What is considered stable?

[Mongox]

I've been playing around with simple Overclocking on my GigaByte_GA-MA785GM-US2H with Phenom 550 BE unlocked. I've been using the Tom's "How To: Overclocking Your AMD Processor" as a guide for what I'm changing.

In that guide, it says that they want any settings to go thru at least 30 mins of Prime95 to be considered stable. Does that mean that if it fails after 2 hrs, it's still OK? Or is one error while running Prime95 for 5 hours too many?

What's the standard here? If the system will complete 30 mins of Prime 95, is it considered OK for running days of Excel or games?

 

[normski78]

Hey man!


I usually let prime 95 run overnight min 12hrs. The first time i overclocked i let prime run for 2 days straight. You should get no errors to consider your rig stable. Make sure you run real temp when you are running prime. That will tell you the max temp your cpu reached during your testing. Check your cpu max safe temp from AMD, or just search it in here. Good luck!

 

[jsc]

I used to test for 12 hours. Then a couple of time that I ran P95 overnight, it failed once at 14 hours and once at 18 hours. Now I test for 24 hours.

 

[randomizer]

I do about 2 hours for a 24/7 overclock, 10 mins for a higher-than-24/7 overclock and none at all for suicide runs (because they don't usually pass the first 3 seconds).

 

[Mongox]

I guess the non-24/7 and suicide runs refer to OC things done that are just for fun, rather than trying to create a computer that works?

I'm interested only in getting the most out of my computer - and the computer is designed to accomplish tasks, not to be a performing bear.

So, sounds like I DO want the computer to complete an "endless" series of Prime95 tests. That is, an actual working computer should be able to run Prime95 for weeks at a time.

 

[Mongox]

I usually let prime 95 run overnight min 12hrs. The first time i overclocked i let prime run for 2 days straight. You should get no errors to consider your rig stable. Make sure you run real temp when you are running prime. That will tell you the max temp your cpu reached during your testing. Check your cpu max safe temp from AMD, or just search it in here. Good luck!

Thanks! I've put CPUID's Hardware Monitor in startup while I'm messing around with this, so I don't forget. I started that before installing my new cooler to get a feel for my baselines. My cooler is the Scythe Katana, for me a nice compromise between crazy cooling and not enough. My stock AMD heatsink/fan would reach 73C when running Prime95 (and maybe more, I'd shut down then) - too hot to consider overclocking. With the Katana installed the at-rest temps lowered to as low as 35C but usually ~38C. When running Prime95 it has reached a peak of 64C but generally ~60C.

 

[randomizer]

I follow the logic that if it can do the tasks it's used for without error then it's stable "enough." Many prefer to stress it to breaking point before considering it stable. Stability is in the eye of the beholder.

 

[jsc]

True. If you are building a gaming machine, and it fails a really complex floating point calculation, so what? And for office apps, any modern computer is fast enough without overclocking. And if you are running software requiring mission essential calculations, you shouldn't be OC'ing the thing to the ragged edge of stability anyway.

 

[jitpublisher]

Since by far and away, most overclockers are gamers, I would say that if you can play your favorite game for an hour without a problem, it's good enough.

 

[cjl]

Definitely not. Even if it never crashes, it could corrupt data and eventually cause your OS to stop working. If you want it to truly be stable (and this includes minimizing any chance of data corruption), you want it to be able to pass 12+hrs of Prime95 blend.

Oh, and yes, a fully working computer should be able to run Prime95 indefinitely without errors.

 

[sportsfanboy]

I fully agree with cjl...

It may take a while, but even small instabilities can and most likely will manifest themselves in the form of operating system issues/bugs, programs suddenly crashing, things like that.

 

[Mongox]

Thanks everyone - good discussion!

For my uses - business apps, some graphics design, scenario games - my system out of the box(es) would be plenty. I only upgraded because I could buy a new Phenom-based motherboard, 4GBs, CPU for about $100 more than putting 3GBs of RAM into my old Athlon system. The cpu cooler, 2 case fans and WD 1TB drive are the normal extra things you end up with when "saving" money this way!

