Never Using Prime95 Again...It's Ineffective

[MrControversial]

I was having a problem last night with my overclocked CPU/GPU. I didn't know what the problem was because the thing ran Prime stable for 24 hours with no problems. I uninstalled Far Cry, reinstalled it, uninstalled it again, patched it, unpatched it...everything. Everytime I'd run 3DMark03 and I'd crash and restart. I didn't know what the problem was. Everytime I'd play Far Cry the game would crash within seconds.

So I finally broke down and ran SuperPI. The damn thing failed on the 16K test. Talk about unstable! So I lowered the memory to DDR333 (with virgin timings) and voila passed the 32M test with flying colors. 3DMark03 and Far Cry don't crash anymore. Yay for me and for
SuperPI. Why run Prime95 for hours when PI can break your processor in a matter of minutes?

 

[CaBoOse999]

Prime95 tests the stability of the CPU for the most part depending which test you run. So if your ram was the problem it wouldn't necessarily show up in prime95.

 

[Slaimus]

Maybe you are running P95 with the wrong test settings?
If you want to stress memory, select 4096k FFT and do not run them in place, and set the memory to use to about 3/4 of total.

 

[LTC8K6]

Yeah which P95 torture test did you run?

 

[MrControversial]

I ran the one that stresses memory. I don't have a floppy drive so Memtest86 isn't an option.

 

[Thor86]

Originally posted by: MrControversial
I ran the one that stresses memory. I don't have a floppy drive so Memtest86 isn't an option.

You can use a bootable usb drive or cd rom.

 

[Koyanisquatsi]

you can download the bootable iso off memtest's webpage.

 

[MrControversial]

Bootable ISO for my flash drive?

 

[CaBoOse999]

Originally posted by: MrControversial
Bootable ISO for my flash drive?

To burn it to a cd

 

[MrControversial]

Thanks n00b. Nice rig u got there.

 

[THUGSROOK]

there is nothing wrong with p95

 

[Duvie]

Originally posted by: THUGSROOK
there is nothing wrong with p95

xactly!!!You just need to know how to use it...The one thing I have said forever is that it is not a good stress test of memory and that is why I run test #1 and #2 to test the cpu and memtest (with cache on) to test the memory subsystem....

More specifically run about 10 passes full and then about 50 passes each of #5 and #6...this will dial it up for you...any restarts or freakiness after this would likley we do to heat over prolonged period of time...

 

[zakee00]

memtest 86 didn't do it for me, i pass all night long but i run windows based memtest and i fail after ~30 minutes.

 

[Avalon]

Originally posted by: MrControversial
I was having a problem last night with my overclocked CPU/GPU. I didn't know what the problem was because the thing ran Prime stable for 24 hours with no problems. I uninstalled Far Cry, reinstalled it, uninstalled it again, patched it, unpatched it...everything. Everytime I'd run 3DMark03 and I'd crash and restart. I didn't know what the problem was. Everytime I'd play Far Cry the game would crash within seconds.

So I finally broke down and ran SuperPI. The damn thing failed on the 16K test. Talk about unstable! So I lowered the memory to DDR333 (with virgin timings) and voila passed the 32M test with flying colors. 3DMark03 and Far Cry don't crash anymore. Yay for me and for
SuperPI. Why run Prime95 for hours when PI can break your processor in a matter of minutes?

Pi is breaking your memory, not your processor
Although, I'm wondering how Prime95 didn't fail if you really did do the blend test, since if Pi fails the 16K test, your memory must have been SUPER flaky!

 

[Duvie]

Originally posted by: zakee00
memtest 86 didn't do it for me, i pass all night long but i run windows based memtest and i fail after ~30 minutes.

that doesn't make sense....The windows test is far less accurate...

reasons??

1) It doesn't test all the memory...Only test non used ram which in most cacses can be 25%..

2) Any background apps or programs suddenly using th eram can possibly cause an error....

This program is not even ballpark...I suggest you get the memtest86+ cdrom iso version and make sure you are properly configuring it...YOu want to make sure you do all the test and make sure cache is on always....

If you do it like me you will never see an issue, unless heat is a factor. It seems like running it exclusively at night is not necessarily the best either. Likely the room is cooler at nigh and thus the stress may not be building up the max ram heat.

