Sign in with
Sign up | Sign in
Your question

Prime95 failure on Intel Core i7 960 Quad 3,2Ghz with ASUS P6X58D-E

Tags:
Last response: in CPUs
Share
September 17, 2011 2:06:27 PM

I was merely trying to stress my cpu with Prime95 to figure out the max temperature at heaviest workload, but I got an error about 4 or 5 minutes in there;

FATAL ERROR: Rounding was 0.5, expected less than 0.4

I've looked around a bit, and it seems that most likely this has something to do with my CPU Core Voltage? I'm a bit afraid to modify those values, and it's set to auto (which is 0.944v... isn't this low?). Can i increase it? And what will happen to my cpu temp in that case? It was 78 celcius when the error occurred (didn't seem like it was going to increase more... pherhaps just a a couple of degrees more). I took the test "in place large ffts".

Thought my system was okay by now... I've recently flashed the BIOS to the latest version, and downgraded from 12 to 6 gigs of RAM (because of memtest failures and BSODs), and adjusted memory timings and voltages. Memtest runs fine (just tested it for 1 pass though).

This is the rest of my pc-specs:

ASUS GeForce GTX 560Ti 1GB PhysX CUDA

2x Samsung 1 TB 7200RPM S-ATA2 32MB

OCZ 60 GB SSD 2,5" Vertex 2

Mushkin DDR3 6GB Blackline Black FB
PC3-12800 3x2048MB 1600Mhz. (9.9.9.24)

Intel Core i7 960 Quad 3,2Ghz
8MB, BOXED m/vifte 4,8GT/sec CPU

ASUS P6X58D-E, X58, Socket-1366

CMPSU-750TXV2EU
Corsair power supply, TX750W

This is my CPU-Z Dump:

http://valid.canardpc.com/show_oc.php?id=2003567
September 17, 2011 2:31:19 PM

I don't know if it is of any importance or not, but the vcore increased to 1.23 when I stress-tested it again (with BLEND test - some of everything). Anyway, it failed after about 4 minutes (CPU max temperature is 81 degrees celsius under this workload/test btw)
September 17, 2011 2:48:42 PM

Abstinent said:
I don't know if it is of any importance or not, but the vcore increased to 1.23 when I stress-tested it again (with BLEND test - some of everything). Anyway, it failed after about 4 minutes (CPU max temperature is 81 degrees celsius under this workload/test btw)

the cpu might be overheating as well as your ram, do you have the stock cpu cooler, if so tell me your temps while running p95 and you can ease off your ram a little, to 1333mhz with the same latency and see what happens
Related resources
September 17, 2011 3:17:31 PM

megadelayed said:
the cpu might be overheating as well as your ram, do you have the stock cpu cooler, if so tell me your temps while running p95 and you can ease off your ram a little, to 1333mhz with the same latency and see what happens


I don't know what kind of CPU cooler I have, to be honest. I had the PC built by the retailer/store I bought it from (and I'm beginning to wonder if this was a big mistake). Anyway, it just says "Intel" something, on top of the CPU fan. So I'm guessing stock cpu cooler? Anyway, I took the small FFTs test with Prime95 as well, and I ran it for about 5 minutes without any errors. CPU temp about the same. I just read that if this test does well, it probably doesn't have anything to do with the CPU (know I should've run the test longer, but some of the threads/cpu core tests had stopped and restarted, and I've just wanted to see if anything happened about the same time it did with the other tests). So I ran Memtest again, and it failed :/  So I guess my memory is the fault after all, as it were to begin with.

Thanks, I'll try to ease off to 1333 mhz, but what should the QPI/DRAM voltage and DRAM Bus voltage be set to when i decrease the frequency to 1333 mhz? They are respectfully at 1.35 and 1.65 (is this a bit high?)
a b à CPUs
September 17, 2011 3:20:07 PM

The voltage is a little low at load, but you might just have a good chip. What is your QPI/Dram voltage? You can try setting the QPI/Dram to 1.300 and test again. If that works, lower it one notch until you get an error than raise it back one.
September 17, 2011 3:29:56 PM

Cygnus x-1 said:
The voltage is a little low at load, but you might just have a good chip. What is your QPI/Dram voltage? You can try setting the QPI/Dram to 1.300 and test again. If that works, lower it one notch until you get an error than raise it back one.


