Prime95 one core failure

[kingfoot]

So my system;

ASUS Crosshair V Formula
AMD Phenom II X6 1090t
G.Skill Ripjaw series DDR3 1600 1.5v
ASUS Radeon HD 6950 DCII/2gb
Corsair 600w Gamer series
Corsair H60 (With added fan for a push-pull radiator system)

The current 100% stable setup;

3.8ghz@200x19.5x1.45v (with a .05+/- load variance)
2000mhz HT
DDR3@1600x1.55v

Core reaches 54c after 24h


I can reach 3.9 with the above settings, and pass a 10m prime95 6 Test run. Not ready to call that 100% stable until a 24h test passes.

Now, before I do a 24h test on that, I want to try to reach 4.0 to run a test and if I can't get that to reach the 2h test, then I will fall back to 3.9 for 24h testing.

Here's the issue;

I get the settings for the 4.0 going. All my HW monitoring shows it running right where I set it so no math errors when in OS. But then I startup Prime95 to run a blend test. About 30-45 seconds in, core04 (5th core) fails test1. All 5 other cores pass Test1 and begin Test2. After about 2 minutes Test2 begins to complete on the cores in random order and sometimes core03 (4th core) fails, sometimes it passes. All remaining cores pass up to Test10 before I stop Prime95.

Is this a physical defect? I read somewhere that upping my vcore is all that's needed. Unfortunately under load at 4.98v in the BIOS will place the .05 variance to nearly 1.55v which I should NEVER go over. So what's going on here? Why can't I hit 4.0 stable? People have reported reaching 4.3 stable with this chip. I realize there are variances and I'm not a lucky person in that regard. But I've yet to see someone with ample hardware and cooling NOT reach 4.0 minimum.

Just a note; the ram is running at the stock settings, manually input by me to ensure their reliability.

 

[kingfoot]

[strike]Is there any way to test an individual core? Can't figure it out in Prime95 settings.[/strike]

Ok I got Prime95 to run on specified cores by setting the process affinity in task manager and adjust the cores to run on to 1 when initiating the blend test. The 2 cores in question pass to Test10. So I'm relatively confident the hardware is fine. Which leads me to believe it's the voltage settings.

 

[kingfoot]

Anyone got anything? The computer temps are great at idle; 27c at the core. Under load even at 1.54v@4.01 it won't go above 54c. It's just that one core keeps failing when all the others are running.

Did another test; set affinity to the first 2 cores (core0 and core1) and it passed to test15, but when I switched it to core3 and core4 (4th and 5th) one of them failed by test5. What's going on?

 

[Rustyy117]

it seems like some of the cores on you CPU aren't as strong as others, happens on my 3930K aswell (core 10 will fail before all others), I expect more voltage would solve it but obviously your running quite high Vcore already. I assume the same thing happens with other stress testing programs like OCCT?

 

[kingfoot]

Could it be that my ram needs more voltage? 1.55v seems pushing it but that or my cpu/nb voltage is also possibly not fast enough. My HT is set to auto which puts it at 2000mhz because my core clock is unchanged. Would I have better results doing a lower ratio and higher core clock? Scaling back my HT and ram speeds?

I know on a black edition I want to change multi and not clocks. But on non-black editions people have to change the clocks to OC at all and they get decent results. Stands to reason that a combination should suffice. Is it possible my PSU isn't up to par? 600w seemed fine when I had my old nVidia 260 superclock. But with the new 6950, this thing is a power eater.

 

[kingfoot]

Also; no, I've just used Prime95. Don't know what OCCT is.

 

[Rustyy117]

If your RAM is rated at 1600Mhz @ 1.5V then 1.55V is more than enough, just make sure your timings are set correctly. Also your HT frequency is fine.

Do you get a BSOD at all? if so what's the bluescreen code?

OCCT is another stress testing program like prime95.

 

[kingfoot]

I don't get a BSoD. I haven't done anything but run Prime95 (and now OCCT) and normal browser/music use while I let the system just be at 4.01 for a little. Prime95 just fails 1 core sometimes 2. Did some more single affinity and dual affinity testing. When testing any 2 cores EXCEPT the 2 that fail, they will run past Test15 fine. When I throw the 2 misbehaving cores in, one of the two will fail.

I ran OCCT for 10m and it passed just fine. Voltage readings were way off in the program but each program is different. The CPU-Z rog flavor for my mobo reads it fine. Concurrent with BIOS readings.

Anyways I still can't figure out what's going on. The failure is a rounding error. Something like getting .5, expecting .4 or less.

 

[Rustyy117]

I've been overclocking my memory recently and always get that error if the memory is unstable, maybe try small FFT's in prime.

Also check your memory timings in your BIOS are correct.

