

[8 Pack]

Use the PC for what you actually use it for day to day as a stability test.

Stable for what is the key?

I have seen several machines run this kinda stuff for 24hrs then crash when opening running a game.

 

[Deleted member 66701]

Don't use Prime95 for stability testing, use the computer for what you will actually use it for and stability test against that usage. This is why Asus RealBench is a nice stability test as it uses real world applications.

If you search for prime numbers, then sure, make sure Prime95 is stable. But if you don't, then bin the app - it's nigh on useless.

^ I'm sure the people who run it 24/7 searching for prime numbers would be interested to learn that.

How many people OC'ing their CPU's actually do that?

 

[Vish Petrol]

Just run aida 64 for an hour, then open some games, and run applications that you would normally use.

 

[joeyjojo]

Don't use Prime95 for stability testing, use the computer for what you will actually use it for and stability test against that usage. This is why Asus RealBench is a nice stability test as it uses real world applications.

Does RealBench do any error detection? Or does it just crash? Genuine question, as the FAQs don't say.

If it doesn't then it may well be a good stress test but an almost useless stability test, as errors could be silent.

How many people OC'ing their CPU's actually do that?

I don't know. There are 156,170 people working on GIMPS (the internet prime search), some of them are presumably overclocking.

 

[titaniumx3]

I ran IBT for 3hrs on max stress levels, hit 89c but no errors. I then ran x264 stability tester and it crashed within 2 min. Tried many things to get it stable but eventually had to reduce overclock by 100mhz (which was bizarre in itself). This made it stable even after 4 hrs of x264. So then I moved on to realbench just for laughs, but almost started crying when it crashed after 10min. Bumped up voltage and now it's stable for 4hrs.

I am still totally lost as to what is stable and not stable, but through out this entire process I've not had a single game crash or error in anything else other than the aforementioned stability testers.

In my opinion, if you really want rock solid stability, you need to test with a wide variety of programs. I don't think p95 or IBT give you complete assurance, despite being the hardest, there's more to it than that.

 

[Danny75]

I ran IBT for 3hrs on max stress levels, hit 89c but no errors. I then ran x264 stability tester and it crashed within 2 min. Tried many things to get it stable but eventually had to reduce overclock by 100mhz (which was bizarre in itself). This made it stable even after 4 hrs of x264. So then I moved on to realbench just for laughs, but almost started crying when it crashed after 10min. Bumped up voltage and now it's stable for 4hrs.

I am still totally lost as to what is stable and not stable, but through out this entire process I've not had a single game crash or error in anything else other than the aforementioned stability testers.

In my opinion, if you really want rock solid stability, you need to test with a wide variety of programs. I don't think p95 or IBT give you complete assurance, despite being the hardest, there's more to it than that.

On Adaptive, or Auto (which is often a type of Adaptive by default), P95 and IBT add 0.1v to Vcore and that helps pass those tests. Realbench doesn't add much if anything at all to Vcore, even on Adaptive, so sometimes it's not enough to pass it. Of course, there could be a number of other reasons for each case, including yours. I don't know if your Vcore is Fixed or not.

I've had similar experiences at times. Can be frustrating for sure.

 

[titaniumx3]

That probably explains it then, as I am using adaptive voltage mode. Still a bit strange though. After I got my system IBT stable, I used the computer for a few days, played games for several hours with no issue, but soon as I ran the x264 stability tester I got a bsod almost immediately. I tried increasing the voltage slightly but no luck. In the end I dropped OC from 4.8ghz to 4.7 and it was perfectly fine. I'm fact I was able to drop voltage a fair bit too in the process.

 

[Silent_Scone]

This thread makes me sad. Stop running Prime, you are killing the CPU. The current drawn from Prime 28+ AVX 2.0 routines will degrade your CPU. You've probably done a fair bit already.

I think you're a bit confused.

I don't know about aida or realbench, but recent Intel Burn Test (or linx) uses AVX. If used properly it should get hotter than anything else, even Prime95 small FFTs. It's very sensitive to the problem size so you need to experiment with that.

As for "Is OP using his PC for AVX? I doubt it" - plenty of applications are compiled with AVX(2) extensions. Handbrake (specifically the x264 encoder) is one. It's highly optimised and will get things hot. The problem with using it as a stability test is it won't flag any errors (other than catastrophic ones), unlike Prime95 or IBT/linx.

If you think x264 is a "realistic" load then so are IBT and Prime95.

He's not as confused as you might think, actually. Less so than you. Prime 95 AVX 2.0 routines on the version he is running draws more current than any other test or even professional application. EP processors will reduce speed and voltage automatically when it detects these types of instruction sets for this very reason. The current drawn can be in excess of 400W with the right (or wrong) overclock.

So if you want to keep giving that kind of advice without doing the proper research first, you carry on.

X265 4k bench is a good stability test for most workloads, but not brutal. AIDA FPU test also has a reasonable amount of AVX2.

 

[joeyjojo]

^ This is exactly the kind of muddled thinking that seems to have become the norm in this subforum.