1.35 atm :)  But I'm really freaked out about adjusting the voltage - either on the CPU or RAM. Sure it's not going to harm anything on my pc? I've got a bit of electrophobia or something :p  :S
a b à CPUs
September 17, 2011 3:39:13 PM

LOL@ electrophobia...Your not touching the cpu voltage and your just setting your ram voltage to it's factory specs. 1.3 is more than safe on the QPI/Dram, in fact my ram recommends 1.35 qpi and 1.65 volts for the ram. I did some testing and have mine stable at 1.28750 @ 1667ish. Try 1.30 and work your way down from there, don't worry it's safe :) 
September 17, 2011 4:32:21 PM

I adjusted the frequency to 1333mhz and sat the voltage to 1.332 (or something) for QPI/DRAM, and 1.62 for DRAM Bus voltage. And stress tested with Prime95 for about 20-25 minutes with zero errors and a max temp of 77 (though mostly 76) :)  Also ran a Memtest for 1 pass without any errors. Acceptable! It seems thought that things are running a bit slow, but maybe it's just my internet connection, or my pc is tired form stress-testing :p 

Anyway, how long should I run Memtest and Prime95 to ensure there's not any errors? 24 hrs with Prime95, and during the night with Memtest? And can I lower the voltage even more to make my system more stable? And I do have 12 gigs, but as previously mentioned I took out 6 of them to make my system more stable. Though, as it seems it works for now, can I insert the rest and try with this timings and voltages? Or should I make some adjustments to timings/MHz/voltage?

And a final question - do I really need more than 6 gigs RAM for any application or game?
a b à CPUs
September 17, 2011 8:40:42 PM

I would run memtest for two or three runs, then blend for like 4 hours. Some say run prime for 8 hours minimum, but i think that's overkill. You might want to run prime95 in smallfft for a while as well just to make sure the cpu is stable. i would try putting the rest of the ram back in, can't hurt. It's always a good idea to find the lowest stable voltage setting, so yes.

You might not need 12 gigs unless your using large spread sheets, or cad work, video editing, things like that. It's nice to have though, and your board should handle it, just might take some tweaking.
September 17, 2011 9:32:41 PM

Cygnus x-1 said:
I would run memtest for two or three runs, then blend for like 4 hours. Some say run prime for 8 hours minimum, but i think that's overkill. You might want to run prime95 in smallfft for a while as well just to make sure the cpu is stable. i would try putting the rest of the ram back in, can't hurt. It's always a good idea to find the lowest stable voltage setting, so yes.

You might not need 12 gigs unless your using large spread sheets, or cad work, video editing, things like that. It's nice to have though, and your board should handle it, just might take some tweaking.


Not anyone of those things really, though I might just be needing it for video editing someday :)  For now, the resource/ram hogs are games, but as far as I can tell, none of the current games require more than 6, even at the highest setting. Anyway, as you say, good to have onboard :) 

I am currently running memtest as we speak, and it's up to 409% coverage with no errors :)  . Thought I'd try to run the boot-version at night :)  And as you suggested I'll try blend for 4 hours or more, and smallfft for a while :) 

Thank you very much for the help! :D  Been going a bit crazy over this for the past few days... invested a lot of money in this computer, and want it to run smoothly of course :) 

Just a tiny question at the end though - does it have/should it be a ratio/fixed voltage difference between QPI/DRAM voltage and Dram Bus voltage?
a b à CPUs
September 17, 2011 10:08:57 PM

Abstinent said:
Just a tiny question at the end though - does it have/should it be a ratio/fixed voltage difference between QPI/DRAM voltage and Dram Bus voltage?


