Is Prime95 still the best stress test??

LarryJoe

Platinum Member
Oct 22, 1999
2,425
0
0
Been running it successfully for over 24 hours. Is this still the best test to determine stability?
 

Pecos

Member
Jun 11, 2001
72
0
66
CPUBurn is good at raising the temps, but IMO, Prime95 and the Serious Sam looping demo are best for finding if your CPU/system is actually stable. Any little quirks, and both of those will halt.
 

ssanches

Senior member
Feb 7, 2002
461
0
0
Another vote for CPUBurn-4. It really raises the temps high. Just be careful that you have adequate cooling...
 

Rand

Lifer
Oct 11, 1999
11,071
1
81
For sheer temperatures then CPUBurn is best, for sheer stability testing Prime95 is still much better though.
 

oldfart

Lifer
Dec 2, 1999
10,207
0
0
Prime95 is a better stress test than programs like CPUBurn. Prime95 tests for CPU and ram stability. If you have an unstable overclock caused by either, you will get an error. The CPU heat up type programs dont do that.

With Prime95, I was able to determine my ram is stable @
-->152 FSB 2-2-2
-->158 FSB 2-3-2
--> ??? FSB 3-3-3 (never tried above 166 FSB)
 

Duvie

Elite Member
Feb 5, 2001
16,215
0
71
I agree with 2 above me...prime95 will pick up many of the little quirks as mentioned...sometimes it isn't a heat issue bit power or ram timings...
 

DolfanJay

Junior Member
Feb 12, 2002
22
0
0
Try Monkeys Audio at www.monkeysaudio.com its a lossless audio compressor. Just take some wavs and compress them at high compression and then verify the ape files it creates. This program would error about 20% of the time on my system until I increased my voltage a little bit more. This is when my system would run Prime95 and 3dmark2001 at the same time for hours without any crashes. I gotta say I was shocked, thinking I had 100% stability. I know it was my overclock and not the program because returning to default speeds eliminated the errors.
 

AGodspeed

Diamond Member
Jul 26, 2001
3,353
0
0
Prime95 is the best IMO.

Running the GUI version of SETI@home while multitasking really stresses your system's stability as well.
 

ssanches

Senior member
Feb 7, 2002
461
0
0


<< For sheer temperatures then CPUBurn is best, for sheer stability testing Prime95 is still much better though. >>



Rand, doesn't CPUBurn also test stability? I've seen the voltages are pushed to the max. My +12V rail goes upto 12.44 with it. Prime95 just pushes it upto 12.31 to 12.37. Sisoft stress test (CPU multimedia test) using normal priority also pushes it upto 12.31 to 12.37. So could CPUburn also be a great stability test because it pushes the system to the limits as observed with the voltages....

 

oldfart

Lifer
Dec 2, 1999
10,207
0
0
A CPU is but one part of the system. If I run CPUburn yup, the temp goes up. It has no error reporting at all. You have to use it in conjunction with another application that can actually detect and report an error. CPUburn does not test ram stability which when FSB overclocking, is critical to test.

CPUBurn is OK to test cooling capacity and maybe power supply capability, but is does not test for system stability like Prime95 does.
 

Rand

Lifer
Oct 11, 1999
11,071
1
81


<< For sheer temperatures then CPUBurn is best, for sheer stability testing Prime95 is still much better though. >>



Rand, doesn't CPUBurn also test stability? I've seen the voltages are pushed to the max. My +12V rail goes upto 12.44 with it. Prime95 just pushes it upto 12.31 to 12.37. Sisoft stress test (CPU multimedia test) using normal priority also pushes it upto 12.31 to 12.37. So could CPUburn also be a great stability test because it pushes the system to the limits as observed with the voltages....[/i] >>



It does test stability also, in fact that's pretty much the whole point of CPUBurn.... that said temperatures and voltage are hardly a determinating factor for the amount of stress the processor is under. Prime95 is simply a much more well rounded and thurough test then is CPUBurn, and tests far more variables.

Voltage alone is but a vague indicator of stress the system is under, many factors can influence that considerably.
There is a reason AMD and Intel recommend Prime95 for stress testing ;)
Short of NTBenchS*, or whatever program(s) Intel uses for in-house stress testing your not going to find a more consistently reliable stress test then Prime95.

BTW-NTBenchS is AMD's own in-house testing program of course.... though naturally it's not publically available.

If I want to test for adequate cooling then I run CPUBurn because it will evoke the highest temperature of any program I've yet seen.
But if I want to test sheer processor/RAM/motherboard stability then I'll definitely look towards Prime95 first.

I've found it will consistently error out long before any other program will show even the slightest signs of instability.
 

Divebrake

Member
Dec 6, 2001
40
0
0
Prime95 is great for exposing problems with RAM timings. I haven't tried CPU Burn, but Toast heats things up way better than SETI and really stresses the 5V rail on your PSU.
 

ssanches

Senior member
Feb 7, 2002
461
0
0
Thanks Rand, really good info you've provided over there...You mentioned about AMD's benchmark? Aren't you talking about NBench? I thought those were available on many sites? Distributed unofficially, right? I think a newer version was released around a month back....

 

Rand

Lifer
Oct 11, 1999
11,071
1
81


<< Thanks Rand, really good info you've provided over there...You mentioned about AMD's benchmark? Aren't you talking about NBench? I thought those were available on many sites? Distributed unofficially, right? I think a newer version was released around a month back.... >>



NBench is AMD's benchmarking solution that they'd like reviewers to use... though naturally coming from AMD one would expect it to be somewhat biased for AMD's own processors.
NTBenchS is an entirely different program, and not publically available. I doubt if even influential sites like AnandTech could get ahold of it, NTBenchS is specifically designed to promote optimum stress upon code paths throughout the K7 design.
Intel has a similar program, if not more then one.... also not publically available.
 

CSoup

Senior member
Jan 9, 2002
565
0
0
run as many looping tests as you can. prime95, 3dmark, unreal flyby, quake3 timedemos, cpu stability test, sandra benchmarks, compress/decompress large files, and other things like that. You want to execute as many different instruction types as possible to test as much of the chip as possible.
 

ssanches

Senior member
Feb 7, 2002
461
0
0
Rand, I hadn't heard of NTBenchS before. Thanks for the info... I'll be happy with Prime95, Seti and Cpuburn for now.... :)