Prime95: Which torture test mode?

[Speedo]

Which torture test mode do you guys usually use when testing system stability? The default "Blend" mode?

I had my new machine running "the first" test for 8 hours without problems, and really thought it was stable. But I let it go, and after 15 hours, I got an error and it halted. I didn't know stability issues would show up after such a long time...

 

[Assimilator1]

They can do but its rare.
I still use the older version which doesn't give you a choice (1 less decision to make )

I'd go for blend though

 

[BlackMountainCow]

Blend test is ok if you're in kind of a hurry to test a system

Do the one where it says minimal memory testing if you want to test your overclocked CPU

Run memtest86+ (a different programe than Prime95) if you want to test your memory.

If both memtest86+ and Prime95 run for 24 hours without a problem, I'd consider the PC to be stable and well done.

:beer:

 

[Speedo]

Thanks a lot. That pretty much answers all my questions.

Btw, could it be a good idea to run for example 3DMark simultaniously, for one night? Reason being the added heat generated from the graphics card could make the otherwise stable CPU unstable...?

 

[BlackMountainCow]

You actually should run 3DMark anyway for at least a couple of hours. It's a good test for memeory, CPU and GFX. But I personally wouldn't run it together with Prime95. Cause if one of them fails, you don't know exactly which one cause the error and thus it'll only bring you back to run the tests independently. In my experience, after you've completet Prime95 and memtest86+ successfully, just play a couple of hours of your current favorite game (Doom3, HL2) and if that goes without any problems, than you're on the safe side.

:beer:

 

[Speedo]

BlackMountainCow: Yea, that makes sense. About memtest86+, I've just taken a quick look at it once. How do I set up the test (options and stuff), and does it halt the same way Prime95 does, if an error is detected?

 

[Rattledagger]

Running through some tests in prime95 and other programs doesn't mean your computer really is stable in a DC-project, and being stable in one project doesn't mean changing to another project will not lead to crash after a couple of minutes or hours...

Also, even if a project seems to run without any problems and with normal run-times, this doesn't mean the results really is correct. In example SETI@home "classic" the only time you know you've overclocked too much is if every wu terminates after a couple of minutes, but for the rest you really don't know excempt if you re-crunches every wu...

Under BOINC on the other hand, you have much better control, since if you've overclocked too much so you're only returning garbage, you'll not get any credit for this garbage either... :evil:


BTW, overclocking so much that you'll getting an ocassional error in a game or seti "classic" isn't really a problem. But, in other projects like CPDN there a wu often can take a month, and even more so in a Sulphur Cycle Model with expected run-time 2.5 months a crash due to overclocking 90% out in the wu really means you've just wasted 1-2 months crunching...

 

[BlackMountainCow]

@ Speedo: You basically download the thing and make either a bootable floppy or a bootable cd from it. It's all pretty much explained on the site. All you need to do then after creating the bootable floppy/cd is to have it in you cd/foppy drive and boot from it. It'll run itself and test your memory for as long as you want it to. There are some options but all I usually do is run the standard tests for 24h straight. If that goes through, your mem & timings should be ok.

@rattledagger: I agree with you. Some DCs are much more sensiitve to a bad overclock than others. But after passing both 24h of memtest96+ and Prime95, plus a couple of hours of complete system stree with a hardware demanding game, there's really nothing more one can do to test a system for its stability in my opinion.

 

[Assimilator1]

RD

Also, even if a project seems to run without any problems and with normal run-times, this doesn't mean the results really is correct. In example SETI@home "classic" the only time you know you've overclocked too much is if every wu terminates after a couple of minutes, but for the rest you really don't know excempt if you re-crunches every wu...

Prime 95 will pick up errors long before SETI trips up.
So if it passes P95 it'll work ok with S@H

Good point about CPDN ,I guess you'd want to run it for several months on other projects 1st

 

[Rattledagger]

SETI@home "classic" crunching to 100% with reasonable times doesn't mean the result is correct, so even is "stable" in prime95 you don't know without re-running every seti-wu if you're not producing an error in every 100 wu or something... For some more info, look at http://www.teamlambchop.com/bench/archives/012001.htm


More than one user have run a computer stable for a long time in different projects, but still crashes every wu in CPDN. Different projects uses the cpu, memory and even disk-subsystem differently, so running error-free in one project doesn't mean another project doesn't use the "wrong" part of the computer so much that you pushes it over the edge...

 

[Speedo]

So it seems as if you really want to be sure the system is completely stable, it should pass 24h+ Prime95, 24h+ memtest86, 24h+ 3DMark, and when this is done, you should back a couple of FSB steps, alternatively up the core voltage one step to make sure you have some headroom.

I have never been this serious when testing for system stability before, but maybe I will have it as a routine from now on, although I guess it could cause some problems sometimes if I'm in a hurry

 

[Assimilator1]

Originally posted by: Rattledagger
SETI@home "classic" crunching to 100% with reasonable times doesn't mean the result is correct, so even is "stable" in prime95 you don't know without re-running every seti-wu if you're not producing an error in every 100 wu or something... For some more info, look at http://www.teamlambchop.com/bench/archives/012001.htm


More than one user have run a computer stable for a long time in different projects, but still crashes every wu in CPDN. Different projects uses the cpu, memory and even disk-subsystem differently, so running error-free in one project doesn't mean another project doesn't use the "wrong" part of the computer so much that you pushes it over the edge...

Hmmm,ok but I'd say that only probably applies if your running your CPU or RAM to the absolute edge,but not so likely if your CPU tops out at say 2.8GHz but you run it at 2.7GHz.Hard to prove though ,did Roelof ever release the program he used to check results?

And for the record I submitted 3 results to TLC & they were accepted & added to the table (& the fastest one at no17 was an o/c to the edge!) ,I must be doing something right