Prime95 fails only AFTER 24 hours up.. need ideas!

[David James]

Hi all,

 

I am having some interesting overclocking issues, and need some older and wiser heads to guide me.

 

Cut a long story short, I am consistently failing Prime95 - but only AFTER 24 hours! (I run it in the background all the time, and work in the foreground)

 

Failed after 47 hours mixed use (mostly prime, bunch of other tests) at 2684MHz

Failed after 30 hours continous use at 2640MHz

 

 

* I have never had a single error in any memtests.

* ONLY fails while I am working at the machine - never overnight.

* Can play games as long as I want, passes every torture test I can throw at it - except long term Prime95

 

It smells like something other than just too little voltage, it seems a bit weirder than that. I have tried a few memory timings to no avail, but there are so many variables to test there...

 

I saved my MBM high low page after the 2 day failure, and all my reported voltages had varied within 1.6% ,which I thought was good. (vcore-0%, LDT-1.6%, 3.3v=0.9%, 5v=1.6%, 12v=1.1%

 

The trouble is, I don't know how long this has been a problem, because I haven't run Prime95 longer than a day until this week - I assumed if it made it to 24 hours it was stable.

 

So right now I am going back to stock speed and will try that for a few days straight to try to eliminate if it is an overclock problem and narrow it down from there. But while I wait...

 

But has anyone seen something similar? or do you all quit Prime95 after 24 hours and call it done?

• 
  

[Angry_Games]

why are you worrying about this is my question...you've been past 24 hours of Prime95...do you ever plan on using your machine for 35 hours straight without a break, punishing it with 100% cpu useage and 98% memory useage at the same time during this 35 hour period that it will eventually crash in?

 

the Overclocking Database only requires 8 hours of prime95, because honestly no one I know sits and uses 100% cpu and memory for 8 hours straight.

 

the Stock Speed Database will require 24 hours of Prime95...just because it had damn well better be able to do 24 hours @ stock speeds.

 

but

 

no amount of prime95'ing will guarantee you true stability...no amount of benchmarking or testing will guarantee you stability...

 

thats why we choose to run at least 8 hours...if your rig can pass torture for 8 hours and still do all the 3d benching...that is going to be enough that you will probably not see a system crash unless you...install something bad, get a virus, or do something bad....

 

I assumed if it made it to 24 hours it was stable.

 

you assumed correctly.

 

stop fooling with it and begin enjoying it (you did pay a good chunk of money for your rig yes? did you spend all that money to worry over failing prime95 after 30+ hours? or did you spend all that money to play games, or whatever you bought it all to do?)

[David James]

why are you worrying about this is my question...you've been past 24 hours of Prime95...do you ever plan on using your machine for 35 hours straight without a break, punishing it with 100% cpu useage and 98% memory useage at the same time during this 35 hour period that it will eventually crash in?

 

Ummm yes actually. And have done so since 1997. Surely you have heard of 3D animation?

 

I realise most of your customers are insane modding gamers, but dual core is opening up new markets – namely 3D animators tired of expensive “workstations”? (double the price for less gain)

 

For our renderfarm ALL we care about is bang for buck. Right now my renderfarm is dual Xeons and dual AMD MPs. But dual core drastically alters the price performance of our hardware costs, and I am looking to upgrade our render farm. With overclocking, that opens up a new factor that can save/gain another 10-20% Which is why I am testing it so much before I buy more machines.

 

Back in the early Win2000 days (before Windows Update) I could log uptimes of months on my main work machine. (2 months 2 days was my record) And that was Photoshop, Digital Fusion and 3DSMAX during the day, and usually always a 24/7 background render of 3Dsmax.

 

So in answer to your question, yes. My CPU load would be close to 100% day in day out, and 1-3G of RAM in use. Only reboot when I need to.

 

the Overclocking Database only requires 8 hours of prime95, because honestly no one I know sits and uses 100% cpu and memory for 8 hours straight.

 

Well, allow me to introduce myself… ;-)

 

Our network renders are running 24 and 7 when we have animation jobs on. (most of the year) 20% overclock means a sequence that takes 7 days became a 6 day render. A good thing.

 

thats why we choose to run at least 8 hours...if your rig can pass torture for 8 hours and still do all the 3d benching...that is going to be enough that you will probably not see a system crash unless you...install something bad, get a virus, or do something bad....

 

Great, that is what I need to know.

 

stop fooling with it and begin enjoying it (you did pay a good chunk of money for your rig yes? did you spend all that money to worry over failing prime95 after 30+ hours? or did you spend all that money to play games, or whatever you bought it all to do?)

