Overclock.net banner

1 - 12 of 12 Posts

·
Registered
Joined
·
1,366 Posts
Discussion Starter #1
i was running a p95 blend test (5min per fft) and got en error on the 224k test. now without changing anything i started a 224k test again and the error doesn't appear anymore?

i was thinking that instead of testing all over again after adjusting my settings i could just jump to the 224k test to see if the changes fixed the problem that happened with the last settings and if so go to the full testing.

but without being able to reproduce the same error with the exact same settings that wont be possible.

1. why dont i get the same error again when testing the same fft with the same settings?
any ideas why this could happen?

2. if i test this fft for like 1-2 hours (which i dont have yet) how can it be a good stability test when testing 12-24h?
one fft cycle takes about 8h (15min per fft) so each fft gets tested for about 15-45min, so if im not able to reproduce one error in like 45min then i doubt 12-24h are enough:O

i always liked testing with p95 and always used to test the problem size that caused an error. but i never tried it with the same settings... so i never noticed that issue:/
 

·
Registered
Joined
·
2,160 Posts
i have found that when running blend if i get an error in either large or small fft's if i then try just running say small fft's(if thats where i errored)it will not fail.theres something about the switching between large and small that will show the error that running just large or small wont show?ive tested this thoroughly and it always does the same thing.i get an error while running blend during the large fft's,then i run a few hours of large and no errors
 

·
Registered
Joined
·
2,160 Posts
no i dont think so,my temps are always in check i use AC.just look at this SS
the switching between large and small fft's errors has been well documented by others as well.you can error by having vcore slighty too low and make it through the 1024k's in blend and have it fail during the 8k fft's.then i stop the blend run and then just run an hour of 8k with out an error.then try blend again and it fails during the 8k ftt's again,and its right after it switches,doesnt make it to the second 8k fft without error.ive done it many times
 

·
Registered
Joined
·
1,366 Posts
Discussion Starter #6
hm very interesting.... but if it fails when switching from large to small fft there is clearly a stability problem?

the range where i got the error is very small, not like 1024 to 8k..

Code:
Code:
32
40
[B]224[/B]
192
i got the error with 224, so im gona see if it happens again when i run a 40-224 range.

edit: got the error already in that range but reduced the test time to 0.1 so it will jsut do one of each ffts and go on.
if this is where the errors come from in blend tests wouldnt it be better to run them with minimal time per fft to produce the problem?
 

·
Registered
Joined
·
2,160 Posts
Quote:

Originally Posted by kromar View Post
hm very interesting.... but if it fails when switching from large to small fft there is clearly a stability problem?

the range where i got the error is very small, not like 1024 to 8k..

Code:

Code:
32
40
[B]224[/B]
192
i got the error with 224, so im gona see if it happens again when i run a 40-224 range.

edit: got the error already in that range but reduced the test time to 0.1 so it will jsut do one of each ffts and go on.
if this is where the errors come from in blend tests wouldnt it be better to run them with minimal time per fft to produce the problem?
yes there is a stability problem for sure.try upping vcore one notch and see if it fixes it.below i put a quick basic explanation of prime95.im not familar enough with AMD cpu's to determine what you should or need to adjust for large fft problems(memory/NB/Vcore) to really help with your settings and what to adjust.

basics are like this.

errors in small fft's are most likely cpu related
errors in large fft's are memory and NB/IMC related
if vcore is way too low it can error in large or small or crash
small fft's are these 8k,14k,20k and so on
large fft's are these 768k.986k,1024k and so on
running blend switches between small and large,so depending
on what size it fails at will determine what needs to be done
to fix it.

heres a great link on how to use prime95.

Torture test your CPU with Prime95

^this link has great info for Prime95

 

·
Registered
Joined
·
1,366 Posts
Discussion Starter #9
so i did some intensive testing about this problem and it turns out that this happens if the CPU-NB is unstable:O this might be a good method to test this because errors occur really quick when setting the FFT time to 0.1.

@radaja: i have a ram stability problem when i run it over 1400mhz, and it always errors @ 8k FFT, but this range is supposed to be a CPU problem indicator. so why does this problem occur at that FFT?
 

·
Registered
Joined
·
2,160 Posts
the small fft/cpu and large fft/memory/NB thing is just a basic
guideline but sometimes like if the vcore is too low it can fail
in the beginning of blend during the 1024K fft.
ive even had it fail during the 8k because of my NB ref voltage
being slightly off.im not too familiar with these types of quirks
with Prime95 and AMD but im sure they happens too.
so basically if something is off,voltage wise and its unstable
it can fail anytime,so more of a guideline and not a 100% rule

just need to really test each thing thoroughly.like try dropping
the cpu multi to rule out vcore and test each vdimm and NB voltage.
once you can prime stable and rule out the memory and NB
then work on the cpu frequency and visa versa.
thats the only real way to determine whats happening is to just
really be meticulous and test thoroughly
 

·
Registered
Joined
·
369 Posts
Cool+rep will try that to reproduce errors i get to see if tweaked settings solve it.

Quote:

Originally Posted by kromar View Post
so i did some intensive testing about this problem and it turns out that this happens if the CPU-NB is unstable:O this might be a good method to test this because errors occur really quick when setting the FFT time to 0.1.

@radaja: i have a ram stability problem when i run it over 1400mhz, and it always errors @ 8k FFT, but this range is supposed to be a CPU problem indicator. so why does this problem occur at that FFT?
 

·
Registered
Joined
·
1,366 Posts
Discussion Starter #12
Quote:


Originally Posted by ITaoI
View Post

Cool+rep will try that to reproduce errors i get to see if tweaked settings solve it.

if you are able to reproduce this please let us know. i also let a smp client run during priming, it shows errors a lot faster for some reason...
 
1 - 12 of 12 Posts
Top