Prime95 kills CPUs is a myth. Any voltage and heat degrades microprocessors, not one program or another.

Some codes are better at utilising the vector units on modern CPUs (AVX/2 being the widest) so obviously these get the CPU hottest because such a large % of the die is now devoted to them.

IMO if you're not stability testing with the greatest computational load you can (to utilise as much of the CPU as you can), then why bother? When a load comes along that's greater than the testing you've done it'll fail.

Specific issues with what you've written:

The current drawn from Prime 28+ AVX 2.0 routines will degrade your CPU.

This is true for all programs. Efficient codes just get hotter than others.

Prime 95 AVX 2.0 routines on the version he is running draws more current than any other test or even professional application. EP processors will reduce speed and voltage automatically when it detects these types of instruction sets for this very reason. The current drawn can be in excess of 400W with the right (or wrong) overclock.

Literally complete nonsense.

X265 4k bench is a good stability test for most workloads, but not brutal. AIDA FPU test also has a reasonable amount of AVX2.

What is a "reasonable amount of AVX2"? Have you profiled it?

 

[Silent_Scone]

Anything that will get through to you will likely result in a ban on my part. You are completely wrong.



All people need to know (for their own good and for those who aren't chuckleheads), is that Prime95 with small FFTs specifically will cause damage to the CPU. This isn't a myth, this is specific to this platform and these CPU. People don't seem to be aware of the current this test generates when the CPU is overclocked. Which is evident by the fact you keep talking about heat and voltage which is exactly why you are lost.

 

[joeyjojo]

Sorry, voltage was a typo, I was thinking of power consumption models not MTTF.

This isn't a myth, this is specific to this platform and these CPU. People don't seem to be aware of the current this test generates when the CPU is overclocked. Which is evident by the fact you keep talking about heat and voltage which is exactly why you are lost.

So you're saying this specific software on this specific platform "will cause damage". Why isn't it a problem for socket 1150/1151? Why not other efficient AVX codes?

Also are you saying heat isn't a factor in TTF?

Edit: what I'm getting at is you can't pick out Prime95 as 'dangerous' when there are so many codes that are similarly intensive. If Prime95 is off limits then you'd better throw out all the other mathematical solvers, video encoders, raytracers, compressors, etc.

Also Intel would have to alert people like me that we better not utilise all the vector units in their CPUs for fear of damaging them.

 

[Silent_Scone]

I'm not picking out anything, this is just the fact of the matter. I've seen numerous users with moderate overclocks on 5820k degrade both cache and core as a result of hours of Prime. It comes down to the routines involved in Prime and Prime alone, most other application that use this extension do not come close. How you aren't able to discern this difference is probably part of why this is doesn't sit well with you.

 

[zipp0r]

I'm not picking out anything, this is just the fact of the matter. I've seen numerous users with moderate overclocks on 5820k degrade both cache and core as a result of hours of Prime. It comes down to the routines involved in Prime and Prime alone, most other application that use this extension do not come close. How you aren't able to discern this difference is probably part of why this is doesn't sit well with you.

so there we have it. p95 the cpu killer!!!!!!!!111 not excessive heat or vcore!!!11 nevermind that ibt or occt also do the same thing.

 

[Silent_Scone]

so there we have it. p95 the cpu killer!!!!!!!!111 not excessive heat or vcore!!!11 nevermind that ibt or occt also do the same thing.

So to clarify, you think Prime 95 has the same AVX 2.0 routines as the two latter tests? This place has really gone down hill.

Maybe this will help:

http://www.overclock.net/t/1510388/haswell-e-overclock-leaderboard-owners-club/2390#post_22900116

By all means, keep on priming...

lol, ffs.

 

[setter]

I haven't tried p95 on my 5820k. But having owned a 4770k and x2 4790k's. Even at stock under a high end air cooler. Temps on all three chips were into the mid 90's within 2 seconds of starting small fft test on prime ver 28.5.

 

[Justintime]

Use v26.6 for Haswell as 28.7 uses AVX and heats up more and requires more voltage to be stable.

On my 4770K @ 4.5 temps hit 100C within minutes on 28.7 as the voltage required and heat created is much higher than any other application I run.

Isn't that like lying to oneself? Giving the cpu an easy test just to see a pass? I have built around 100 Haswell rigs and delidded about 70 chips so far, 40% or so razor and 60% vice, that tames the temp issue once used with a proper cooler. I can overclock about 2-300mhz more in general if I ignore the Elephant in the room that AVX 2 prime can't pass but I don't. Have not seen a single case of degradation and can't say I've heard of any, also always manually set the voltage. I get that with Haswell-E you can't delid but even my 5820K passes prime @ 4.5Ghz 1.295v with temps under 88C, with a Noctua NHD14 and rather aggressive 110CFM fans that are silent in normal operation, can do 4.7Ghz with old prime and never break 79C testing but oddly a few games do lock up occasionally even with 1.3v+, never does that at 4.5 1.295