After 15+ hours of Prime95 computer restarts

[Ackbar]

What's the cause? It just restarts, no notice like suddenly failing with errors. I've been monitoring the temperatures and I never exceed 35 degrees Celsius on the CPU! Any ideas on what's causing the random restart?

Take a look at: http://forums.anandtech.com/messageview...atid=28&threadid=1511410&enterthread=y for more information about my situation.

Specs:
Chaintech VNF4 Ultra (newest beta bios)
1 GB of OCZ PC3700 Platinum TCCD
AMD 3200+ Winchester Retail (with retail HSF)
X700 Pro (waiting for preordered X800 XL to get rid of this POS)
36.7 GB WD Raptor + 120 GB WD
Neopower 480W PSU
Silverstone SST-TJ05-ST (with Thermal Controller)

 

[imported_kouch]

how many times did this happen?

 

[Ackbar]

Originally posted by: kouch
how many times did this happen?

This is the second time I have let it run this long and it suddenly restarts. The first time I wasn't so sure since the power might've gone out or something, but once it's happened twice I've been pretty sure that's what it is. Am I o/cing too much? I figure that after 12+ hours without an error would clear that part. Is this indicative of my temps being too high and eventually causing a freeze?

 

[imported_kouch]

what are your temps, you said they stayed below 35C?

btw: no way your load temps are 35C with air cooling, not possible even if you are in antarctica living in an igloo

 

[Ackbar]

Originally posted by: kouch
what are your temps, you said they stayed below 35C?

Yup, never exceed 35C and in fact even under load are usually below that.

 

[imported_kouch]

are you using air cooling? if so you are looking at the "system" temps most likely. that is the temperature of the air inside your case, not your cpu core. Look at the CPU temp.

 

[BW86]

Originally posted by: Ackbar

Originally posted by: kouch
what are your temps, you said they stayed below 35C?

Yup, never exceed 35C and in fact even under load are usually below that.

Looks like even the new Chaintech VNF board has a temp problem. :laugh:

anyway, 14 hours is enough. I usually let prime run 9 or 10 hours for me and I consider it stable. Run loops of 3dmark and see if it fails.

 

[Fern]

Originally posted by: Ackbar
What's the cause? It just restarts, no notice like suddenly failing with errors. I've been monitoring the temperatures and I never exceed 35 degrees Celsius on the CPU! Any ideas on what's causing the random restart?

Sounds like you've got an older version of P95. I had the exact problem and updated to the new version, that solved it.

Fern

 

[Ackbar]

Originally posted by: Fern
Sounds like you've got an older version of P95. I had the exact problem and updated to the new version, that solved it.

Fern

I just downloaded this version about a week ago. Let me check for updates.

Edit: Just checked, definitely have the newest version of Prime95 (v23.8).

 

[Ackbar]

Originally posted by: kouch
are you using air cooling? if so you are looking at the "system" temps most likely. that is the temperature of the air inside your case, not your cpu core. Look at the CPU temp.

I'm using air cooling with the stock HSF. I'm definitely looking at the CPU temp (using nTune Monitor or from the bios gives same temperature at idle) on nTune monitor and using the same "diode" in MBM5 that everyone else is using to get the CPU temps.

 

[Ackbar]

Originally posted by: BW86
Looks like even the new Chaintech VNF board has a temp problem. :laugh:

anyway, 14 hours is enough. I usually let prime run 9 or 10 hours for me and I consider it stable. Run loops of 3dmark and see if it fails.

I need my computer to be totally stable since it's primary usage is for CPU intensive computations for my research.

 

[Fern]

Originally posted by: Ackbar

Originally posted by: Fern
Sounds like you've got an older version of P95. I had the exact problem and updated to the new version, that solved it.

Fern

I just downloaded this version about a week ago. Let me check for updates.

Edit: Just checked, definitely have the newest version of Prime95 (v23.8).

OK, my next suggestion is to put it (computer) back to stock settings (no OC) and uninstall and reinstall P95. If it just kicks off I'm thinking bad install of P95 since you have checked and confirmed you have the new version

 

[Ackbar]

Originally posted by: Fern
OK, my next suggestion is to put it (computer) back to stock settings (no OC) and uninstall and reinstall P95. If it just kicks off I'm thinking bad install of P95 since you have checked and confirmed you have the new version

Ok, I'll try that. In the meanwhile, does something like this seem likely due to instability from o/cing or from something else? It doesn't "feel" like a CPU induced instability since those usually cause the calculation errors. I could be wrong though.

 

[GuitarDaddy]

Reboots are most often RAM or power supply related.