You just have to keep them within .5v or less of one another. So if your ram voltage is 1.65, the qpi/dram voltage should be at least 1.15. You easily have that covered, so your good.
September 17, 2011 11:44:44 PM

Cygnus x-1 said:
You just have to keep them within .5v or less of one another. So if your ram voltage is 1.65, the qpi/dram voltage should be at least 1.15. You easily have that covered, so your good.


Yeah. those can't be that far apart anyway :)  Thanks again for helping, both to you and megadelated. you've made my computer and gaming life way much easier :D 
September 18, 2011 2:29:01 AM

Ok, I'm not going to claim that this is my last question :p 

I've tested it some more, and it passed small ffts with no errors (ran it for an hour). Then i tested inplace large ffts, and I got an error in worker#8 after 15 minutes (same as before). So then it's safe to say it's memory? Anyway, I'll try to lover the voltage some more. And btw, since I've lowered the frequency to 1333mhz, can i lower the timings also? How much?
a b à CPUs
September 18, 2011 10:37:38 AM

lol np... I wouldn't tighten the timings until you get your computer error free. lets not rule out the memory just yet, maybe the board isn't supplying enough voltage to the chip at stock. Your at 3.2 correct? Try setting the cpu voltage to 1.15 and run some test. If that doesn't work you'll have to go back to the basics.

What I mean by that is underclock everything and test one thing at a time. So set the ram to it's lowest multi, qpi link under 6000, loosen timing on the ram etc. But try the cpu voltage first and let me know how it goes. When your ram is at 1333, what is the uncore set to? I have found that it should be double that of the ram speed, just fwi if you didn't know.
September 18, 2011 2:58:23 PM

Abstinent said:
Ok, I'm not going to claim that this is my last question :p 

I've tested it some more, and it passed small ffts with no errors (ran it for an hour). Then i tested inplace large ffts, and I got an error in worker#8 after 15 minutes (same as before). So then it's safe to say it's memory? Anyway, I'll try to lover the voltage some more. And btw, since I've lowered the frequency to 1333mhz, can i lower the timings also? How much?

it depends on the set of rams you've got and looking at what you've been saying the stock seems to be 1600mhz@9-9-9-24 so you might get to 8-8-8-21 or maybe a few 7's in there if your lucky but since cl9 isn't terribly bad for 1333 you shouldnt notice a massive difference anyway
September 18, 2011 4:28:53 PM

Cygnus x-1 said:
lol np... I wouldn't tighten the timings until you get your computer error free. lets not rule out the memory just yet, maybe the board isn't supplying enough voltage to the chip at stock. Your at 3.2 correct? Try setting the cpu voltage to 1.15 and run some test. If that doesn't work you'll have to go back to the basics.

What I mean by that is underclock everything and test one thing at a time. So set the ram to it's lowest multi, qpi link under 6000, loosen timing on the ram etc. But try the cpu voltage first and let me know how it goes. When your ram is at 1333, what is the uncore set to? I have found that it should be double that of the ram speed, just fwi if you didn't know.


Tried to reduce CPU voltage to 1.15, and CPU temperature went down about 10 degrees! (Even if it was 0.94 when it was idle, it rose to 1.25 at most in heavy load - now it's constant at 1.14). But I still got the same error in inplace large ffts about 18 mins, though it was in worker 6 and not 8 this time (it that makes any difference). as for further underclocking, what about lovering cpu voltage to 1.1 for starters, or/and lowering qpi and dram bus after that?

What is Qpi link? And what do you mean by the lowest multi?

The uncore is set on auto. Should i set it at the double of ram speed myself, or does it do that by itself in auto?
September 18, 2011 4:30:34 PM

megadelayed said:
it depends on the set of rams you've got and looking at what you've been saying the stock seems to be 1600mhz@9-9-9-24 so you might get to 8-8-8-21 or maybe a few 7's in there if your lucky but since cl9 isn't terribly bad for 1333 you shouldnt notice a massive difference anyway


Ok. But, when I get that far after testing system stability, should I test one timing at a time? Like go from 9-9-9-24 to 8-9-9-24?
a b à CPUs
September 18, 2011 5:42:04 PM

Ok so run your current settings by me again. I wouldn't lower the cpu voltage until you get stable, you might have to raise it a little. Lowest multiplier for the ram or the lowest frequency setting for now, just until we figure out the problem.

