haswell z87 freezing prime 95 in memory tests

[oneofusjustin]

4770k
sabertooth z87
16bg trident x 2400

i have a somewhat stable 4.7 ghz over clock it can run pcmark and 3dmark and a prime 95 blend overnight. but when i run a custom blend and scale the memory used to 80% im freezing the entire system. no BSOD just a hard lock up. ive raised the vcore even more to 1.39 and even clocked the memory down to 1600 upped dram voltage to 1.7 even dropped the oc to 4.5 and its still locking up in under 18 min with temps under 72c

is the a better memory test for haswell memory controller than prime 95?

 

[SOFTengCOMPelec]

You could try memtest86.
It's not necessarily better than prime95, or perfect.

But if it finds a problems, it will be more likely to identify it (as a memory issue).

From what you describe, it sounds like a memory issue to me, but I am not 100% sure.

Here it is

 

[oneofusjustin]

usually a OC issue will BSOD or reset the box. but havent really overclocked for a few generations. Ill try memtest... but thats more of an error checker isnt it? does it detect stability issues or just defective memory? Im running the aida 64 memtest with 9 gigs and its been going fine for 45 mins which is several times longer than prime 95, but not as much memory used

 

[SOFTengCOMPelec]

The memtest86 would be checking the memory, WHILE the cpu/memory are "hurt" by your overclocking settings, such as voltage and frequency.
So memtest is NOT specifically trying to overheat/overload your memory, but it will tell you if it can reliably read/write to memory, while under the overclocking settings you have applied.
If it was me I'd be backing off the overclocking, or even worried that my hardware is a bit flakey (needs to run at all=stock, before assuming hardware faults).
As I would want all prime95 tests, and anything else to pass ok, BEFORE saying that the overclock is ok.
Even then, I would back things down a lot, to maximise system stability and life expectancy of components/reliability.

 

[oneofusjustin]

no point in dialing back until you firmly establish the edge if it was crashing out in other applications or tests i would simply dial back. But when nothing gives it problems besides prime 95 which is a non signed app then i have to keep digging at it. but it is just annoying because i want it to pass everything and right now no mattter what i throw at it prime 95 with memory loaded to 14-15 gigs just wont play nice.

 

[oneofusjustin]

7 instances of memtest with 14.8 gigs total testing is going fine 0 errors 60% coverage so far

 

[SOFTengCOMPelec]

But when nothing gives it problems besides prime 95 which is a non signed app

Usually the reasons that causes prime95 to be the only thing giving trouble, is because prime95 is so good at detecting issues, in the first place.

Also prime95 tends to get the cpu to a much high temperature than most other software, which also tends to make problems much more likely.

(Take the following as super rough and highly inaccurate) If prime95 causes/detects one error every 30 .. 60 minutes, then you probably would have to do 5 .. 100 hours of normal activities to get each BSOD/error/problem.

I.e. It is saving you testing time and hassle.

 

[sm625]

You have to dial back the OC until it can run prime95 without freezing. Who knows, maybe it will do the same things at stock. Maybe your memory controller cannot run the RAM at your timings when it is beyond a certain temperature. In which case you'll never see the problem unless you run prime95 or are in the middle of a heated battle in a very cpu intensive game. (And that is not when you want it to fail)

 

[oneofusjustin]

i was 50 lines passed in LinX max settings when i left for work and i ran OCCT with AVX and 90% memory last night a couple hours... if the linX makes 100 passes ill just chalk it up to prime95 being strange

prime 95 is freezing on the following test... if anyone knows what this test stresses let me know

Test2, 800000 Lucas-Llehmer iterations od m172031 using FFT Length 8k

 

[SOFTengCOMPelec]

The most likely cause is that you have set the overclock TOO high.
If you think your overclock is 100% fine and dandy, and you now have the most reliable and stable computer ever!
And you want to blame prime95, then we will have to agree, to disagree.

 

[Rvenger]

Haswell is real goofy with prime 95. I can be stable in the small fft for hours and then move over to blend and it will BSOD 0x124 within minutes. AVX requires more voltage to be stable on Haswell so you will need to drive up your voltage more unfortunately. Its not a prime issue unless you are using the wrong version.

