Hi, I did all my tweaking... ran prime95 for about 6 1/2 hrs without any problems, then suddenly my comp randomly restarts. What exactly does this usually mean? Obviously thats not good? Whats consider a good stable time mark? I chose Small FFts btw. Thank you.
Is running 6 1/2 hrs of Prime95 considered being stable?
- Thread starter pRada
- Start date
Increasing the vcore one notch (0.0625v) should do it. Either that or your power supply or motherboard couldn't handle it.
I'd try to figure out what is stable and what isn't.
Run the CPU at its minimum multiplier (e.g. x6) with the RAM and FSB runing at
your desired test levels and everything else set up as you hope to have them.
Run memtest86+ for 24-48 hours. If you get no errors in that case, your memory and motherboard are pretty stable at those settings.
Now increase the CPU multiplier leaving and try Prime95 again and see if you get any failures. If you do, chances are it's something relating to the CPU and its interaction with the motherboard etc. Either adjust Vcore or lower your frequency a bit;
check your temperatures to make sure they're not getting too high etc.
I wouldn't consider 6 hours stable. For a casual use PC I'd say maybe 72 hours with no errors is pretty stable and unlikely to be rebooting on you at random times (which of course could cause data loss so you really don't want it to be happening to you).
If you are going to run the PC for extended periods doing useful work, e.g. distributed computing projects, acting as a file server / media center, etc. then I wouldn't consider it usefully stable unless it could run stress tests for 1 week+ with no errors and show no evidence of crashing under normal use for 3 months+.
Maybe you should turn OFF any settings that try to "fight" Vdroop in the BIOS if you have those.
See this page of this article and the next few following pages to explain:
http://www.anandtech.com/cpuch...howdoc.aspx?i=3184&p=4
Run the CPU at its minimum multiplier (e.g. x6) with the RAM and FSB runing at
your desired test levels and everything else set up as you hope to have them.
Run memtest86+ for 24-48 hours. If you get no errors in that case, your memory and motherboard are pretty stable at those settings.
Now increase the CPU multiplier leaving and try Prime95 again and see if you get any failures. If you do, chances are it's something relating to the CPU and its interaction with the motherboard etc. Either adjust Vcore or lower your frequency a bit;
check your temperatures to make sure they're not getting too high etc.
I wouldn't consider 6 hours stable. For a casual use PC I'd say maybe 72 hours with no errors is pretty stable and unlikely to be rebooting on you at random times (which of course could cause data loss so you really don't want it to be happening to you).
If you are going to run the PC for extended periods doing useful work, e.g. distributed computing projects, acting as a file server / media center, etc. then I wouldn't consider it usefully stable unless it could run stress tests for 1 week+ with no errors and show no evidence of crashing under normal use for 3 months+.
Maybe you should turn OFF any settings that try to "fight" Vdroop in the BIOS if you have those.
See this page of this article and the next few following pages to explain:
http://www.anandtech.com/cpuch...howdoc.aspx?i=3184&p=4
Im going to try and up my vcore a notch and I'll run the test again, if it continues to do the same problem, i'll do what Quix told me. I really appreciate the input. Thank you.
I'd also consider using the "Large FFTs" option instead. It has proven to be much more likely to reveal stability issues on most of my systems.
It might be a BSOD aswell and your OS is set to automaticly restart if a BSOD event occur. Try to disable automatic reboot (Right click My computer, go to Advanced, System Startup and Recovery.)
no, 6.5 hours is not enough, 12-24 hours is more like it.. i've seen P95 fail after 18-20 hours or so
large fft also does not confirm everything.
I usually run 6-10 hours of in-place large fft. to test CPU and solid NB connections. as well as power draw.
then do a and 24 hours of blend fully test ram.
I usually run 6-10 hours of in-place large fft. to test CPU and solid NB connections. as well as power draw.
then do a and 24 hours of blend fully test ram.
K, I disabled the automatic restart. Hopefully this might make a little impact. Thank you. Once again, thanks for the feedback and input.
- Sep 28, 2005
- 21,130
- 3,661
- 126
no that means your prime failed at the last possible second like it does on quads.
This is why i ask people at least 7-8 hours stable.
You need to fix your settings or find out why it did a reboot.
Possibly check windows update.
Assimilator1
Elite Member
- Nov 4, 1999
- 24,185
- 528
- 126
Yep agreed, needs to be 24 hrs really,and now I run F@H afterwards too because that can show up errors even after its passed P95 for 24hrs!:Q
QuixoticOne
Good points, you should add F@H as the ultimate stress tester too
Come on you need to prime for at least one year to figure in the seasonal ambient temp variations.
no.. priming doesn't test for temperatures.
it tests for system stability so seasonal variation in temperature has nothing to do with Prim95 stability.
you people must have nothing to do on your computers during all of these super long tests...
if its stable during what you do then dont worry about it... screw what they say...
if its stable during what you do then dont worry about it... screw what they say...
Well he sort of had a good point. If your PC is so overclocked and under-cooled in the winter that it is already red-line on the CPU / PWM / GPU / RAM / NB temperatures, but is stable in a cold room, you likely WILL have instability in the summer when the ambient room temperature might climb to 100F on hot days.
So it is a valid concern to artificially limit case airflow a bit so the interior case temperature is something like 110F and THEN see if everything is stable and operating within acceptable temperature limits under stress. If not, your PC WILL crash on hot days without a lot of air conditioning.
In the extreme overclocking forums it isn't that atypical to see better results from people who live in areas where the outside temperature is like -15C to +10C and they test with a window open or the PC just brought in from the cold.
garage overclocks the best......but damn it is cold....... my fingers get too frozen to
push down those heatsink pins
flexy
Diamond Member
- Sep 28, 2001
- 8,464
- 155
- 106
running "prime" without to know WHAT to test does not prove stability AT ALL.
