Prime95 Fails - Now what?

[Konowl]

Complete newbie to this forum here... so take it easy...

Got my new computer setup. Northwood 1.6A, P2B266-C, Antec 1040B, 4 case fans, etc. Purchased 2 sticks of OCZ3000... however, when I run Prime95, it fails after approximately the 4th try. Now, I've isolated the stick that fails Prime 95 - my main question IS, HOW reliable IS Prime 95? Should I take what it says as fact and return both sticks of OCZ? The store I got it from is out of stock on it, so I'd have to get some Crucial PC2100. I can't seem to find any good high quality RAM in the Ottawa area, which is suprising considering the amount of high-tech companies in the area...

So, once again... OCZ stick fails Prime 95 before I even overclock... does this mean it IS bad?

Thanks in advance.

 

[Konowl]

Update:

After taking out the bad stick of OCZ.. thought I'd mess around a bit.

I've currently got the RAM at 2-2-2-5, CPU at 2239 1.6 Volts... wonder how high this puppy can go. I just wish the stupid ASUS Probe program would monitor the CPU temperature correctly... it's consistantly reporting 40 degrees, BIOS as well.

 

[Monaco]

hi Konowl, welcome to the fourms!

if it was me, I would go back to the store and swap the RAM first chance I got, if it wasn't too much of a hassle. Crucial makes good stuff, I wouldn't mind taking that over OCZ.

Prime95 is only one app, but i would trust it for this. If you want, try running memtach or Sandra's memory benchmark loop, and see if they pass it just to be sure.

 

[Konowl]

hi Konowl, welcome to the fourms!

if it was me, I would go back to the store and swap the RAM first chance I got, if it wasn't too much of a hassle. Crucial makes good stuff, I wouldn't mind taking that over OCZ.

Prime95 is only one app, but i would trust it for this. If you want, try running memtach or Sandra's memory benchmark loop, and see if they pass it just to be sure.

Yeah, but how good would normal plain-jane crucial be for overclocking compared to the OCZ stuff? I"ve heard good stuff about OCZ... and then bad stuff....

 

[Monaco]

crucial isn't exactly plain-jane. The modules are made by micron, and sometimes OCZ uses micron modules too- there is a chance the crucial will be the exact same as the OCZ stuff! A lot of folks here have had really good luck with Crucial. Personally I hate OCZ from way back when they were in the scamming business, that might color my opinion a bit.

I have some Crucial SDRAM here that has performed just as good as any high-end stuff I've seen in reviews. I don't have first-hand experiance with their DRR tho.

 

[takiwa]

you could use memtest86 from DOS to check your RAM...but prime is a pretty good indicator of errors...run memtest86 if you are unsure

 

[vw671]

My crucial P2100 does 170MHz @ cas 2.5 while my OCZ PC3000 does 210MHz @ cas 2.5.

 

[sod]

I was just about to ask a similar question. How to test the stability of a given system. Let's take some Windows programs. How many hours should one run SiSoft Sandra's burn-in test, 3Dmark 2001se, prime 95, toast etc... (not at the same time ) in order to be sure his system is stable (and then maybe try to squize some more Mhz)?

 

[Monaco]

hi sod, welcome to the forums!

I usually run stress tests for at least 6 hours before I consider something stable- overnight is even better, if I have the time. A single hour is probably enough if you are still trying to find your max OC, but once you close in you need to be more sure.

That memtest app is very cool, but I don't think it is perfect- my RAM passes that test up to 180MHZ, but fails other tests at 167. It's good to use but you should use a few other apps as well, just to be sure.

The best memory test I have come across is to simply loop 3DMark2001- that stresses everything in the system, including memory. It's good because a system might be stable while the memory is stressed, and while the CPU is stressed, but not both at once.

Or just let your favorite 3D game play all night, a Q3 DM full of bots for 12 hours is a good stress test too.

 

[zervun]

I'd ditch the OCZ and get a samsung 512mb pc2700 chip, they overclock amazingly with little to no variation. I've only heard aweful stuff about OCZ

 

[sod]

Ok, my systems are able to run prime95 for 12 hours, 3Dmark2001SE 1600x1200x32bit for 10 hours and SiSoft Sandra's burn'in for about 40 times without a single crash (I had to inatall Windows to run these tests, although the prime95 test was ran under Linux). In Linux I have a method of my own of stressing the system. I archive over and over again about 500 megs of docs, using the bz2 algorithm. As this test puts teh CPU+mem+IDE on a serious load, I think that any problem should at some point cause a crash in my program. It never did.

When I'll need some more computing power, I'll try to overclock my P4 even more, since I'm running it stable now without any voltage increase.

 

[Freeflyfreak]

Another good testing program is "super Pi"

 

[Crash893]

FFF:
i wasnt aware that super pi had any error correcting abblities
im sure it has some basics

ive been running a prime95 team for about 2 years and i have never had prime wig out on me when there wasnt something wrong with the hardware

perhapse you set it to aggresively

if you could back off to stock and see how long you can runn it

and then keep bumping it up and see how long it takes to fail
( you could graph it if you really wanted to)

the only other suggestion i can tell you is get a program that logs your computer data ( like mbm5) log you voltages and your temps so on

then compare the time of falure in prime with what was going on in the log ( i had a computer once that was failing every day while i was at school couldnt figure it out did the log thing and relized it was going wiggy at about 2 pm so that saturday i made sure to be around at 2 pm and the sun was coming threw the window and bakeing my case)

if you have anyother questions about prime feel free to pm me

 

[Freeflyfreak]

I dont know that super Pi has any error correcting ability. All I know is that if I run the super Pi 32M calculation and my system is clocked too far it will not finish the calculation and give errors such as "non convergent in square" and "incorrect round". The prime 95 I have but the boxes for the self test and the torture test are greyed out, how do you activate these tests?

 

[Crash893]

the nedxt menu over to the left

should say something like password

going off memory may have to look around for it

password=9876

then go back and you should be able to do the tests

i still say the best way to test your system is to do a complete work unit but many people are way to impasont to do that

 

[Freeflyfreak]

Hmmmm, OK put in 9876, now the password selection is grayed out, and the torture test and self test are still greyed out.

 

[Crash893]

screeennnnn shottttttttt

 

[sod]

I don't trust prime95 as a test anymore. I just had a nasty experience
with a Mushkin stick (in fact even after RMA and receiving another stick I
still say this stuff really sucks). Memtest86 (which by the way
is not a DOS app; boy, if you ppl see text mode it's DOS to you) was
giving me constant errors at a single address. At the same time I could run
Prime95 for 22 hours without a single crash! I said to my self
that Memtest86 has a bug and my memory was just fine. It turned out that
memtest86 was right, and the stick was really bad. BTW, it could
also run 3DMark2001SE in loop mode for many hours. Nevertheless, the stick
had a way of random giving kernel panics when doing simple things like
playing mp3's or so.... So the only valid mem test in my opinion
is memtest86. And after a very close look at the source code I know
this for sure. In just a few words, let's just say that the memory
patterns in apps like prime, seti and 3DMarks are not likely to be
affected by some errors in RAM sticks that cause the system to crash when
doing simple things. For details please read the memtest manual.

 

[Crash893]

have you tried the stage 1 and 2 factoring of a primary digit

that always give my comptuer a work out

it uses most of my memory when its doing that (226 mb)