Prime 95 & stability questions

[MoeStooge]

I'm having some trouble running Prime95 with a new system:

P4C 2.6GHz (SL6WS) - currently using stock fan/hs
2x512MB Mushkin L2 PC3500
Asus P4C800-E
WD360GD SATA HD
Geforce2 MX - old card I stuck in temporarily

Once I had everything hooked up and all the Windows XP updates installed, I decided to try a quick overclock just to check how well it did with the stock fan. I bumped it up to 3.2Ghz (without adjusting any voltages) and it booted into windows without any problems. I ran the Prime95 torture test and then switched over to my other PC to play CS for a couple of hours. I switch back and it is still chugging away. The temperature seemed to max out at 58, but averaged in the upper 40s so I figured it was safe to let it run overnight. I come in to check on it in the morning and Prime95 has crashed- giving me the "Would you like to report this error?" XP dialog box. I figure that it's no big deal- I haven't purchased a new HS/FAN yet and I didn't bump up the voltages. I decide to set everything back to the defaults and run Prime95 again just to make sure my system is stable at its rated settings. Same thing happens- it runs fine for 3 hours, I go to sleep, next morning Prime95 has crashed again.

If my system is not stable at default settings and Prime95 crashes, do I have a hardware problem? I might be able to swap in some different memory but other than that I'm not sure how I could isolate the cause of the instability. I don't have another P4 motherboard or another P4 to try. Anyone out there have some ideas for me?

 

[CTho9305]

www.memtest86.com - it's worth running to check.

 

[MoeStooge]

I've isolated it to the memory. One of my sticks is bad. Since I bought the "matched" pair I'm gonna have to send both back to Newegg.

 

[Shimmishim]

also, when running prime95 for stability, you need to run 2instances of it.

because HT, xp and 2000 recognize HT processors as 2 processors for some reason or another...

1 instance only stresses your cpu 50%.

2 instances does 100%


if you do however want a better test of stability, try doing some encoding with a program like TMPGen

from my experience, it'll lock up your cpu in 15 minutes if it's unstable or run for a few hours before you get a nice reboot.

i really like using encoding now as a form of testing stability... thanks to pj, duvie, and thugs who gave me the idea to test stability this way!

 

[ponyo]

When running Prime, make sure you set your Priority to 10 instead of the default 1. You can find Priority under Advanced setting. You will need to enable password if the option is not enabled. Password is 9876. All this can be found in the readme file.

Running Prime with Priority set to 10 is proper way to stress. It places maximum stress on your system and is the true test. Most people run Prime with default priority of 1 which is like doing reps with 10 lb weight instead of 200 lb.

Credit goes to Hoot at Overclockers forum for pointing this out.

 

[MoeStooge]

Thanks for the info. As soon as I get the new memory in I'm gonna give it another go.

I would try TMPGen, but I don't have any video capture devices or video to encode.

 

[silencerius]

Get a dvd, rip the VOBs off using DVD Decrypter, download DivX 5.1 or 5.1.1 and encode the VOBs twice simultaneously at 1000bps/slowest using TMPGenc or Virtual Dub Mod or anything you like. Another way of stress testing that I tend to like, is running Super Pi @ 32M and video encoding. If it ain't stable, it's gonna crash

Running two instances of Prime95 and having one of them test 1st CPU and the other the 2nd CPU is very good in my humble opinion. Although I guess your CPU will never be under so much load by an ordinary app (like encoding etc) and I think that's why Prime is so controversial.

 

[MoeStooge]

Update:

I received my replacement memory on Monday. I pop it in, run a single instance of prime95, and the program crashes two hours later. I try each memory stick individually and it still crashes. So I'm thinking there's no way both replacement sticks are bad. I decide to burn memtest86 to a CD and use it to test the memory. Within 5-10 minutes it locks up with both sticks installed or just one at a time. I try a friend's Corsair memory and it does the same thing (no matter what mem settings I use). After wasting many hours messing around with memtest I read on these forums about disabling USB legacy support (wtf?) and Voila! my replacement memory passes the test. I run it for 8 hours with no errors.

I decide to try Prime95 again. It crashes after two hours again. So now I'm thinking it's got to be a motherboard or CPU problem. I've read a few blurbs about the voltages being flaky on the ASUS P4C800E so I decide that maybe I should start playing with the CPU voltage even though I'm not overclocking yet and just using standard settings. Instead of setting everything to AUTO I decide to manually set the Vcore to it's default level and try Prime95 again. This time when I try it I decide to run two instances (as many have suggested) with the hope that it will crash faster if there is a problem. When I run two instances of the blended test I notice that the CPU is only at 50% usage and that the temperature is basically the same as it is at idle. The test seems to run much slower too. Since I'm interested in testing the CPU anyway, I decide to not use 'blend mode' and pick the smaller FFT option that is primarily for CPU stressing only. I figure that memtest86 has already proven that my ram is good. Running two instances of Prime95 using small FFTs gives me 100% CPU utilization and brings the temp up to a steady 54C with the stock cooler. I let it run for 12 hours and have absolutely no crashes or errors at all.

So what the hell is going on?

When I go home tonight I'm going to narrow down whether the manual setting of the Vcore fixed my problem or if my problem is only with blended mode. I just want everything to work out so I can starting overclocking this damn thing.

Also, when running two instances of Prime95 in blended mode the swap file gets huge (= 1.5 GB). It makes me wonder if my problem could be related to the HDD or the SATA interface on the motherboard. I might check the swap file size when running one instance in blended mode.