Yeah i would just set the uncore to double what ever you set the ram to.

There should be something that says QPI link or something to that effect. Try lowering that to under 6000 for now.
September 18, 2011 9:08:38 PM

Ok, some new info here... been posting on the Mushkin forum about this as well, and I've been given the advice to test each of the membricks separately. So I've tested all six now, and I've got errors on 4 out of 6! Ok, I didn't run Memtest, but I ran the in-place ffts, and on three of them it failed after a minute or so (on stock settings that a Mushkin-official gave me, he seems to now what he is talking about), and the fourth failed after 18 mins or so (all separately). The two left have run for an hour without any faults (separately), and now I'm running them both together (4gig) to test them again, just to double-check, and no errors so far (it's almost been an hour).

So... 4 out of 6 memory modules are faulty? Jesus Christ... and is this a fabric-fault or from the store I bought this pc? I bought it from Netshop - a norwegian retailer, which built, and supposedly checked the computer. Well, not good enough. Only problem now is that I don't know if I'll get 4 memory modules back on the warranty. Suppose I erased that option when I opened my computer... but taking out and inserting ram-modules isn't exactly rocket-scientist work either. Guess I'll have to check with them. And luckily this ram isn't expensive at all :)  Glad it isn't the CPU or something else (not that I know a 100%, but I'm 95% sure... have to do some more testing though).

Puh, a lot of work to figure this out, but I think I'm there. Thanks again very much for the help and assistance! :D 
September 18, 2011 10:12:26 PM

What the ***? I put on the chassis after I've tested all the memory, and then the BIOS told me that my "overclocking has failed". So i went into Bios and sat the cpu voltage up to 1.65 from 1.50, thinking that maybe the vcore voltage is too low (only thing I can think of, everything else is at factory settings). So, it booted okay, then when i ran Prime95 the pc-speaker started beeping almost right away. Short beeps. I panicked, pulled out the plug and I'm afraid to turn on the pc again. What the *** is happening? It can't be just the memory? Is something seriously wrong since it's beeping like this in Windows environment?
a b à CPUs
September 18, 2011 10:16:08 PM

Whoa not cpu voltage ram voltage right?
September 18, 2011 10:21:16 PM

Cygnus x-1 said:
Whoa not cpu voltage ram voltage right?


No, CPU voltage. What have I done? :S :S
September 18, 2011 10:23:14 PM

Sorry, I wrote wrong... from 1.15 to 1.165. I think, and hope.
September 18, 2011 10:26:07 PM

Sorry again, turned it up from 1.5 too 1.65 or 1.75! Gah, been spending entirely too much time in front of my computer :p 
a b à CPUs
September 18, 2011 10:27:10 PM

lol, ok had me thinking you fried your chip. does your ram support XMP in bios?
September 18, 2011 10:30:58 PM

Uhm... maybe I put it from 1.15 and up to 1.65 instead of 1.165. Would that have killed my mobo and CPU? :S
September 18, 2011 10:32:46 PM

It didn't report any significant change of temperature, ASUS PC Probe II, so it's got to be 1.165. Yes, I have XMP support in BIOS :) 
a b à CPUs
September 18, 2011 10:38:28 PM

I think maybe go back to the XMP setting with the current cpu voltage and see what happens. If no joy add a little voltage to the cpu. Also just for kicks, what does the voltages say in bios for the 12 volt rail and others?
September 22, 2011 2:37:43 PM

Hey, sorry for no updates or answer to your suggestion, Cygnus X-1. I was really tired of trying to get things to work. Anyway, I've located the issue (after several days of testing and troubleshooting) - 3 or 4 of the ramsticks are faulty. So I'm replacing them with redline instead of blacklines, 3x4 instead of 2 3x2 3x2 and a tad better timings :) 