 

[oneofusjustin]

Haswell is real goofy with prime 95. I can be stable in the small fft for hours and then move over to blend and it will BSOD 0x124 within minutes. AVX requires more voltage to be stable on Haswell so you will need to drive up your voltage more unfortunately. Its not a prime issue unless you are using the wrong version.

thats how mine is, have you figured out how to get it stable in prime 95 blend? i dont see how its not a prime 95 issue if its the ONLY test that makes the thing freeze..and its only when i do the custom blend and max the ram out. every other crash ive had is a BSOD. people say LinX ans linpack stress the cpu beyond anything possible outside of syntheitcs so if i can pass them no problem i dont see how im not stable, just because 1 prime 95 test fails in a software lockup, not bsod

 

[2is]

Does it lock up when everything is set to default values? If it doesn't, it's certainly not a software issue.

 

[Idontcare]

thats how mine is, have you figured out how to get it stable in prime 95 blend? i dont see how its not a prime 95 issue if its the ONLY test that makes the thing freeze..and its only when i do the custom blend and max the ram out. every other crash ive had is a BSOD. people say LinX ans linpack stress the cpu beyond anything possible outside of syntheitcs so if i can pass them no problem i dont see how im not stable, just because 1 prime 95 test fails in a software lockup, not bsod

This debate is extremely simple to settle - if it is a software-induced issue then it will happen regardless the OC values for your hardware. (it will happen at stock, at under-clocked clockspeeds, and at over-clocked clockspeeds)

So do you get the exact same error when you run Prime95 blend the exact same way when the CPU and hardware are all set to stock?

If yes, then you *might* have a valid software issue on your hands (further partitioning steps and verification are necessary because some hardware can be bad even at stock)...but if the problem goes away as you relax the OC clockspeeds and go towards stock then you most certainly do NOT have a software bug/issue on your hands.

 

[oneofusjustin]

This debate is extremely simple to settle - if it is a software-induced issue then it will happen regardless the OC values for your hardware. (it will happen at stock, at under-clocked clockspeeds, and at over-clocked clockspeeds)

So do you get the exact same error when you run Prime95 blend the exact same way when the CPU and hardware are all set to stock?

If yes, then you *might* have a valid software issue on your hands (further partitioning steps and verification are necessary because some hardware can be bad even at stock)...but if the problem goes away as you relax the OC clockspeeds and go towards stock then you most certainly do NOT have a software bug/issue on your hands.

no your 100% right... i have had this rig built since Saturday and have done nothing but stress test it.. its so frustrating especially when the errors can take so long to find. i dialed back a bit and was able to make it to 8 hours instead of 15 minutes in the prime 95 custom blend and ... so there is progress.

but in all actuality if it takes 8 hours to find an error at 100% load i think ill be okay just gaming, right?

 

[Rvenger]

thats how mine is, have you figured out how to get it stable in prime 95 blend? i dont see how its not a prime 95 issue if its the ONLY test that makes the thing freeze..and its only when i do the custom blend and max the ram out. every other crash ive had is a BSOD. people say LinX ans linpack stress the cpu beyond anything possible outside of syntheitcs so if i can pass them no problem i dont see how im not stable, just because 1 prime 95 test fails in a software lockup, not bsod

I went from 1.15v to 1.20v and it was perfectly stable. So I increased it .05v to give a buffer for AVX workloads.

 

[oneofusjustin]

I went from 1.15v to 1.20v and it was perfectly stable. So I increased it .05v to give a buffer for AVX workloads.

lol im alot higher than 1.20v thats the thing thats so frustrating, OCCT has AVX and im stable in that for over night.... its just prime 95 thats giving me shit in custom blend tests

 

[2is]

but in all actuality if it takes 8 hours to find an error at 100% load i think ill be okay just gaming, right?

Wrong... Different apps stress the processor in different ways. What took 8 hours on prime may never happen in a game, or it may happen a lot sooner. You're also opening up yourself to silent data corruption since you're clearly not stable. If your temps are good, increase voltage. If they're up there, decrease your overclock. It's not worth turning your computer on one day and your OS fails to boot because of data corruption. As for me personally, it would drive me crazy if every time something out of the ordinary happened, I'm always questioning if it's because I'm not stable. Not to mention, failing after 8 hours is still pretty poor. Maybe not atrocious like 15 minutes, but certainly poor.

 

[Rvenger]

Are you using manual, offset, auto, or adaptive voltage?


EDIT: Noticed you are trying to achieve a 4.7ghz OC. Good luck because your temps got to be sky high and most haswell chips can't even handle 4.7ghz. Some chips have a very weak uncore and IMC at higher clocks. What is your Uncore/cache ratio?

 

[Deders]

Could be something to do with the timings, but if it passes occt I'd be happy.