Is Prime 95 really a useful stress benchmark?

[craftech]

I ask this question because I have used this stress test and monitored temperatures, yet I can't find a single application that I use which even comes close to generating the temperature levels Prime 95 does.

So why stress the system way more than it will actually be stressed under normal use?

John

 

[Flipped Gazelle]

Originally posted by: craftech
I ask this question because I have used this stress test and monitored temperatures, yet I can't find a single application that I use which even comes close to generating the temperature levels Prime 95 does.

So why stress the system way more than it will actually be stressed under normal use?

John

Many people agree with you.

Some folks are strict about what "stable" is - they demand that their PC is able to run 24/7 @ 100% CPU load. As you've pointed out, most people never come anywhere near this, but they do like having the overhead.

However, some fairly common applications really do work the CPU for long periods of time. Video processing and batch image processing can monopolize CPU cycles. My g/f spends some time in Second Life, and that will run both of her PC's CPU cores at over 70% constantly - I mean for hours - and often at 90% or higher. Heck, I installed SL on my PC and spent about an hour looking around, and the whole time all 4 cores were bouncing between 70-93% usage.

 

[craftech]

Originally posted by: Flipped Gazelle

Originally posted by: craftech
I ask this question because I have used this stress test and monitored temperatures, yet I can't find a single application that I use which even comes close to generating the temperature levels Prime 95 does.

So why stress the system way more than it will actually be stressed under normal use?

John

Many people agree with you.

Some folks are strict about what "stable" is - they demand that their PC is able to run 24/7 @ 100% CPU load. As you've pointed out, most people never come anywhere near this, but they do like having the overhead.

However, some fairly common applications really do work the CPU for long periods of time. Video processing and batch image processing can monopolize CPU cycles. My g/f spends some time in Second Life, and that will run both of her PC's CPU cores at over 70% constantly - I mean for hours - and often at 90% or higher. Heck, I installed SL on my PC and spent about an hour looking around, and the whole time all 4 cores were bouncing between 70-93% usage.

Video Editing is what I do for a living. Some complex renders can take 10 hours or more to complete. But, even utilizing all four cores with renders in Procoder 3, Sorenson Squeeze 4.5, or Sony Vegas the temps don't come anywhere near the temperatures generated by Prime 95.
However, I don't game so maybe the games stress it more than video rendering does. Not sure.

John

 

[J van E]

Interesting question, because I also wonder about that. I have my i7 920 OC'ed to 3,33 with the stock cooler and so far temps are great. Only when I use Prime temps get higher (but still not too high).

 

[ajaidevsingh]

In my book Prime 95 falls under the so-so category since now a days i solely use LINPACK an it seams better.

 

[Flipped Gazelle]

Originally posted by: ajaidevsingh
In my book Prime 95 falls under the so-so category since now a days i solely use LINPACK an it seams better.

We're not really discussing Prime vs something else, we're discussing the applicability of what Prime/Linpack/whatever does vs real-world usage.

 

[Idontcare]

Originally posted by: craftech
So why stress the system way more than it will actually be stressed under normal use?

To value what Prime95 stress testing does for the typical overclocker you have to understand what the phrase accelerated failure test conditions means.

Some nitty gritty is captured in the contents of this link: http://www.weibull.com/hotwire/issue67/hottopics67.htm

Suffice to say what you are using Prime95 to accomplish is to increase the fail-rate of a certain class of instability issues with your CPU. Maybe under your typical usage pattern a particular failure (instability event) will not occur but once every 6-7 days...well instead of waiting 6-7 days to find out you still have a low-level of instability you place the hardware into a stress environment where the fail-rate is increase by a factor of 10x...now you'll incur a failure in 1/10 the time, wasting less of your time to find out how stable your system is.

For people who don't really care about random stability fails with their OC then it isn't such a big deal to you to worry about accelerated fail-rates. However you do have to accept in this case that your definition of stable falls outside the criterion of what your fellow AT posters consider stable, like using non-WHQL compliant drivers when trying to claim a 3dmark05 WR...the standards exist for reason and no one should feel frustrated by this when they choose to operate outside the standards.

For many of us OC'ers around here though we like to know that under accelerated test conditions our rigs do not incur a stability failure in a 24hr period of say prime95 small FFT, or 10hrs of OCCT, etc...this gives us a reasonable expectation that under normal operating conditions we aren't likely to experience a stability related failure for maybe 3-4 weeks if not even less frequently than that.

It's not about making sure our rigs are stable while running under tortuous conditions themselves, but the tortuous conditions accelerate fail-rates so we can uncover them in a reasonable time-frame, make BIOS tweaks as necessary, retest, and repeat until we are satisfied that under our accelerated test conditions we are not incurring a failure within the given test time-span.

This is SOP (standard operating procedure) for characterizing fail-rates and mechanisms in nearly all fields of engineering, it isn't a thought process invented by OC'ing geeks on the AT forums. We are just borrowing/leveraging from best known practices.

It isn't mandatory that you do the same, just as it isn't mandatory that we believe your OC is stable when/if you try and convince it is without having done a 24hr prime95 run, etc.

 

[betasub]

:thumbsup: Nice explanation IDC, and it pretty much covers the how-and-why I like to stress-test my OCed systems.