Just review your logic of what you said...It makes no sense...

 

[chinkgai]

Originally posted by: THUGSROOK
there is nothing wrong with p95

did they fix that page file bug yet? heh

also its seems the most recent version wont run ever since i upgraded from my 9800 pro (i've gone 6800gt then x800 series since then)

all it does is bounce my cpu usage from like 5 - 35 percent or something...forever basically, without ever failing or proceeding

im stuck using one version back =

as for memtest...i loop memtest # 5 for a least half an hour then choose test # 8

test 5 would pass the half hour+ and then test 8 would give me errors

 

[1EZduzit]

When I'm testing my overclocking limits, I use SuperPi first for at least 15 minutes, if you have a memory proplem it usually shows up before that. If it runs stable there, I go to Prime 95 and do the full torutre test. I run it for 2 or 3 hours when I can keep an eye on it and if it has no problems the I let it run overnight.

I also always keep my cpu/mem ratio at 1:1. It doesn't seem to make much of a difference in the SuperPi tests and seems to be much more stable on all the mobo's (which isn't that many) I've had.

 

[sharad]

24 Hours is not be enough for stability testing.

If SuperPi found the problem within a few minutes that doesn't mean it is any better than Prime95. What if tomorrow Prime95 finds the problem faster than SuperPi?

The test also depends upon the ambient temperature in the room etc.

The paging problem with the "Blend test" is because Prime95 is asking for way too much RAM from Windows. Windows will allocate the RAM but in the swap file not in the physical memory. For this reason the CPU usage may never reach 100% in blend test and you might see a lot of disk thrashing. To work around this problem you need to select "Custom test" and specify a lower amount of memory (half of physical ram maybe).

Prime95 has got its own forums Mersenne Forum. The author of Prime95 answers a lot of questions there.

For RAM quality testing memtest86 is very good. It has picked up all the faulty sticks I have ever bought. But remember its only testing your RAM for defects. It may not pick up the stability problems.

Use more than one application for more than a day if you much have an absoute stable system.

 

[Duvie]

Originally posted by: 1EZduzit
When I'm testing my overclocking limits, I use SuperPi first for at least 15 minutes, if you have a memory proplem it usually shows up before that. If it runs stable there, I go to Prime 95 and do the full torutre test. I run it for 2 or 3 hours when I can keep an eye on it and if it has no problems the I let it run overnight.

I also always keep my cpu/mem ratio at 1:1. It doesn't seem to make much of a difference in the SuperPi tests and seems to be much more stable on all the mobo's (which isn't that many) I've had.

15 minutes is not enough....I can pass 32mb and 27minutes of it and still get an error in memtest....oh by the way and get no errors in P95 for 8-12 hours....

When are you ppl going to learn P95 is not a good stresser of memory. On toip of that the blended test (full test) actually doesn't hold my cpu at 100% so it actually is the test my cpu gets the least hot and seems to draw the least voltage according to my monitors....

I see that probably about 80% of ppl here I wouldn't trust their "so-called stable OC.....

 

[classy]

I have said this before and I'll state it again. I believe the best test to run is looping 3dmark 2001. I have seen many pass memtest as well as Prime 95, but then try running a program and the thing crashes like a crackhead coming down from a high. And believe it or not the old system killer is still great, looping the old Unreal Flyby. These two really stress the entire system and I am confident that if your system is unstable it will show up looping either of these apps for a few hours.

 

[THUGSROOK]

there is no such thing as a program that can tell you if your system is stable.
however, there are plenty that can tell you if your system is not stable

 

[Duvie]

Originally posted by: classy
I have said this before and I'll state it again. I believe the best test to run is looping 3dmark 2001. I have seen many pass memtest as well as Prime 95, but then try running a program and the thing crashes like a crackhead coming down from a high. And believe it or not the old system killer is still great, looping the old Unreal Flyby. These two really stress the entire system and I am confident that if your system is unstable it will show up looping either of these apps for a few hours.

Except genius how are you going to know where the issue comes from then?? drivers??? heat of gpu??

You need to run test to isolate components....