 

Actually I am still amazed I paid about half what my last workstation cost. Don’t worry, I am enjoying it, it’s a dream compared to my last machine, noisy dog of a dual Xeon. This is background testing between jobs.

 

I am new to DFI and Prime95. If the answer is “one Prime95 error every day or three is perfectly acceptable”, then great. But I have not heard anyone say that here – and you get a lot of variety in the advice given here! I want to know before I buy several more like it. And having Corsair ram makes me paranoid. (I started out in August on a ASUS A8N SLI that was a disaster, and swapped MB to DFI)

 

So I will see how my stock testing goes, but if I still fail Prime95 at stock, that’s a showstopper - right?

[Angry_Games]

here's a serious answer that you should consider very much...

 

if you are needing this for mission critical things like rendering...then you really need to look into not overclocking at all. Period.

 

I can never advocate overclocking when you need the machine for ANYTHING serious. Period.

 

too many times I get users here that start screaming and crying about how our board is costing them money because they decided on a Lanparty board for a mission critical use like a server, a business machine, etc...and that is about the dumbest decision anyone can make.

 

You want a mission critical workstation? Use a Pentium4 Xeon, or Opteron system with registered, ECC memory, and forget you ever heard of overclocking.

 

If you can't take this advice seriously, then...I don't know what to tell you. I've been doing this for years and years and it still amazes me the crazy (and usually stupid) things that people try to get away with, when the 'mission' is very critical and very important.

 

don't take this as an insult to you...it isn't. This is one serious tech to a very serious sounding user who needs a solid stable machine. I think you know where I am coming from. Bust that overclock back to stock speeds and forget you ever thought of overclocking if it is as serious as you tell me it is.

 

I am new to DFI and Prime95. If the answer is “one Prime95 error every day or three is perfectly acceptable”, then great. But I have not heard anyone say that here – and you get a lot of variety in the advice given here! I want to know before I buy several more like it. And having Corsair ram makes me paranoid. (I started out in August on a ASUS A8N SLI that was a disaster, and swapped MB to DFI)

 

no errors in prime95 are acceptable, no matter how long you run it. Period. 24 hours is perfect for most of us, as again, no one other than a VERY FEW SPECIALIZED USERS (who really should have bought a P4 or Opteron workstation setup and put overclocking out of their heads completely) uses their machine @ 100% cpu/memory for more than a few hours at a time, and no one uses it for 24+ hours at a time (other than again, the few specialized users who should have avoided any kind of 'tweaker/enthusiast' setup to begin with).

 

Having corsair memory...well...thats probably not so bad if you can get 35 hours of prime95 out of it.

 

So I will see how my stock testing goes, but if I still fail Prime95 at stock, that’s a showstopper - right?

 

at risk of repeating myself too much...yes...you should have avoided any enthusiast setup completely and went with a Xeon/Opteron workstation with ECC registered memory for what you are needing it for. (this is how you got away with only paying half of what your normal systems usually cost...because professionals like you rely heavily on stability, they don't overclock, and they run workstation-class rigs like Xeon/Opterons)

[MsB]

This question is for david james and anyone. The other day I was in my nearby computer shop(friends) and there was a guy there that had a new dual opteron work station that he was getting help building and it was for the purpose of rendering video. This was no cheap setup dual 180's i think and 4 gigs of ram and a 7800gt vidcard. So my question is this. Why if in most new video cards the gpu does the work now and not the cpu do you video guys end up buying dual systems and only one video card? It seems like a nice SLI gtx setup would improve times if its true the gpu does all the work. Sorry I saw this thread and couldnt help asking a question that has been nagging me to ask for some time.

[Funkytown]

I'd have to agree....if stability is that critical to you, you shouldn't be overclocking.

 

If my machine runs prime for 1 hour, I consider it stable - because in reality I never have the CPU at 100% load for an hour. Even when playing games, the matches end and begin again, which gives the CPU a break.

 

I'm confident my computer is stable, but would I do that to my work machine - no!

 

IMHO, overclocking should only be a hobby for enthusiasts, not professionals who rely on the machines to make a living.

[soundx98]

I'm not too convinced that Prime95 is actually doing anything if you are running it in the background all the time. If using the default priority of 1 I'd think it was only priming when you were doing nothing else.

 

I'm with AG, don't believe "mission critical" components should ever be OverClocked.

What version of Prime 95 are you using? The latest P95v2414? And is this on Blend or different settings?