But as you explore the possiblities and do the research like I've been doing here, you make some hopefully good choices. Like getting the 550 BE rather than the 545 I had originally planned. Without the comments of Tom's forum users, I'd have saved $10 and regretted it for a couple of years. So now I have a 4 core CPU that's running over 3.5GHz for $100.

In return for all the great help I got here over the past month, I've been slaving away in the Memory forum trying to help those I can. I've been working with PCs since '83 and although my knowledge is a bit rusty, it comes back fast with exposure to you guys!

And yes, I'm not immune to all the new gadgets. My pair of 3-LED Antec case fans make me smile every time I look at the old ugly case they're in - and my legally-blind Dad's "Is your computer supposed to be glowing?" comment was the high point of my week!

So thanks for the help and for helping all the lost souls who wander in here!

 

[Shadow703793]

I run P95 Small FFT for 6-7hrs. Then I run OCCT Linpack for 6hrs. This imo, ensures the highest stability.

 

[michaelmk86]

I used to test for 12 hours. Then a couple of time that I ran P95 overnight, it failed once at 14 hours and once at 18 hours. Now I test for 24 hours.

Prime does not reflect realistic use scenarios. It could easily fail at 25 hours or at 48 hours so what are you going to do? leave your pc forever running prime to make sure? of course not.

So leave prime for 10 min and then enjoy your overclock benefits in realistic scenarios.

 

[jsc]

True. Small fft's loads the processor cores. But I think 24 hours is a good point to stop testing.

All a ten minute test will do is max out the temperatures - admittedly, a worthy goal.

And notice what we are saying. We are telling the OP how long we test for. I scanned back over through the thread and I didn't see anyone saying, "You should test for ___ hours."

 

[Mongox]

Well, what I've learned in playing around with the OC numbers is that if it's gonna fail Prime95, it does it within 2 hrs. My plan is that once I have a nice outcome that seems reasonable compared to my failure sets, I'll let it run for a day.

Of course, now I have to wait until my new much better G.Skill RAM comes in!!!

Most of my OC settings put my Phenom II B50 up at a max of 60C during Prime95 - and otherwise it sits at 38C. Nothing I do normally seems to make it budge over 40C - so clearly the Prime95 is a LOT more taxing than anything I run. I've tried running the cooler/fan at max all the time and under PWM control - I like the control better. Nice to have no noise for normal usage and I like the run-up sound when I fire up the Prime95!

Now, why doesn't Antec make those cool tri-color fans in 92mm for my cooler?

 

[randomizer]

And notice what we are saying. We are telling the OP how long we test for. I scanned back over through the thread and I didn't see anyone saying, "You should test for ___ hours."

Well he never actually asked that, he wanted to know the "standard." I'm sure it's pretty clear that there isn't one

 

[Mongox]

I have to wonder how many of those early IBMs and IBM clones I used to build would pass a test like the current memtests, not to mention Prime95?

The early systems would require the "technician" to install anywhere from 27 to 82 individual RAM chips into tiny sockets, either on the motherboard or add-on cards. Most clones took 36 chips. It took a fair bit of skill to even get the dang things in without bending any of the 16 (?) legs. We did have some tests, but I can't recall if they did more than show that the banks were full and operating.

I remember one test that was great because it would often actually show which chip of the nine in a bank was bad - wasn't totally accurate but you couldn't just remove a row and toss em or send them all back. Usually you didn't know for sure and put the suspicious 9 chips in a tube of their own to be tried again - changing out one chip at a time in a "good" setup. All memory was parity on IBMs so you swapped out the 9th chip to test for bad ones. Failed parity test and you had it.
--------------------------

Back to my OQ, I wasn't happy with the idea of 30 mins of no errors being OK for a computer I might not test again. From this discussion, I got enough replies to understand that 30 mins - or 2 hrs - may be enough to get confidence the configuration is stable. But once you quit fiddling with the settings, run it for 24 hours.