Prime95 question

[geekified]

I've ran my X2 3800+ for 24 hours on the small test, so is it stable?


Quoting from two different forums I've read (not this one):

"As for prime95, if you don't care about memory just use small FFT. Blend won't stress the CPU enough so it will take longer to fail."

"small fft's is NOT a good measure. i can run small fft's all day long with an unstable computer."

Zebo's guide here says for dual core to use small FFT, so is there any common consensus on this?

 

[Mattd46612]

u run 2 instances of prime to test both cores?

 

[MADMAX23]

Originally posted by: geekified
I've ran my X2 3800+ for 24 hours on the small test, so is it stable?


Quoting from two different forums I've read (not this one):

"As for prime95, if you don't care about memory just use small FFT. Blend won't stress the CPU enough so it will take longer to fail."

"small fft's is NOT a good measure. i can run small fft's all day long with an unstable computer."

Zebo's guide here says for dual core to use small FFT, so is there any common consensus on this?

You have to run the blend test for sure if you want to test stability, that's the king test, no doubt about it!!
Forget about small fft test!!

 

[VStrom]

Originally posted by: MADMAX23

You have to run the blend test for sure if you want to test stability, that's the king test, no doubt about it!!
Forget about small fft test!!

I wouldn't say forget about it. If you are testing the CPU only (isolation) this is the better test - you don't want your RAM causing the failure because you won't know if it is the RAM or CPU.

 

[nealh]

my experience is run memtest to check ram...if that is ok and you get 24hrs of small fft stable with same ram timings you should be fine

I ran small fft and had a memory issue and it cause prime to fail

You can go crazy with this testing..I doubt if memtest is ok and small fft you will find many places where you are unstable

blend test does not stress cpu enough but places more stress on ram...I prefer large fft over blend

 

[Thor86]

For dual core or Hyper-Threaded cpus, you should test using two instances of LARGE FFT testing for max heat/cpu stability.

It is not the be-all, end-all for stability testing, but is a good start for cpu-specific stability testing.

 

[MADMAX23]

Do not use small ffts test as the reference test for cpu stability because it's stupid to know your cpu can work "alone" for hours at a certain high speed without an error and then realise that when you run the blend test, it fails after a few minutes/hours (less than 6h usually) because at that high speed your cpu cannot communicate/work properly with the rest of the system although the rest of the system parts, I mean, RAM,HTT,nForce4 Chipset work correctly and without an error.
I have tested that in my system.
Use the Blend Test!!!

 

[Wentelteefje]

Originally posted by: Thor86
For dual core or Hyper-Threaded cpus, you should test using two instances of LARGE FFT testing for max heat/cpu stability.

QFT... This one gives me the most heat as well... Although it doesn't seem you can let it run on both cores (maybe with Task Manager?), OCCT is also a very stressing torture program...

 

[nealh]

Originally posted by: MADMAX23
Do not use small ffts test as the reference test for cpu stability because it's stupid to know your cpu can work "alone" for hours at a certain high speed without an error and then realise that when you run the blend test, it fails after a few minutes/hours (less than 6h usually) because at that high speed your cpu cannot communicate/work properly with the rest of the system although the rest of the system parts, I mean, RAM,HTT,nForce4 Chipset work correctly and without an error.
I have tested that in my system.
Use the Blend Test!!!

so lets see...do not test your cpu to make sure it truly is stable at a particular speed..only use the blend test which doesnt stress the cpu enough....Ill bet if you do not use small fft to test cpu you will find apps that will crash that are cpu intensive under full load

I personally run memtest for 6hrs at the ram speed I will use and then run both small fft and large fft for 8-15hrs to ensure I am good to go....

This has proven to work very well..to show my ram is up to stuff, cpu is stable and system works well ....

 

[themusgrat]

When OCing, most people use small to find out if the CPU itself is stable, because it has major bragging rights. Blend and large should also be run for at least 4 hrs., though, before you leave it there permanently. And memtest is always the RAM king.

 

[EvertonB]

I personally run memtest for 6hrs at the ram speed I will use and then run both small fft and large fft for 8-15hrs to ensure I am good to go....

This has proven to work very well..to show my ram is up to stuff, cpu is stable and system works well ....

DO you run small and large at the same time, one on each core or seperately for 8-15 hours?