Prime95 will stress cpu/cache and power on system (albeit not fully)....

I like to run this and loop 3dmark for 2-3 hours when I am starting to stress test the gpu...as the end all test by itself??? Not even close....Again I have looped 3dmark on a system that was not stable...IE a HT p4....how is 3dmark test that??? You have to run it in conjunction with another cpu intensive app....

I use memtest to test the memory....


When all of that is done I run some FH and make sure I can do several WU's without EEU....then I run my encoding apps I use all the time..If I get any errors that is also a sign...

I have seen about it all...I have seen these apps say nothing is wrong to have something be wrong...all you can do is try to cover all your bases and do that by making sure you cover the major components..there is no 1 test that can do this...plain and simple....

 

[classy]

Originally posted by: Duvie

Originally posted by: classy
I have said this before and I'll state it again. I believe the best test to run is looping 3dmark 2001. I have seen many pass memtest as well as Prime 95, but then try running a program and the thing crashes like a crackhead coming down from a high. And believe it or not the old system killer is still great, looping the old Unreal Flyby. These two really stress the entire system and I am confident that if your system is unstable it will show up looping either of these apps for a few hours.

Except genius how are you going to know where the issue comes from then?? drivers??? heat of gpu??

You need to run test to isolate components....

Prime95 will stress cpu/cache and power on system (albeit not fully)....

I like to run this and loop 3dmark for 2-3 hours when I am starting to stress test the gpu...as the end all test by itself??? Not even close....Again I have looped 3dmark on a system that was not stable...IE a HT p4....how is 3dmark test that??? You have to run it in conjunction with another cpu intensive app....

I use memtest to test the memory....


When all of that is done I run some FH and make sure I can do several WU's without EEU....then I run my encoding apps I use all the time..If I get any errors that is also a sign...

I have seen about it all...I have seen these apps say nothing is wrong to have something be wrong...all you can do is try to cover all your bases and do that by making sure you cover the major components..there is no 1 test that can do this...plain and simple....

Your analogy of isolating only works if your trying to find a the reason for a problem. Overclocking is a not problem. Thats why so many guys build oc'ed systems run Prime 95 declare it stable, then the the pc crashes to the desktop. System components run in-sync with each other. Thats why isolation programs like memtest and prime 95 may help to detect and isolate a problem, but as in this individual's case like so many it doesn't help test the system stability as whole, which is what you should look for. And thats why 3dmark and as I mentioned the old stand-by Unreal flyby are so good. If you have memory issues Unreal flyby and 3dmark 2001 will crash to the desktop. And these two programs create a real world scenario. You should test your system as whole when overclocking. And I never said one test is a do all. But Prime 95 is not a very good test and I speak from quite a few years of overclocking.

 

[THUGSROOK]

different platforms react differently depending on the stress test used.

the point is~
you need to run prime95
you need to run m86
...and you need todo more then just run those two tests.

on my current rig, m86 has the final word on stability. (in other words it crashes first before any other stress test or program)

:beer:

 

[Duvie]

I recommended isolation then consolidation in terms of real world apps and looping 3dmark2k1 with prime95....

P95 is a good test when understaning what its limitations are...most ppl who complain about it do so cause they have errors they cant explain and just dont want to declare their system unstable...

Unreal flyby will not crash to desktop even in a semi stable system for me without looping it for 2 hours....I looped the demo and even played for hours once and was fine...cmae back played a few hours more later on and had a freeze up...went into memtest and let it run a few hours while I went to bed and came back with about a dozen or so errors in test (not this current A64..older cpu).....

NO one test will do it...that is my point....

In conjunction with several apps can we better look for stability.....

One thing I have noticed since coming from p4 to A64 is that superpi 32mb would run complete on a very flaky overclock yet fail prime (running an older version 22 before the 3 test split up) within 5-10 minutes.....with the amd I can run blended test and run for 2-3 hours yet fail 32mb superpi......

No one app works all the time in all situations...many apps taking their stengths and using them in conjunction with others can do a better job.....

A non gamer such as myself will not test gaming for the hours most will...I dont play it...I encode so I run 4-6 hours 2 pass encoding test...I do cad so I I run several test of specviewperf test, and povray benchmark....