 

[kingfoot]

What's the most reliable way of reading EVERY setting the ram is programmed to be at? My settings are what the BIOS reported, and cpu-z reported. Though some settings are missing. It would be nice to have all programmable settings the ram is supposed to be at.

 

[Rustyy117]

CPU-Z's memory tab will show you the memory amount, frequency, channel and basic timings, if you want to see the more advanced timings then I think you'd have to go into the BIOS, you'll find the CAS latency, speed and voltage on the side of the memory sticks, just make sure there the same as what CPU-Z is reporting.

 

[kingfoot]

Yeah those are all the right setting right now. I'll run the smaller fft's. OCCT Passed on large ftt's though which is odd.

 

[deadjon]

Best way to run Prime, set the FFTs to custom, from 4k to 128k for small FFTs, and set the RAM usage to ~75% (leave some headroom for windows).
I currently run FFTs from 128k - 2048k, they seem to be the killers for my chip.

 

[Rustyy117]

Strange that OCCT is fine, may also be worth fully testing your RAM with memtest.

Let us know how you get on with Small FFTs

 

[kingfoot]

If it's an issue with my ram, will the 1.55v oc void my warranty? They're supposed to have a limited lifetime warranty.

 

[Rustyy117]

How will they know you overvolted it?

 

[kingfoot]

Well a 10 minute small FFT's test ran without a hitch. test15 passed on all cores. Temps well within range. In-fact, I'm going to reduce the voltage as it doesn't need to be this high for the small test. Indicating my cpu is stable, but my memory is causing the hiccup?

 

[Rustyy117]

I wouldn't like to say for sure, but its looking like it.

 

[kingfoot]

Actually, looking back at the log looks like 1 core had an "Illegal sumout". 0 errors, 100 warnings. Didn't see it at first in the completed window.

 

[Rustyy117]

I dont think OCCT has round off checking so that maybe why it passes, :/ seems like its not completely stable on small FFTs either then. I'd still run memtest for a few hours just to eliminate it from the equation. or you could try underclocking the RAM.

 

[kingfoot]

Ok running me test right now. Not a single error and I'm more than 60% through after an hour ten. Takes a lot longer with 16gb huh? Haha

researching a little shows my timings are crystal in the bios for my ram. I just can't figure out how .5x causes so many problems. Also noticed after letting the computer just sit overnight at 4.01 that when I ran the small fftS test, it got Much hotter. From 54c to 63!!! I halted pretty quickly cause that was alarming. What would cause such a drastic heat degradation when I changed nothing at all??

Note: I had a log going for temps and all night wi no activity it was at nearly 26c with little to no change. So can't figure out why it would jump up so much compared to the blend test.

 

[kingfoot]

So the test took 2 hours and completed with no incident. My ram seems to be ok at the clocks that it should be set at according to the manufacturer. I still don't understand why I'm getting these kind of readings. The minimum voltage to be stable at 3.81 is 4.62500 in the bios with a .03 variance. The minimum for 3.9 "stable" (not 24h but 2h stable) is 1.472 so it stands to reason that 1.49/1.50 should be proper voltage to be stable. But it isn't as far as Prime95 is concerned. Really don't understand what's up.

 

[kingfoot]

As a side note; I went back to stock and was able to run 100% stable at only 1.33v with a .02 variance. My temps were AMAZING. 25c min 41c max. Really love this cooler, and as the clocks and cores go up with the build and size going down. I expect this cooler will last me through the end of the AM3+ series which is just getting off the ground! All in all I'm happy with my performance and heat. I will take what I can get and call it good. Afterall, a successful overclocker knows when the limit is reached, right? Then again, I have yet to move to clock tweaking. Any good guidance for taking what I have to make it better through clock tweaks?

 

[kingfoot]

Hey, not to revive a dead topic but...

I had a thought. Could my PSU simply be not enough power? I did a few of those PSU Calculators around the web. Turns out my PSU's peak is about right where my computers' draw is. Could it be just stressing it out too much? Trying to pump the power to the CPU when testing/load, not having enough for the memory causing the symptom of failure because the memory? Could the choke really be the PSU?

 

[rage33]

[strike]Is there any way to test an individual core? Can't figure it out in Prime95 settings.[/strike]

Ok I got Prime95 to run on specified cores by setting the process affinity in task manager and adjust the cores to run on to 1 when initiating the blend test. The 2 cores in question pass to Test10. So I'm relatively confident the hardware is fine. Which leads me to believe it's the voltage settings.

Adjust your LLC (load line calibration) in your bios settings to very high or ultra, I had a similar problem where two cores were showing errors, but my computer wasn't crashing. LLC with help with VDroop and fix your cores from showing errors