Haha, funny store btw... the "pc-speaker" that started beeping while I ran Prime95 (and another time in BIOS) was actually an alarm-clock built into a custom-made lcd-panel at the front of my chassis. A lot fell into place when I figured that out :p  That's electrophobia for ya :p 
And the message that your overclocking has failed, from BIOS when you boot... it's just a standard message from BIOS when you turn off your pc while it's testing and booting BIOS :p 
a b à CPUs
September 22, 2011 3:28:00 PM

LOL that beeping would have had me a little worried also. That sucks about the memory, a lot of time and effort trying to get things right, just to find out you have bad parts. Well post back when you get the new ram and let me know how it goes.
September 22, 2011 8:44:08 PM

Cygnus x-1 said:
LOL that beeping would have had me a little worried also. That sucks about the memory, a lot of time and effort trying to get things right, just to find out you have bad parts. Well post back when you get the new ram and let me know how it goes.


Yeah, especially when the PC is supposedly "tested" and pre-built. No ***... it would have taken them less than one pass at memtest to've seen that something was wrong. And the BIOS recognized 2 out of 16 gig when the pc booted for the first time. The BIOS wasn't even updated. Beginning to seriously doubt either the effort or the capabilities of "computer techs" at Netshop. Last time I'm buying anything from there. At least when they're supposed to test it.

Thanks again anyway, and I'll let you know once I get the replacements :)  Which are these btw;

http://www.mushkin.com/Memory/Redline/998981.aspx

:)  Seems to be a lot of potential issues with 6 sticks, so I'm going with 3 instead :) 
a b à CPUs
September 22, 2011 9:37:36 PM

alright GL
September 28, 2011 10:58:43 PM

Jesus almighty.. I got the same memory bricks back even though they said I could trade/replace with the ones mentioned (998981). Yeah, well, at least they're working ok. Tuned them down to 1333 mhz, since it's 6 of them, but I'd like them to run at 1600. Since they are 1600s, I mean. Should I up the voltage or timings in case? Or just set them at 1600 at stock timings and voltage and test them from there on?
a b à CPUs
September 28, 2011 11:04:27 PM

I would run them at stock voltage and everything first and go from there.
September 30, 2011 12:53:57 PM

I checked with MushkinSean on the mushkin boards, and he recommended to start at QPI 1.325 and increase towards 1.375, while DRAM voltage should be constant at 1.65. If that didn't work I should up the DRAM voltage to 1.7. Done all of that, and even though the system is more stable with QPI 1.375 and DRAM 1.7, It's still unstable. Ran Memorytest 86+ for 6 hours, and somewhere between 2 and 6 hours it found an error. Any suggestions? :/ 
a b à CPUs
September 30, 2011 2:04:16 PM

I would recommend Going back to stock on your overclock until you get the ram issue straightened out if you haven't already. What are the timings set to, are the at spec?
September 30, 2011 2:52:45 PM

Cygnus x-1 said:
I would recommend Going back to stock on your overclock until you get the ram issue straightened out if you haven't already. What are the timings set to, are the at spec?


I have tried at stock with 1600, not stable. And in 1333 everything is stable (at stock).
a b à CPUs
September 30, 2011 3:08:09 PM

Can you return the memory altogether and buy a different brand? You shouldn't be having this much issue running memory in socket 1366. I have found 1366 to be better at handling memory than socket 775 not worse.
September 30, 2011 4:01:27 PM

Cygnus x-1 said:
Can you return the memory altogether and buy a different brand? You shouldn't be having this much issue running memory in socket 1366. I have found 1366 to be better at handling memory than socket 775 not worse.


That's what I tried, and allegedly were supposed to, but the crappy store forgot it or something, and sent the same memory back :heink:  But as I've said, it works fine in 1333 mhz, and as far as I know it's a pain in the ass to get 6 dimms/12g to operate stable at 1600 mhz. And that was partly why I asked Netshop to swap for 3x4gig dimms, but they must've slept through something... last time I'm buying anything from over there. Messy store. Anyway, I'll try to up the QPI voltage a tad and see how that goes :) 
!