Run memtest for 18-24 hours and see if you get errors, if not try another PS.

 

[Ackbar]

Originally posted by: GuitarDaddy
Reboots are most often RAM or power supply related.

Run memtest for 18-24 hours and see if you get errors, if not try another PS.

Thanks for the info! :thumbsup: I'll run memtest overnight since the computer is busy doing it's thing right now and can't be spared at the moment.

Unfortunately, I can't test another PS since I'm making a huge leap on building this computer from my previous computer and parts aren't compatible anymore...

But I just bought a Neopower 480, so I would hope that it would be good.

 

[Duvie]

Do you have automatic restart turned off in XP??? It may have masked a general system error with the rstart...

Also what test do you run??? I would recommend the one with little ram is tested...use memtest86 to test ram more efficiently. Ram test in prime is a horrible cpu stress tester since the cpu does not say pegged at 100% usage and thus does not generate the heat needed to truly stress it...

 

[Ackbar]

Originally posted by: Duvie
Do you have automatic restart turned off in XP??? It may have masked a general system error with the rstart...

Also what test do you run??? I would recommend the one with little ram is tested...use memtest86 to test ram more efficiently. Ram test in prime is a horrible cpu stress tester since the cpu does not say pegged at 100% usage and thus does not generate the heat needed to truly stress it...

Good point! I am just running the normal "primality test". Should I be running the Torture test that uses little ram instead? I believe that automatic restart is enabled. So you're right, I may have gotten a BSOD that I don't know about.

 

[Duvie]

Go into control panel - system - advanced - startup and recovery - uncheck the box...

 

[Ackbar]

Originally posted by: Duvie
Go into control panel - system - advanced - startup and recovery - uncheck the box...

Hehe thanks. Just did it before you responded.

 

[djspl]

Originally posted by: Ackbar

Originally posted by: BW86
Looks like even the new Chaintech VNF board has a temp problem. :laugh:

anyway, 14 hours is enough. I usually let prime run 9 or 10 hours for me and I consider it stable. Run loops of 3dmark and see if it fails.

I need my computer to be totally stable since it's primary usage is for CPU intensive computations for my research.

Well, run your computer within spec if it's that important. In the world of overclocking, 14 hours of prime is plenty stable. Diagnosing a "post 14 hour" prime crash is totally hit and miss. Trying to keep your computer from crashing by running an easier test seems a little backwards when your ultimate goal is 100% stability for "research".

I agree about your temps, you are not running at 35C with stock air cooling, overclocked, after 14 hours of prime95. Your sensors are misconfigured, miscalibrated or broken. My P4 1.7 ghz Williamette running at stock runs 7C more after 30 seconds of prime.

 

[Shimmishim]

which prime test are you running?

the small fft's? blend? custom?

in order to test purely cpu stability, you're supposed to use small fft's and use memtest to test your ram

 

[Ackbar]

Originally posted by: djspl
Well, run your computer within spec if it's that important. In the world of overclocking, 14 hours of prime is plenty stable. Diagnosing a "post 14 hour" prime crash is totally hit and miss. Trying to keep your computer from crashing by running an easier test seems a little backwards when your ultimate goal is 100% stability for "research".

I agree about your temps, you are not running at 35C with stock air cooling, overclocked, after 14 hours of prime95. Your sensors are misconfigured, miscalibrated or broken. My P4 1.7 ghz Williamette running at stock runs 7C more after 30 seconds of prime.

From what I have read, there's no guarantee that you'll be Prime95 stable even at stock speeds! Since my computations are CPU intensive, I'm trying to compromise between speed and reliability. If once every 30 days my computer has an error, no big deal. The point is that I want to have a high level of reliability.

Edit: Actually, I think my temperatures are too low as well! But I've been speaking to others with my mobo and have been told that this is what they report as well...
It seems that the temps are just reporting low uniformly across everyone who has this board. So I believe my sensors are fine physically, but moreover the general placement on these boards needs to be fixed (something I can't physically do). But, I think the temps are a good indicator to compare against your own numbers not necessarily against other people's numbers. I remember that in car tuning everyone would place an O2 sensor in a different spot and try to compare numbers, but really all you need to know is how your own numbers change. And although an absolute reference point would be nice, in this case, it's not possible.

 

[Ackbar]

Originally posted by: Shimmishim
which prime test are you running?

the small fft's? blend? custom?

in order to test purely cpu stability, you're supposed to use small fft's and use memtest to test your ram

I originally was just running Prime95 the regular part, now I'm running the small FFT test and I'll follow with memtest for the ram later.