prime 95, which test do you run it??

[jjyiz28]

there is small, inplace, and blend, and custom. i noticed that blend is the one that catches errors the most quickly. which do you run ?? or do you guys run them all?

 

[Shimmishim]

i run the small fft's... it stresses the cpu the most...

to test your memory, use memtest86 v.1 and higher

 

[BlueWeasel]

Interesting....

I haven't updated Prime95 in years, and I didn't know the latest version had the different tests. Good to know, as I am in the process of benchmarking my new XP-M 2600.

 

[Fricardo]

From what I have heard of other people's testing, its best to use custom. If you set the fft's to 2048-4096 it will catch errors much sooner, or that's what a lot of people around here have been saying anyways. That's the setting that I have been using lately too.

 

[Zebo]

I run torture test. The documentation says thats most demanding.

 

[jjyiz28]

Originally posted by: Zebo
I run torture test. The documentation says thats most demanding.

http://pics.bbzzdd.com/users/jjyiz28/prime.jpg

yeah, there are 4 types of torture tests to run. its a new feature in the later release. i wonder what test it was running in previous versions.

 

[ectx]

Here is my experience:

1700xp oc to 2300mhz 11.5 x200 at 1.7v. (It was at higher speed but I used a lower speed fan to reduce noize). I thought it was 100% stable in older version of prim95 tests. But since I started using new prime95 in my other pc's, I start to have some doubt: here is my tests over a few weeks:

a. small fft - no error in 36 hrs.
b. in place large fft - no error in 36hrs.
c. blended - error out in 12- 14hrs.
d. custom - large fft (2048K and above) - errors out in ~ 10 hrs.

I have reprpeated the cycle of tests 2 test and the results are largely reproducable.

My guess is that most problem occurs in L-2 cache or in the pipelines from chip to cache to memory (the memory sticks have been proved to run at much higher fsb so it is not the problem of the memory sticks). So running larger FFT tests will detect the problems sooner. If you really oc the chip excessivley, I bet any of the torture tests will error out in a few min. When your chip is almost stable, I'd recommend the large fft custom test.

Maybe I should try to force the min fft to 4096 and see if it forces errors even quicker.

Also, I have tried the tests on p4, barton and observed that large fft custom tests find errors sooner.

Also, you are advised to at least try a 3-D test. I have systems that is stable in >10 hr in prime95 (it would ultimately error out in prime 95 but takes much longer) but could not run 3dmark2003 for > 5 min. I also have systems running 3d all day long but have problem with prime95 in a few hours. In general, I guess 3dmk detects NB problems.

Anyway, it is a frustrating process and you should try a few different tests as the preliminary tests. As you get close to the stable speed, I beleive there is no better tests other than running prime95. But I could be wrong.

 

[MercenaryForHire]

Originally posted by: ectx
Here is my experience:

1700xp oc to 2300mhz 11.5 x200 at 1.7v. (It was at higher speed but I used a lower speed fan to reduce noize). I thought it was 100% stable in older version of prim95 tests. But since I started using new prime95 in my other pc's, I start to have some doubt: here is my tests over a few weeks:

a. small fft - no error in 36 hrs.
b. in place large fft - no error in 36hrs.
c. blended - error out in 12- 14hrs.
d. custom - large fft (2048K and above) - errors out in ~ 10 hrs.

I have reprpeated the cycle of tests 2 test and the results are largely reproducable.

My guess is that most problem occurs in L-2 cache or in the pipelines from chip to cache to memory (the memory sticks have been proved to run at much higher fsb so it is not the problem of the memory sticks). So running larger FFT tests will detect the problems sooner. If you really oc the chip excessivley, I bet any of the torture tests will error out in a few min. When your chip is almost stable, I'd recommend the large fft custom test.

Maybe I should try to force the min fft to 4096 and see if it forces errors even quicker.

Also, I have tried the tests on p4, barton and observed that large fft custom tests find errors sooner.

Also, you are advised to at least try a 3-D test. I have systems that is stable in >10 hr in prime95 (it would ultimately error out in prime 95 but takes much longer) but could not run 3dmark2003 for > 5 min. I also have systems running 3d all day long but have problem with prime95 in a few hours. In general, I guess 3dmk detects NB problems.

Anyway, it is a frustrating process and you should try a few different tests as the preliminary tests. As you get close to the stable speed, I beleive there is no better tests other than running prime95. But I could be wrong.

You also have to consider that big-power 3D cards can drastically affect stability by being both the source of heat or the component that fails to it.

- M4H

 

[myocardia]

Originally posted by: MercenaryForHire

Originally posted by: ectx
Here is my experience:

1700xp oc to 2300mhz 11.5 x200 at 1.7v. (It was at higher speed but I used a lower speed fan to reduce noize). I thought it was 100% stable in older version of prim95 tests. But since I started using new prime95 in my other pc's, I start to have some doubt: here is my tests over a few weeks:

a. small fft - no error in 36 hrs.
b. in place large fft - no error in 36hrs.
c. blended - error out in 12- 14hrs.
d. custom - large fft (2048K and above) - errors out in ~ 10 hrs.

I have reprpeated the cycle of tests 2 test and the results are largely reproducable.

My guess is that most problem occurs in L-2 cache or in the pipelines from chip to cache to memory (the memory sticks have been proved to run at much higher fsb so it is not the problem of the memory sticks). So running larger FFT tests will detect the problems sooner. If you really oc the chip excessivley, I bet any of the torture tests will error out in a few min. When your chip is almost stable, I'd recommend the large fft custom test.

Maybe I should try to force the min fft to 4096 and see if it forces errors even quicker.

Also, I have tried the tests on p4, barton and observed that large fft custom tests find errors sooner.

Also, you are advised to at least try a 3-D test. I have systems that is stable in >10 hr in prime95 (it would ultimately error out in prime 95 but takes much longer) but could not run 3dmark2003 for > 5 min. I also have systems running 3d all day long but have problem with prime95 in a few hours. In general, I guess 3dmk detects NB problems.

Anyway, it is a frustrating process and you should try a few different tests as the preliminary tests. As you get close to the stable speed, I beleive there is no better tests other than running prime95. But I could be wrong.

You also have to consider that big-power 3D cards can drastically affect stability by being both the source of heat or the component that fails to it.

- M4H

That and the fact that many of the newest, fastest video cards use in excess of 100 watts of power. If that happened to be power that your cpu needed, then you've also got a problem.

 

[ectx]

Good points.

The heat on my 9500 - 9800 class video cards is considerable. The hot spot on my 9500 is well over 140f (I cannot remember the number clearly now) when I run 3d loops and oc the video carrd (I have since scale back the oc).

Demands on PS are getting higher and higher. A lot people solve their oc problems after switching to a better psu. I have never had this kind of experience but I believe it could be is true due to the reason you guys pointed out.