[grunjee]

I'm 100% with AG on this one... and no offense at all to you. But overclocking inherently means taking a risk. On the one hand you basically claim 'I can't take any risks, this is mission critical', yet by your overclocking you're taking a risk... it really doesn't make sense.

[Pyr0]

hehehe my cpu is at 100%, 99% of the time

i prime95'd for 48 hours to check stability

[ROcHE]

This question is for david james and anyone. The other day I was in my nearby computer shop(friends) and there was a guy there that had a new dual opteron work station that he was getting help building and it was for the purpose of rendering video. This was no cheap setup dual 180's i think and 4 gigs of ram and a 7800gt vidcard. So my question is this. Why if in most new video cards the gpu does the work now and not the cpu do you video guys end up buying dual systems and only one video card? It seems like a nice SLI gtx setup would improve times if its true the gpu does all the work. Sorry I saw this thread and couldnt help asking a question that has been nagging me to ask for some time.

 

Because the cpu does the render, not the video card.

[Angry_Games]

cpu calculates triangles and positions, the gpu fills in with textures, shading, lighting, shadows, etc

[David James]

Thanks Angry, I hear you. I’m not insulted, straight talk is good. I realise overclocking is taking a risk, and you will never hear me classy lady about DFI ate my data and I want to sue blah blah. Think of it as wanting to learn the rules before I bend them a little, with my eyes open. And we are pretty specialised, I agree.

 

People here seem to taking this as either black OR white, of risk/no risk but IMHO overclocking is a exponential curve of calculated risk. I don’t plan or living past the toe of that curve. Right now I am learning and plotting that curve.

 

A couple of points:

 

My previous career as a photographer taught me to never have a single equipment failure cost you the job. Two of everything, and backup. If I overclock, I assume I will break something and plan for that. It is not like we are unfamiliar with hardware or software unreliability! Sheesh ,we are on Windows!

 

If this machine melts and dies I won’t enjoy moving down the bench to my noisy 2 year old Xeon, but it is there. We have multiple backups of our data. The server is NOT overclocked! Because of past experience with burglary, I have designed it so that ANYTHING can happen to any of our machines and we will not go broke. (ok, if we burnt down completely we might have trouble with a few deadlines, but any one or two boxes can go bang)

 

People are putting words in my mouth saying 'I can't take any risks, this is mission critical'. I never said anything of the sort. In fact I said bang for buck is the most important thing for us in hardware.

 

Our data is fuzzy! It is pretty pictures. It’s a bunch of pixels. If it looks right it is right. It is not as if a zero turns into a one a patient dies or the space shuttle explodes or we get sued into oblivion.

 

Unlike the vast majority of users for us there is actually a productivity benefit to overclocking. Most computers CPUs would be near idle for most of the day. Hence a 15% boost makes an almost zero boost to their productivity. In our game, given that we run all day and night, 10% is noticeable.

 

I think my wistful talk of the uptimes I used to get has been giving the wrong impression – I do not NEED machines that have continuous uptimes of months! (Not to say I don’t appreciate that). You can forget that running Windows XP, no matter how stable the hardware. When we render an animation, every frame rendered on the farm is a discreet chunk, and independent of the others. The renderfarm machines do one thing and one thing only, and it is easy to get them stable for that. (no sound or video issues etc) I remember working on MacOS in the bad old days before OS-X was even a dream, and then you lost a lot of time and hair due to constant crashes. But you didn’t go broke. We coped.

 

Run the numbers. If a 15% overclocked render box is producing valid frames and not crashing, it could be out of action for nearly four hours every day and still be equal with it’s stock speed twin.

 

Our philosophy has always been to leverage consumer or near consumer level gear. This is possible because we are a tiny company and can do it ourselves. (dammit, it is fun. Why have your own company if you can’t play with your toys?) Given the unreliability of computers generally, I would now rather 2 cheap loaded PCs than one workstation. Far better bang for buck. (Having owned and run both)

 

I have to disagree on the ECC RAM thing (for a workstation) I have not noticed a much better experience using “workstations” versus “PCs” over the years. What you gain in ECC you lose in flaky proprietary drivers, and you are still running Windows. The software you use is by far the weakest link. YMMV.

 

Stock for me is 2200MHz. I know that there is a point somewhere between 2201 and 2650 that will work perfectly for me. And as we speak I am 27 hours into a Prime95 test at stock, so soon I will know if my original post was a overclocking problem, or something I actually need to worry about.

 

PS: Soundx98:– while testing I am not doing 24hrs a day rendering – just the interactive foreground work – so prime95 would get probably 70% of a day’s CPU.