 

[MADMAX23]

Steps to test stability:

1) Run Memtest86+ 1.65 tests 5 and 6, 25 times each, then run 5-10 times the whole Memtest tests.
2) Run Prime95 24.14 blend test for at least 12 hours, although 24h or more are recommended.
3) Run Prime95 24.14 blend test with 3DMark2001 running at the same time in loop mode again and again for at least 6 hours too so that all your system gets stressed to the MAX. (MAX CPU, GPU, HTT and RAM Stress)

If your system passes succesfully through those tests it can be considered 99,99% stable!!!

 

[nealh]

Originally posted by: EvertonB

I personally run memtest for 6hrs at the ram speed I will use and then run both small fft and large fft for 8-15hrs to ensure I am good to go....

This has proven to work very well..to show my ram is up to stuff, cpu is stable and system works well ....

DO you run small and large at the same time, one on each core or seperately for 8-15 hours?

No I run Memtest first to confirm ram settings...this last time I ran 100 passes on test 5, 3hrs each on test 7 and 8

then I run 15hrs of small fft dual prime 95, then 15hrs dual prime 95 large fft

This what I do now...not sure if it is the best or not but I have no issue if this passes

 

[nealh]

Originally posted by: MADMAX23
Steps to test stability:

1) Run Memtest86+ 1.65 tests 5 and 6, 25 times each, then run 5-10 times the whole Memtest tests.
2) Run Prime95 24.14 blend test for at least 12 hours, although 24h or more are recommended.
3) Run Prime95 24.14 blend test with 3DMark2001 running at the same time in loop mode again and again for at least 6 hours too so that all your system gets stressed to the MAX. (MAX CPU, GPU, HTT and RAM Stress)

If your system passes succesfully through those tests it can be considered 99,99% stable!!!

When running a blend I notice it will use with 2gb of ram..like 2450mg..which means alot of pagefile use which seems a waste

anyone have trouble running a blend test with 2GB of ram...how do you set it up

I think it is important to realize we can do all the "hardware stress testing" we want but once you add apps, drivers etc...stablity can be affected

so even if you pass Prime 95 you may not be stable unless you restress you system after each change which I think is absurd

So beware any software or hardware change can affect stablity....The issue of stress testing becaomes a chasing of the tale

I persoanlly stress my setup on the major hardware revision new cpu/mobo/ram(videocard I do separately as I upgrade less commonly)...and if I get my stable parameters I am done

Also..unfortunately. I have been to lazy to reinstall winxp this upgrade..kinda of stupid as I add and remove crap all the time, so if I end stable with my testing this time...it really is as I am sure I have numerous little driver/app issues

I run a few programs never designed or patched to run in winxp...ie Pagis Pro(this has always been a little unstable on XP for every system I used it on...about 6 now)

I should add there are probably users out there that do very stress stuff and critical stuff and overclocking may not be the best since any failure can be a problem....but as above even stock setups can have huge issues becasue of OS, software, hardware conflicts

 

[MADMAX23]

Download Prime95 version 24.14 and you should'nt have that weird problem with the amount of Ram you say it uses; 2gb...in my case it only uses 766Mb, it is de default value in Prime95 for the blend test if you have 1 giga of ram.

You own 2 gb of ram, don't you? Then, if you want to stress the 2 gigas in blend, specify 1700Mb of ram to be tested...(in custom mode)
I mean, do this: Select Blend test, after that, select custom mode and change the amount of ram specified by the one you want, 1700Mb in your case is a safe choice)

The pagefile bug was corrected in this version of Prime95.

 

[nealh]

Originally posted by: MADMAX23
Download Prime95 version 24.14 and you should'nt have that weird problem with the amount of Ram you say it uses; 2gb...in my case it only uses 766Mb, it is de default value in Prime95 for the blend test if you have 1 giga of ram.

You own 2 gb of ram, don't you? Then, if you want to stress the 2 gigas in blend, specify 1700Mb of ram to be tested...(in custom mode)
I mean, do this: Select Blend test, after that, select custom mode and change the amount of ram specified by the one you want, 1700Mb in your case is a safe choice)

The pagefile bug was corrected in this version of Prime95.

ok thanks...not sure if the pagefile bug is fixed...I am pretty sure I am using 24.14 and blend test use far exceeds my ram by like 500mg

 

[Strife2101]

Well just so you guys know the Custom test with 80-90% of your memory being used in it, is the hardest test to pass. And any of you saying it doesnt stress the CPU dont know what ur talking about.


I can be compeltely stable in memtest and in Small FFT for 24hrs and still fail the Custom Test. This happened to me on many occasions and it wasnt the memory failing it was the CPU and the volts had to be upped. This test is by far the hardest to pass and your system is not stable unless it pases it.

 

[edmundoab]

Originally posted by: MADMAX23

Originally posted by: geekified
I've ran my X2 3800+ for 24 hours on the small test, so is it stable?


Quoting from two different forums I've read (not this one):

"As for prime95, if you don't care about memory just use small FFT. Blend won't stress the CPU enough so it will take longer to fail."

"small fft's is NOT a good measure. i can run small fft's all day long with an unstable computer."

Zebo's guide here says for dual core to use small FFT, so is there any common consensus on this?

You have to run the blend test for sure if you want to test stability, that's the king test, no doubt about it!!
Forget about small fft test!!

ditto,
the blend test will 99.9 % confirm that your system is stable for that many hours you have ran it

remember to allow at least 8 hours of continuous running, like through the night

general test wise, even less then an hour you would know if ur PC is stable or not
i've personally experienced my comp failing after 6 hours or so..
therefore, it really is safe to run it 8 hours or more

 

[nealh]

Originally posted by: Strife2101
Well just so you guys know the Custom test with 80-90% of your memory being used in it, is the hardest test to pass. And any of you saying it doesnt stress the CPU dont know what ur talking about.


I can be compeltely stable in memtest and in Small FFT for 24hrs and still fail the Custom Test. This happened to me on many occasions and it wasnt the memory failing it was the CPU and the volts had to be upped. This test is by far the hardest to pass and your system is not stable unless it pases it.

ok but why would it be the cpu...supposedly small fft is max cpu stress test, right..per prime95

seems the blend stress the memory controller and ram more ..did you by any chance lower ram speed and retest without increasing the vcore???

what exactly is the custom/blend testing....seems it is the cpu-memory interaction that is failing not the cpu..because it should fail the small ffts

with that said what exactly is large fft test

so if blend is the only test to use why have the others???

 

[Strife2101]

yes i did lower the ram speed and it still failed until the vcore was risen.


Small FFT should be used to get a rough estimate of your maximum CPU speed at a current voltage, its primarily used for this because it doesnt mess with the memory so you will know if you fail its soley cause of the CPU and has nothing to do with the memory.

The Custom Test uses memory and CPU, and it does have something to do with the memory controller but that controller is on the CPU so thus if the memory controller is failing so is the CPU and thus the voltages must be increased. This is why this test is so important.

I gave up using memtest because i could pass 11hrs then still fail in windows or games, I recommend using a 32mb SuperPi to test memory I found this to be very effective.

 

[nealh]

interesting stuff..32mg super pi seems ok...but I can not seem to get dual super pi to run

hmm..maybe the cpu really needs more vcore...how much vcore increase did the custom blend test require to be stable???

 

[MADMAX23]

Steps to test stability:

1) Run Memtest86+ 1.65 tests 5 and 6, 25 times each, then run 5-10 times the whole Memtest tests.
2) Run Prime95 Small FFT test to find out aproximately where your cpu hits a brick wall with a certain vcore value.It must be stable for at least 6 hours. Once you think your Cpu is stable at that speed and with that voltage, proceed to step 3.
3) Run SuperPi 32M test 2 consecutive times, if it finishes both successfully, then proceed to step 4. If it doesn't, go back to step 1 or 2 and change your settings.
4) Run Prime95 24.14 blend test for at least 12 hours, although 24h or more are recommended. If it passes this test, proceed to step 5.
5) Run Prime95 24.14 blend test with 3DMark2001 running at the same time in loop mode again and again for at least 6 hours too so that all your system gets stressed to the MAX. (MAX CPU, GPU, HTT and RAM Stress)

If your system passes succesfully through those tests it can be considered 99,99% stable!!!

 

[Shimmishim]

i'm not a fan of the blend or large fft.

i've had a cpu do 3.0 ghz with large fft but crashed small fft in 2 minutes.

my method for complete stability.

memtest loop test #5 6+ hours
prime95 small fft 6+ hours
prime95 large fft 6+ hours
Superpi 32M
OCCT

Mem usually messes me up so I like to make sure it's stable.

 

[Pollock]

I learned from someone to do custom and set the min FFT size to 2048 and run that for a while (at least 12 hours over night for me). So that's pretty much what I do.