funky stress test computer issue

AladainAF · 09-15-2007, 09:52 PM

I am burning a system in a stress test in linux using mprime on each of the 4-cores (Q6600 G0 proc). If I run mprime on 3 cores (no matter which three) then it runs fine, but as soon as I start it on the 4th, one of the 4 instances at random will error out. Then the other 3 will run forever.

I am using the stock cooler with this CPU, but doing absolutely no overclocking. I'm using an Asus P5N-E SLI board (newest BIOS), 500W power supply, 8GB of RAM and a super-cheap video card (7200GS) running fedora 7 64-bit (newest kernel).

I don't think its thermal because the CPU doesnt even reach 55c. I don't think its memory because memtest86+ runs all night with no errors. Its not CPU because if I swap it with another Q6600 the same problem occurs (I have the B3 and the G0 stepping, both do this).

Anyone have any ideas? Perhaps an mprime error?

Requiem · 09-15-2007, 09:55 PM

Sounds like it's probably a software problem. The only base you haven't covered is the motherboard. I'd recommend coming up with your own synthetic tests that can get you up to 400% CPU usage. Hey, I know! Join the uberguilds rosetta@home team! =P

AladainAF · 09-15-2007, 10:00 PM

heh well this machine is going to be multiplied times 40 and put into a cluster doing number crunching on all 4 cores 24/7. Thats why I'm somewhat concerned, but don't have the exact SW they will be using. I don't know what else will peg all 4 cores other than mprime.

Also the failure is not immediately -- it might do 1-2 more tests and then error, but sometimes its happened immediately.

I'll keep toying with it.

Requiem · 09-15-2007, 10:55 PM

...If you're getting 40 of them, might I suggest xeons and supermicro rackmount systems(or dell servers really are worth it if you can afford them)? Using some shitty consumer motherboard with non-registered/ecc ram for that sort of application is nuts. WAY more trouble than it's worth in savings.

AladainAF · 09-15-2007, 11:14 PM

Thats the plan when the business gets bigger. Right now, on a limited budget. We can get more processing power out of consumer crap for the money than getting some high end shit.

Needless to say, after a few hours, I figured it out. The memory in the system needs to be clocked to 2.2v. I left it at its stock (2.1v) since I was doing no overclocking, but apparently it wanted more. Been running longer now than ever. Oh well.

09-15-2007, 09:52 PM	#1 (permalink)
AladainAF Registered User Join Date: Aug 2002 Location: Texas Posts: 1,966	funky stress test computer issue I am burning a system in a stress test in linux using mprime on each of the 4-cores (Q6600 G0 proc). If I run mprime on 3 cores (no matter which three) then it runs fine, but as soon as I start it on the 4th, one of the 4 instances at random will error out. Then the other 3 will run forever. I am using the stock cooler with this CPU, but doing absolutely no overclocking. I'm using an Asus P5N-E SLI board (newest BIOS), 500W power supply, 8GB of RAM and a super-cheap video card (7200GS) running fedora 7 64-bit (newest kernel). I don't think its thermal because the CPU doesnt even reach 55c. I don't think its memory because memtest86+ runs all night with no errors. Its not CPU because if I swap it with another Q6600 the same problem occurs (I have the B3 and the G0 stepping, both do this). Anyone have any ideas? Perhaps an mprime error?

09-15-2007, 09:55 PM	#2 (permalink)
Requiem Site Administrator Join Date: Jan 2002 Location: Cambridge, MA Posts: 910 +37 Internets	Sounds like it's probably a software problem. The only base you haven't covered is the motherboard. I'd recommend coming up with your own synthetic tests that can get you up to 400% CPU usage. Hey, I know! Join the uberguilds rosetta@home team! =P __________________ Requiem Alloria Mistweave Uberguilds.org, fohguild.org Site Administrator requiem@fohguild.org

09-15-2007, 10:55 PM	#4 (permalink)
Requiem Site Administrator Join Date: Jan 2002 Location: Cambridge, MA Posts: 910 +37 Internets	...If you're getting 40 of them, might I suggest xeons and supermicro rackmount systems(or dell servers really are worth it if you can afford them)? Using some shitty consumer motherboard with non-registered/ecc ram for that sort of application is nuts. WAY more trouble than it's worth in savings. __________________ Requiem Alloria Mistweave Uberguilds.org, fohguild.org Site Administrator requiem@fohguild.org

Thread Tools	Search this Thread
Show Printable Version Email this Page	Search this Thread: Advanced Search
Display Modes	Rate This Thread
Linear Mode Switch to Hybrid Mode Switch to Threaded Mode	Rate This Thread:

09-15-2007, 10:00 PM	#3 (permalink)
AladainAF Registered User Join Date: Aug 2002 Location: Texas Posts: 1,966	heh well this machine is going to be multiplied times 40 and put into a cluster doing number crunching on all 4 cores 24/7. Thats why I'm somewhat concerned, but don't have the exact SW they will be using. I don't know what else will peg all 4 cores other than mprime. Also the failure is not immediately -- it might do 1-2 more tests and then error, but sometimes its happened immediately. I'll keep toying with it.

09-15-2007, 11:14 PM	#5 (permalink)
AladainAF Registered User Join Date: Aug 2002 Location: Texas Posts: 1,966	Thats the plan when the business gets bigger. Right now, on a limited budget. We can get more processing power out of consumer crap for the money than getting some high end shit. Needless to say, after a few hours, I figured it out. The memory in the system needs to be clocked to 2.2v. I left it at its stock (2.1v) since I was doing no overclocking, but apparently it wanted more. Been running longer now than ever. Oh well.