I have a thread http://forums.anandtech.com/me...id=28&threadid=2164153 which you might want to read.
Eg. it is NOT advised to run prime "blend" as an "overall" stability test. Although Prime small FFT is the best test right now for the *CPU*.
Testing your system for stability not only involves a run of p95 (small FFT for CPU) but also testing your memory and chipset/northbridge etc subsystem. Many do not know but some programs can be used for one or the another - but are NOT good for testing ALL of a system.
Eg. there are much more sensitve tests for your memory/NB chipset/mem timings than prime 95.
In my thread i just added that OCCT now updated its CPU testing code (they now use the very good P95 code). For testing of memory you can run OCCT mem test, or (even better) memtest for windows. (Still the best for memory) OCCT2.0 right now actually is pretty good running in "MIX" mode for a overall idea of stabily of a whole system. I see that mem testing in OCCT also is about as sensitive as memtest for windows. But is still advise running memtest for windows.
Its a mistake to run P95, say small FFTs...then it runs fine for 5+ hours and think the whole system is stable. You still could have untested issues with your memory, timings and chipset which a run of P95 using small FFTs would just not find!
I have a thread http://forums.anandtech.com/me...id=28&threadid=2164153 which you might want to read.
Eg. it is NOT advised to run prime "blend" as an "overall" stability test. Although Prime small FFT is the best test right now for the *CPU*.
Testing your system for stability not only involves a run of p95 (small FFT for CPU) but also testing your memory and chipset/northbridge etc subsystem. Many do not know but some programs can be used for one or the another - but are NOT good for testing ALL of a system.
Eg. there are much more sensitve tests for your memory/NB chipset/mem timings than prime 95.
In my thread i just added that OCCT now updated its CPU testing code (they now use the very good P95 code). For testing of memory you can run OCCT mem test, or (even better) memtest for windows. (Still the best for memory) OCCT2.0 right now actually is pretty good running in "MIX" mode for a overall idea of stabily of a whole system. I see that mem testing in OCCT also is about as sensitive as memtest for windows. But is still advise running memtest for windows.
Its a mistake to run P95, say small FFTs...then it runs fine for 5+ hours and think the whole system is stable. You still could have untested issues with your memory, timings and chipset which a run of P95 using small FFTs would just not find!
Interesting stuff guys. Learned stuff I never knew before. Thanks. My comp was built for gaming. I use my comp mainly at night for about 3-4 hours for fps games. I oced it to make my comp a little faster for my games but at the same time I want it to be stable. Thats why I run prime95. So is it really necessary to run prime95 for 12+ hours?
flexy
Diamond Member
- Sep 28, 2001
- 8,464
- 155
- 106
i would just get newest OCCT http://www.ocbase.com/perestroika_en/index.php?Download and run this on Auto for an hour or two, really. This would gurantee you stabilty for 99,99999999% of real life apps/games incl. CPU, mem, chipset - except you have want to control air traffic or nuclear powerplants with your PC, then i would run it 6+ hours, just to make sure
Running stress tests is like playing that game "name that tune".
You could test seeing perfect stability for one minute, but it could be even so that it has a high probability of crashing within 10 minutes and most likely you'd never know since you could run several one minute tests and see no errors.
It is like rolling dice, most often you don't roll (1) (1) but sometimes you will...
Besides when I build a PC I am most likely not doing anything important with it for the first few days, and it will have little data on it. If it is EVER going to crash and lose data or break within the next year, I would MUCH rather it do it during the first week or two of testing time rather than once I am depending on it with my time and data.
Stress testing for 12 hours is 'free' anyway, start the test, have dinner, go to sleep, wake up, have breakfast and your PC has very close to finished.
If you must go to work / school in the morning / daytime then you might as well just leave it running the test for the next 10 hours beyond that until you are able to return to use it again, and in that case you will have done a 24 hour test without much any inconvenience.
If I am rushing to achieve a somewhat high overclock I might study what sort of expected settings I can use with my particular hardware, then adjust the frequency and timings and voltages appropriately and see if that works or not for at least 5 minutes. If you can see it fail within 0-15 minutes then you know you are significantly far away from settings that will be stable for hours, days, weeks, so you just immediately improve your settings to more likely successful ones and restart the test. Once you get it to the point where it does not fail within 30 minutes or so, you can always back off on the overclock by about 10% of the total overclock that can pass several minutes and you at least are close to values that could hopefully work for hours/days.
On the other hand if I just buy new hardware and I am in no rush, I would burn in test the whole system at stock voltage / speed with a stress test suite for 4 days to 2 weeks just to ensure the purchased hardware and drivers are at LEAST stable at their normal operating conditions since I have only a limited time to refund / exchange the purchases if they are not functioning at stock speeds. If they crash during an overclock I could not so much justifiably complain that they are defective, so I would not waste limited evaluation time testing the maximum overclock before I even assured myself that they will work normally (and sadly many things you may buy today may not even work properly at normal speeds)!
Thanks everyone for taking time and giving me advices and tips. I'll definetly take it into consideration.
Assimilator1
Elite Member
- Nov 4, 1999
- 24,185
- 528
- 126
pffft ,that's fine if you don't mind risking data being corrupted, your distributed computing project crashing part way through a data unit (or worse sending false results!) ,crashing in the middle of watching a film, or crashing part way into a mission of a MMORPG (which can take hours) etc , why risk any of that when you can just test for a day or 2?
What? haven't you read any of this thread? ,merely a few hours is almost useless ,as I & someone else mentioned we've seen P95 error out after 18hrs ,24hr is the minimium.
And Flexy makes a good point about testing with other programs too.
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 24K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter