Are three seperate 8-hour runs of Orthos or Prime95 the same as one 24-hour run?
- Thread starter Psynaut
- Start date
That sounds good, man. Play an intensive round of World in Conflict, Crysis, or UT3. If you don't burn-out and crash, you can stamp it with an ok.
boomhower
Diamond Member
- Sep 13, 2007
- 7,228
- 19
- 81
That was a little harsh.
OP no it is not the same. Three eight hour runs is just a waste of time as you are essentially just repeating the same test you already know it has passed. Do you need to run it for 24 hours? That is a decision you need to make for yourself. Some say eight is enough some say 24. I have even ran across a couple people that do 48. Myself I run 1(yes 1) hour of OCCT. In my experience it has proven to do very well. Typically it will give errors much sooner. I will get errors 30 minutes into OCCT that I may not get for 6 hours in Orthos. Stability is in the eye of the beholder. Do what you feel comfortable with. Keep in mind an unstable overclock can and will corrupt data, write erroneous data, and other nefarious things. So just because you are not getting lockups or random reboots does not mean everything is going alright in you system.
OP no it is not the same. Three eight hour runs is just a waste of time as you are essentially just repeating the same test you already know it has passed. Do you need to run it for 24 hours? That is a decision you need to make for yourself. Some say eight is enough some say 24. I have even ran across a couple people that do 48. Myself I run 1(yes 1) hour of OCCT. In my experience it has proven to do very well. Typically it will give errors much sooner. I will get errors 30 minutes into OCCT that I may not get for 6 hours in Orthos. Stability is in the eye of the beholder. Do what you feel comfortable with. Keep in mind an unstable overclock can and will corrupt data, write erroneous data, and other nefarious things. So just because you are not getting lockups or random reboots does not mean everything is going alright in you system.
Nicely put. I myself let Orthos run for a full 2 days prior to doing anything of any importance. I've had many a system fail after 20+ hours...and for me that is just not good enough...but to each his/her own.
I use OCCT now also, I test with prime just to check, but I've been stable in one and not the other. I like the x264 encode bench is a good enough check for me for what i do. If I don't error during that and it puts a lot of strain on the CPU during the second passes. If that passes I'll move to OCCT. But really encoding hd video is the most ima do with my PC and hardly any at all. Mostly I'm going to game on it, if I can get a couple hours of gaming in with no hang ups that's good enough for me. I'm not keeping the PC on 24/7 f@h so I don't really need the 24hr stress test. I've been turning it off when I'm not using it mainly to save on my electric bill since I have 3 pc's running, my 360, and a TV almost all day. The bill was crazy the last two months, so I've been shutting down when not in use.
Thanks, I wasn't sure if Orthos and Prime95 repeated itself after a while or if it kept generating more difficult calculations. I grabbed OCCT and will try it out now. Thanks
Ummm huh?
People, 3 successful (do not crash / give errors) runs of Prime95 is the same as one 24 hour run in terms of validating statistical stability, provided that it is being run in an error checking mode such that it is frequently able to validate that the results it generates is correct. So as long as it can complete enough calculations in 8 hours to know that the results of those calculations (taking most of the 8 hour time) are correct, then the run isn't wasted.
Errors don't magically appear after 254.132 hours continuous running or whatever, they're random errors that just have some certain probability of happening at any time the system is under heavy load, e.g. 3% probability per hour of calculations or whatever the figure is for your particular system. So if you have (as an example) a 3% per CPU hour error probability then after 24 hours that'd be a chance of successful computation of:
97%*97%*97%...24 times or 0.97^24 = 0.4814 = 48%
Now in a given three hour period your chance of successful computation is:
0.97*0.97*0.97 = 0.9126 = 91.26%
Now if you ran that three hour test eight independent times with each time having a 91.26% chance of success,
we get:
0.9126^8 = 0.4814 = 48.14% chance of success in eight total runs of three hours each, the same as if you'd run the test for 24 hours straight.
So it is the same.
For that to be true, all that is necessary is, again:
1) The error is considered *random* and statistically independent of the general circumstances of the computation, otherwise it is more of a 'bug' than a random error/glitch event.
2) The computation that happens over any short run is verified as being either correct or incorrect within the space of that run so that the entire time the test is run counts as a possible test of whether there was an error or not. Depending on how you configure/run Prime95 and how long your test run is this may or may not be true.
Although it is certainly possible to see your first error after, say, 21 hours of computation, that doesn't mean that if you ran the same test again with the same settings it would again crash after exactly (or even close) 21 hours. It is just a random event with some low probability of happening but which can happen at any time,
otherwise it is a "bug" and not a "glitch".
People, 3 successful (do not crash / give errors) runs of Prime95 is the same as one 24 hour run in terms of validating statistical stability, provided that it is being run in an error checking mode such that it is frequently able to validate that the results it generates is correct. So as long as it can complete enough calculations in 8 hours to know that the results of those calculations (taking most of the 8 hour time) are correct, then the run isn't wasted.
Errors don't magically appear after 254.132 hours continuous running or whatever, they're random errors that just have some certain probability of happening at any time the system is under heavy load, e.g. 3% probability per hour of calculations or whatever the figure is for your particular system. So if you have (as an example) a 3% per CPU hour error probability then after 24 hours that'd be a chance of successful computation of:
97%*97%*97%...24 times or 0.97^24 = 0.4814 = 48%
Now in a given three hour period your chance of successful computation is:
0.97*0.97*0.97 = 0.9126 = 91.26%
Now if you ran that three hour test eight independent times with each time having a 91.26% chance of success,
we get:
0.9126^8 = 0.4814 = 48.14% chance of success in eight total runs of three hours each, the same as if you'd run the test for 24 hours straight.
So it is the same.
For that to be true, all that is necessary is, again:
1) The error is considered *random* and statistically independent of the general circumstances of the computation, otherwise it is more of a 'bug' than a random error/glitch event.
2) The computation that happens over any short run is verified as being either correct or incorrect within the space of that run so that the entire time the test is run counts as a possible test of whether there was an error or not. Depending on how you configure/run Prime95 and how long your test run is this may or may not be true.
Although it is certainly possible to see your first error after, say, 21 hours of computation, that doesn't mean that if you ran the same test again with the same settings it would again crash after exactly (or even close) 21 hours. It is just a random event with some low probability of happening but which can happen at any time,
otherwise it is a "bug" and not a "glitch".
Thanks Quix,
I did watch the Prime 95 test names, and I thought it appeared to be repeating the same tests every few hours or so, but I certainly wasn't sure and didn't want to make assumptions about how these programs work. Also, I was running it with error checking enabled.
BTW, I am stable after one hour on OCCT, so that is reassuring.
I did watch the Prime 95 test names, and I thought it appeared to be repeating the same tests every few hours or so, but I certainly wasn't sure and didn't want to make assumptions about how these programs work. Also, I was running it with error checking enabled.
BTW, I am stable after one hour on OCCT, so that is reassuring.
i don't think running 3x8 hour runs of p95 is the same as one 24 hour run, there are many different calculation tests that they run where errors won't be caught in 8 hours
thats just like saying can i do 24 x 1 hour runs of p95. no because then you'll just be doing the same test 24 times.
thats just like saying can i do 24 x 1 hour runs of p95. no because then you'll just be doing the same test 24 times.
HannibalX
Diamond Member
- May 12, 2000
- 9,359
- 2
- 0
I would say no it isn't the same?
Basically you are getting three clean attempts with the eight hour jobs. With the 24 hour job you are getting one clean attempt and whatever happens, happens. It could run all the way through or crash slightly past eight hours.
For your purposes though I believe the three, eight hour runs are adequate to show your system is stable enough to use.
Basically you are getting three clean attempts with the eight hour jobs. With the 24 hour job you are getting one clean attempt and whatever happens, happens. It could run all the way through or crash slightly past eight hours.
For your purposes though I believe the three, eight hour runs are adequate to show your system is stable enough to use.
I'm glad to see your system is starting to look stable. Good luck with it.
Well I didn't want to get too much into the details of under what circumstances short-run computations
are useful in Prime95 or not just because I think there are a lot of different configurations that can be used to make short test runs useful, though there are also some configurations that make short test runs not useful, as well as some configurations that don't do much any verification testing at all (AFAIK) so they wouldn't ever catch errors.
To successfully do a LL test on a very long possible prime like
2^32,582,657
which is the largest known Mersenne prime can take a system about 1-2 CPU weeks of computation or so
as far as I know with Prime95.
When you're running in "non stress test" mode that is one of the more typical / useful types of calculations
Prime95 does, e.g. LL testing of a particular possible prime like M32582657.
I believe it saves its intermediate LL calculation every so often (maybe every half hour? read the docs), or when you tell it to gracefully pause/stop/close down it typically also saves its spot within the LL test it is currently
running so when you start the program again it'll immediately start pack at the same spot in the long calculation so you won't have lost more than around a half hour's worth of calculations even if you stop and restart the program after 2 hours on like 100 different occasions -- eventually it'll be 2 weeks worth of total CPU time and it'll have a result of that particular LL test.
Other types of tests like trial factoring can run much more quickly than the LL test and they may only take a few minutes per possible factor tested or whatever... It depends on your settings and work list and stuff.
In any of these cases it can possibly use certain strategies to double-check the result of calculations it does either by repeating a given calculation, or doing a certain sub-calculation like a multiply with a couple of different algorithmic inputs that it can cross check to see if the result of them in comparison or in analysis can possibly be correct. It can do that pretty much for any modular multiply operation (which is the basis of the whole thing) it does.. it just depends on its settings and work configuration etc.
In stress test mode I know that it isn't trying to produce a useful calculation of an unknown result, it is just trying to perform calculations with known or easily verifiable results to look for errors in the PC. I don't recall precisely what kinds of error checking it does in this case, or how often it checks for errors, or whether it would even bother to save intermediate results and restart calculations at all in this mode of operation. Empirically, however, if you can sometimes see it give you a red "ERROR/FAILED" indication after only a few minutes of runtime on an unstable system, this tells me they're probably verifying results pretty often in this case however they do it, so it seems that if you're talking about run times of more than a couple hours full load they'll have usefully verified many calculations within that span before you interrupt the testing, and, hence, basically the whole time would qualify as being useful test time against a possibly random error, and thus the test time would be cumulatively reassuring even if you only do many somewhat short (hour or more) runs.
I'd ask around at mersenneforums or mersenne.org or whatever if you want to know exactly how long it takes to detect an erroneous calculation when you're running in test mode, or how often it checks for errors in non-test-mode depending on if you're LL checking, trial factoring, or whatever. In test mode the mere fact that it repeats the same SET of computations should not matter since even if you were just calculating something simple like "1+1=? Check to see if result=2" trillions of times you should eventually see an error if the system is unstable as long as your calculation program at least does a decent job at exercising the FPU/ALU/Cache/registers and other parts of the CPU / memory. The specific calculation involved doesn't (to a good approximation) matter at all as to the probability of a CPU error, only that that calculation fully uses certain parts of the CPU (fpu/alu/cache/registers/SSE2/whatever...).
Well I didn't want to get too much into the details of under what circumstances short-run computations
are useful in Prime95 or not just because I think there are a lot of different configurations that can be used to make short test runs useful, though there are also some configurations that make short test runs not useful, as well as some configurations that don't do much any verification testing at all (AFAIK) so they wouldn't ever catch errors.
To successfully do a LL test on a very long possible prime like
2^32,582,657
which is the largest known Mersenne prime can take a system about 1-2 CPU weeks of computation or so
as far as I know with Prime95.
When you're running in "non stress test" mode that is one of the more typical / useful types of calculations
Prime95 does, e.g. LL testing of a particular possible prime like M32582657.
I believe it saves its intermediate LL calculation every so often (maybe every half hour? read the docs), or when you tell it to gracefully pause/stop/close down it typically also saves its spot within the LL test it is currently
running so when you start the program again it'll immediately start pack at the same spot in the long calculation so you won't have lost more than around a half hour's worth of calculations even if you stop and restart the program after 2 hours on like 100 different occasions -- eventually it'll be 2 weeks worth of total CPU time and it'll have a result of that particular LL test.
Other types of tests like trial factoring can run much more quickly than the LL test and they may only take a few minutes per possible factor tested or whatever... It depends on your settings and work list and stuff.
In any of these cases it can possibly use certain strategies to double-check the result of calculations it does either by repeating a given calculation, or doing a certain sub-calculation like a multiply with a couple of different algorithmic inputs that it can cross check to see if the result of them in comparison or in analysis can possibly be correct. It can do that pretty much for any modular multiply operation (which is the basis of the whole thing) it does.. it just depends on its settings and work configuration etc.
In stress test mode I know that it isn't trying to produce a useful calculation of an unknown result, it is just trying to perform calculations with known or easily verifiable results to look for errors in the PC. I don't recall precisely what kinds of error checking it does in this case, or how often it checks for errors, or whether it would even bother to save intermediate results and restart calculations at all in this mode of operation. Empirically, however, if you can sometimes see it give you a red "ERROR/FAILED" indication after only a few minutes of runtime on an unstable system, this tells me they're probably verifying results pretty often in this case however they do it, so it seems that if you're talking about run times of more than a couple hours full load they'll have usefully verified many calculations within that span before you interrupt the testing, and, hence, basically the whole time would qualify as being useful test time against a possibly random error, and thus the test time would be cumulatively reassuring even if you only do many somewhat short (hour or more) runs.
I'd ask around at mersenneforums or mersenne.org or whatever if you want to know exactly how long it takes to detect an erroneous calculation when you're running in test mode, or how often it checks for errors in non-test-mode depending on if you're LL checking, trial factoring, or whatever. In test mode the mere fact that it repeats the same SET of computations should not matter since even if you were just calculating something simple like "1+1=? Check to see if result=2" trillions of times you should eventually see an error if the system is unstable as long as your calculation program at least does a decent job at exercising the FPU/ALU/Cache/registers and other parts of the CPU / memory. The specific calculation involved doesn't (to a good approximation) matter at all as to the probability of a CPU error, only that that calculation fully uses certain parts of the CPU (fpu/alu/cache/registers/SSE2/whatever...).
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 Discussion Threads
- Started by Tigerick
- Replies: 21K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter