Is Prime95 a reliable hardware test?

[daemorok]

I have been having random BSODs some with errors, some where the screen just turns blue and, thats it. They usually happen when gaming (Quake3 rocket areana, and Counter-Strike).

The windows install was fresh as of about a month ago, and ive crashed 3 or 4 times since then. My windows is tweaked a little, but the only major "hack" is the theme dll swap so i can run themes without styleXP or whatever.

So hardware specs:
A7n8x ultra400 standard.
AMD XP 3200+
2x512 Corsair 3200LL 2-2-2-6
Sapphire 9600xt
SB Audigy1
Winfast TV 2000 XP Deluxe (TV tuner)

WD 250 jb
WD 200 jb
lite-on DVDROM
lite-on 52x CDRW

enermax 550w PSU

Nohting is OCd or has been.
The ram is running at 2.7v

I have run memtest, goldmem, and msinst for at least 8 hours each on seperate occasions, with the ram running stock except for the 2.7v
and have never gotten any errors.

When I run prime95(version 23.8.1) tourture test It always fails somewhere along the line.
It could take hours, or minutes, or seconds. Lately it dies on the first test.

When I BSOD from gaming its a random amount of time, I have had it crash minutes after i load the game, and it has crashed after playing for over an hour.

I am thinking that maybe I should get new ram. I have had this ram for awhile, and I have changed the mobo and CPU and still get these issues.

[Wombat]

If memtest (you're using the latest memtest86+ ?) passes, and Prime95 fails, then it is probably not RAM. Prime95 works your CPU, and specifically the FPU and onboard cache, quite hard, but not much else.

Your symptoms sound like the might be heat related. Prime95 will really burn wattage on the CPU, and intense gaming makes the whole system hot. These stresses may be pushing your RAM or other components into a dubious operating range.

[daemorok]

Actually according to the Prime95 documentation the "blended tourture test" which i was running is very ram intensive.

I am currently running the least ram intensive test and it has not failed yet.

I seriously doubt it is a heat issue.
CPU HSF is a swiftech mcx462V with a 92mm vantec tornado
Northbridge has a swiftech MCX259 on it
The case (lian li PC65B) has 4 sunon 80mm fans that move about 50cfm each (100in and 100 out).

according to my bios my CPU tops out around 40C under load and my case temp is usually around 27C-35C

I am not trying to argue with you over the heat, I understand that if it was overheating it could cause this problem, I just dont think it would overheat with my current cooling setup.

[Greebe]

I'll second what Wombat said. Sounds like a cpu heat related issue.

[daemorok]

40C under full load... how is it a heat issue still?

I mean I understand that if it was overheating that it could and probably would cause this problem. But at full load on prime or anything for that matter the CPU tops out at 40C and the case 32-35C depending on room temps..

[Wombat]

With all those numbers, the RAM could still be getting too hot if it's not in the path of the airflow - and it usually isn't. It's usually in the upper front (with the heat from the HDs piling on it), and air flows from lower front to upper rear.

Try lowering the RAM voltage, and/or relaxing your timing.


The one time I saw failures much like yours was when I was trying to OC. It turned out that 99% of my CPU could handle the OC, except for one section of the L2 cache.

Symptoms:
* Games would blue screen when they hit the right address in cache.

* Prime95 would bomb out eventually

* Compressing movies to DivX would result in noticible artifacts in the final file.

*memtest would pass.



Run memtest86+ for like 12hrs or so. Unplug your case fans if you like to accelerate the heat process.

[daemorok]

I will try memtest at the stock timings and
voltage.

The memory tester I have been using mostly is called mtinst.exe, it can be found on the microsoft site.

The reason I have been using that is, I have seen cases where ram passes memtest but fails mtinst.exe

I have run mtinst.exe at
2226 2.6v and 2.7v
2326 2.6v and 2.7v

but never for more than 8 hours

There is a decent amount of airflow over the ram becuase the case has a top exhaust fan that is located directly above the ram slots. However you are probably corrent, I believe that is the hottest area in the case.

[Kurt]

Originally posted by Wombat
The one time I saw failures much like yours was when I was trying to OC. It turned out that 99% of my CPU could handle the OC, except for one section of the L2 cache.
.

I'm curious as how you managed to figure that out. ?!

[Ribbit]

Perhaps he had one of those CPUs with half the cache locked, and had unlocked it?

Please do tell, Wombat.

[Wombat]

Well, first off, I've worked in processor design, and cache testing specifically on the PA-8700. So lots of experience on blaming specific parts of a CPU according to how things failed.

I had a system where changing the CPU speed would cause a problem.

The functional units all would pass tests okay. But I would get memory errors sometimes, or corrupt data (and the RAM was ECC). The video glitches were the biggest hint.

[Jon P. Inghram]

Sounds rather odd, but at least for my system Starcraft is by far the most sensitive program to overclocking.

I had ran Prime95 for hours and never had any indication that there was any problems, but Starcraft would lock up after playing for a few minutes. Either bumping the voltage up a notch, or the FSB down a single MHz made it run stable.

[UtwigMU]

Game crashes and Prime95 would indicate a hardware problem. What's at fault though it's up for you to find out:

Like others have pointed out:
- memory, voltage timings
- heat, what is your case temperature, what's the airflow like, is case clean, which fans are where
- heat - cpu heatsink clogged, poorly applied thermal compound, bad fan, heatsink not good enough
- Power supply - a bad power supply can cause such failures - for instance I had a case, where replacing crappy 200W PSU with good quality 300W PSU cured me of Memtest errors, when I installed 512MB of PC133 on old BX box.
- bad capacitors on motherboard


- Visially inspect motherboard for bulging or leaking capacitors.
- Set memory timings to loose and voltage to default, also set CPU clock to default. Run memtest overnight and Prime95 for an hour
- If you still have problems, clean case, reseat heatsink, make sure you clean it using acetone (100% pure, not nail polish remover as it leaves residue) or isopropyl alcoho and use good quality thermal compond. Be careful, don't turn on computer without heatsink. Also make sure heatsink is actually rated for Athlon XP 3200+. Leave side off and run Prime95 overnight.

If it still has problems, try stripping down system and also try better power supply. If you still have errors, it's time to start swapping out CPU, motherboard and RAM.

[daemorok]

Interesting info:

when running the prime95 Blended (RAM intensive) test if I have the ram voltage set higher it takes longer for the test to fail.

My bios only has 3 settings: 2.6, 2.7, and 2.8

at 2.8 it takes almost 5 minutes for the test to fail but at 2.6 it fails almost instantly.

my 12v adn 5v rails drop 0.1 under load (prime95)

12v from 12.3 to 12.2
5v from 4.9 to 4.8

[Kurt]

You could try borrowing another PSU...


@Wombat: do you have pointers for freely available test programs?

[Wombat]

Memtest86+ has already been mentioned. Running a video encode is pretty intensive.

http://www.ultimatebootcd.com/ has handy tests on it.