Annoying prime 95 problem

[bob4933]

TL;DR -> core 8 failed on p95 at 38 minutes out of a 2 hour test. No other problems noted. Error was .5 instead of .4, all other benchmarks and tests went fine.


Ok, dinked around all day... got a seemingly "stable" setup going.

Intel Burn Test at very high on 7 tests (25 minutes), stayed under 64c. At around three tests it stabilized around 61c.

Geekbench ran great, http://browser.primatelabs.com/geekbench3/321745 pretty solid for me

Idle temp is higher than I like @ 34c, maybe its just heatsoaked and will cool down all the way after it sits idle for a while.

I did an initial 10 minute prime95 to start things, went flawlessly, however I decided to do a nice long stability test. started a movie, turn on p95, and let it go... at around 38 minutes into the test, core eight got a "rounding error" (.5 instead of .4?). Im unsure what that meant, so I let it keep going. Movie was about 2 hours, temps reached 67c, but averaged low 60's.

Its the only test thats causing any kind of errors, and I don't know how to even go about "addressing" it. Any ideas or suggestions would be terrific.

 

[bassnut]

Prime95 is a different beast...... we tend to use it here to assume stability as of all the programs we have available it is the one that pushes the hardest for a couple of us that really need our rigs to be 100% stable. There is plenty of differing opinions on this approach and each individual can make their own minds on that themselves.

My guess is you are getting an ILLEGAL SUM OUT error which generally means you need a bit more VCore ...... bump your CPU voltage a bit and test again. Keep an eye on your temps with HWMonitor .....ignore your Idle temps as AMD has done a crappy job of calibrating the idle temp CPU Socket (CPU in HWMonitor) max temp of 70*C and Core (package in HWMonitor) temps max of 62*C is what we recommend for max temps.

 

[bob4933]

Now, with a lets call it 'mostly stable' overclock, if I add voltages, its creating more heat obviously. If im having no other issues besides that small error, should I ignore it since Im only doing mild gaming or should I keep at it to get this "100% perfect overclock"? Im honestly in the dark with this, my practical side is saying leave well enough alone, and my OCD side is telling me to fix it.

 

[Luke1978]

I would game at that OC for a few hours and see if it's stable in the game. I don't know of a game out that will push your system as hard as P95. BF4 is close, but it's not 100% load non-stop.

I've had problems in games nowhere near as intensive as BF4 on a system that was 24hrs prime stable. Sometimes it's not how much usage the game spikes, but the manner in which it does it.

 

[RGone]

When I am tuning my overclock for stability, I find that I can "get stable" with IntelBurnTest using about 0.02 LESS Vcore but that to get stable by my water-shed of P95 Blend for at least 2 hours error free; I have to up the Vcore depending on the error in P95.

So you have a few choices.
1. Up Vcore and test with P95.
2. But you say raising Vcore will make more heat. Sure it does.
3. Lower overclock to pass P95 Stable since the greater overclock is not stable.
4. Increase cooling capacity.
5. Call your rig stable enough for government work and carry on.
6. About it.
RGone...

 

[bob4933]

Ok, I tried raising CPU voltage, and got the same error. I went all the way up to 1.4 volts, still same error. I brought it down to 4.3ghz and lowered the RAM speed, and it appears to be working. Ill report back I suppose.

That said, do I just keep adding voltage until it works? When do I add CPU/NB or NB voltage?




Ok, I was looking around at the error type Im getting, and its possible its my RAM causing the problem not necessarily the cpu overclock.

That said, my ram is terrible so it makes sense to me. (633mhz ram ) If this is a possible fix, and you guys would recommend it, I can go ahead and fix that one.

That said, my system ran BF4 for a couple hours last night with no issues at all, heat was stable (under 53c) and the system was left on over night without any issues at all.

 

[twocents]

One core, usually the same one (mine is 7) gives '0 errors, 100 warnings' and doesn't turn red? Only happens on blend test?

 

[bob4933]

Let me guess. One core, usually the same one (mine is 7) gives '0 errors, 100 warnings' and doesn't turn red? Only happens on blend test?

nope,

It says

1 error, 0 warnings
Rounding Error 0.5, expecting 0.4

No idea what it means.

 

[Johan45]

OK bob if you post up a screen shot of HWMonitor (non pro) while P95 has been running for about 20 minutes so we can see volts and temps. Also CPU-z main, spd and memory tabs we can have a look at what's going on. Gives us a lot of info and it's about as close as we can get to your keyboard.

 

[bob4933]

OK bob if you post up a screen shot of HWMonitor (non pro) while P95 has been running for about 20 minutes so we can see volts and temps. Also CPU-z main, spd and memory tabs we can have a look at what's going on. Gives us a lot of info and it's about as close as we can get to your keyboard.

Right on. Gimme a bit.

 

[bob4933]

As I said, my temps are more than adequate. I ran this for almost 4 hours last night, temps never get unreasonable. Just the one core will fail around 45 minutes into it, but the rest of them are 100% fine. Makes no sense to me. I feel like I can get much higher than 4.6ghz, but the error is obviously not proper. Changing the voltages seemingly make no difference.


edit: I just noticed, the .924V, is that voltage drooping? Can that be causing my problems?

 

[bob4933]

hour and a half of blend, no errors. I would really like to clock back up, if someone can give me some info. Do i just increase core voltage, or NB or cpu/nb... Im not really sure what Im doing honestly.

 

[keny]

Yep just raise the CPU multiplier until it fails p95, then add vcore to stabilise the oc, keep doing this until it gets too hot again, as I don't think your cooler could take the recommended max vcore of ~1.5v, leave all other voltages on auto for now

 

[bob4933]

Yep just raise the CPU multiplier until it fails p95, then add vcore to stabilise the oc, keep doing this until it gets too hot again, as I don't think your cooler could take the recommended max vcore of ~1.5v, leave all other voltages on auto for now

Thats the thing, if I bump up the multiplier 1 time, no matter what the voltage is, it will fail the 8th core every time.

 

[keny]

Go into your bios and change the percentage rate to core 8 to slightly lower than the others, maybe down to something like -2% and see if that helps with the load on that core, it will be under the clock calibration heading or ACC you should be able to adjust the amount of load to each core, but a usual rounding error is normally fixed with more vcore, what oc are you running when it fails?

 

[bob4933]

I am finishing up a project real fast, Ill update everything in my sig when I get a minute thanks guys.