Games are the best way to test overclocked stability

[Carfax83]

Has anyone else found this? I can't recall how many times over the years I've overclocked a CPU, memory and/or GPU, ran the usual stress tests like Prime95, 3D Mark/FurMark for hours with no problems...

Only to have games or Windows crash, BSOD and freeze later on.

As a result of this, I don't even bother running Prime, FurMark or IBT anymore. I just play the most intensive games in my library for a couple hours or so whenever I can.

Sure, it takes a lot longer, but it's a helluva lot more fun than running benchmarks all day. :thumbsup:

Also, you don't subject your components to unrealistically high workloads that generate lots of heat that can accelerate degradation.

 

[Centauri]

Agreed. I've been able to run Prime 95 for days on end as well as AIDA's system test, but the second StarCraft II hits its main menu...

 

[Concillian]

Yep, I still use the bench test programs, but I usually drop the clock by 5% after finding the limit to give a safety margin. 5% has resulted in enough margin that I don't ever get crashes or blue screens. So the bench test programs still have value to me. But you're right, what ultimately matters is the applications you use.

So for my CPU I pick a voltage I'm comfortable with and use 10x or 15x IBT since it doesn't take that long. Bump frequency until I fail, then back down 5% (200-300 MHz) from there and it's usually been a good 24x7x365 OC for me.

Same with Furmark and Unigine Haven with video cards. Find the limits for a given voltage with those programs, then back clocks down by 5% to give it a margin of safety.

When hanging out in forums like this, it's good to understand that some people are posting benchmarks have stable 24x7x365 systems and some do not. Also that everyone has different uses for their CPUs, so some people that are 24x7x365 stable in the applications they use would not necessarily be 24x7x365 stable in the applications you use.

 

[bononos]

Has anyone else found this? I can't recall how many times over the years I've overclocked a CPU, memory and/or GPU, ran the usual stress tests like Prime95, 3D Mark/FurMark for hours with no problems...

Only to have games or Windows crash, BSOD and freeze later on.
......

Because the mix of instructions is different from the stress testing programs vs games, it can't always predict failures at the very margins of stability. I agree with the poster above to reduce voltages/speeds from tested top overclocks for everyday use.

 

[LagunaX]

Old Prime95 and IBT (at max) decent but not sufficient.

Would pass both but crash HD video encoding (Sothink HD movie maker to burn HD video onto regular dvd in 720p or 1080p quality to play on blu-ray players).

Newer Prime95 with AVX ~ HD video encoding to some degree.

So for me, HD video encoding.

 

[grimpr]

Games hit the integrated memory controller hard, ibt and prime95 small fft dont put pressure on the IMC, hence their failure to work as stress testers. I use y-crunchers fft-test to quickly find out where the cpu memory controller borks and move to integer & fpu tests to test the integer/fpu cores and excess heat with x264 hd video encoding, its quick and 100% assured for stability, no need for ridiculous 24h prime95 runs.

 

[Carfax83]

Games hit the integrated memory controller hard, ibt and prime95 small fft dont put pressure on the IMC, hence their failure to work as stress testers. I use y-crunchers fft-test to quickly find out where the cpu memory controller borks and move to integer & fpu tests to test the integer/fpu cores and excess heat with x264 hd video encoding, its quick and 100% assured for stability, no need for ridiculous 24h prime95 runs.

Yeah, but you can also use large FFTs with Prime95 which is memory intensive. Actually, I never even used the small FFTs when I used Prime95.

Still, I would successfully run it for hours, then load a game of some sort and end up crashing or locking up. Or even in Windows, I'd be browsing the internet and wham! Lock up or BSOD..

 

[escrow4]

I run on any system I build:

IBT, 25 passes Very High
HCI Memtest for an hour at least
Latest Prime 95 for 2hrs, either Large FFT or Blend

After all that, a grunty game for 3hrs if its a gaming PC, and basic multitasking and browsing for an office PC. If nothing falls over or BSOD's, its stable (or more than enough stable for anyone splitting hairs).

 

[aigomorla]

Has anyone else found this? I can't recall how many times over the years I've overclocked a CPU, memory and/or GPU, ran the usual stress tests like Prime95, 3D Mark/FurMark for hours with no problems...

Only to have games or Windows crash, BSOD and freeze later on.

False...

Ive had more games crash on me because of driver issues or something else in cross with new hardware.

Its not accurate to test soley on games.
Also some games wont put enough stress on the cpu to load it 100%.

Testing your system for stability also means testing its max thermal output.. which can not be accomplished via gaming.

 

[JimmiG]

I do at least 12 hours of Linpack, Prime95 etc. and then use my computer normally, including games. Sometimes I've had a Prime95-stable overclock crash when just browsing the web or playing a game.

The CPU is made up of thousands of logical units, spread over hundreds of millions transistors. Even though Prime95 might make your CPU run very hot, and it shows 100% CPU usage, doesn't mean it's stressing every single transistor on the chip.

 

[toyota]

I do at least 12 hours of Linpack, Prime95 etc. and then use my computer normally, including games. Sometimes I've had a Prime95-stable overclock crash when just browsing the web or playing a game.

The CPU is made up of thousands of logical units, spread over hundreds of millions transistors. Even though Prime95 might make your CPU run very hot, and it shows 100% CPU usage, doesn't mean it's stressing every single transistor on the chip.

lol I simply ran IBT at maximum and it did not quite make it all the way at 4.5. I lowered it to 4.4 and it made it the through the 45 minute test and its been there for 18 months now. to be clear I am overclocking just by running all 4 cores at a higher turbo with everything else on auto except where I raised the TDP in the BIOS.

 

[Borealis7]

Testing your system for stability also means testing its max thermal output.. which can not be accomplished via gaming.

which brings up a different question: if games do not stress the CPU so much, isn't it sufficient that the CPU is stable during games and stays comfortably below the thermal limit even if it hits max thermal limit in OCCT or crashes in Prime95?

why prepare for Armageddon against the seasonal Flu?

 

[Carfax83]

False...

Ive had more games crash on me because of driver issues or something else in cross with new hardware.

Same here, but aggressive overclocking introduces system instability which can increase the chances of a program or OS crashing or freezing.

Its not accurate to test soley on games.

I would disagree. Using games and other stressful consumer programs like Handbrake are better for stability testing.

Ask yourself, why do we do stability tests in the first place? To make sure our hardware will run our games, operating systems and programs at a faster speed without instability.

So if our main purpose of stability testing is to make sure our hardware can run our software at an accelerated speed, why SHOULDN'T we use our software to test it with since that's what we'll be using on a day to day basis...

Also some games wont put enough stress on the cpu to load it 100%

But from my experience, and the experience of others who have posted in this thread, "maxing out" our hardware isn't a reliable indicator of system stability.

So I would question the worth of running a program like IBT/Prime/FurMark which puts a very unrealistic workload on a computer, and doesn't mirror how you'll be putting your computer to use.

Testing your system for stability also means testing its max thermal output.. which can not be accomplished via gaming.

But why? Unless you're an electrical engineer that designs semiconductors, I can't imagine why any consumer would be interested in knowing how much heat your processor can output or tolerate..

 

[toyota]

if a cpu can make it through 100% full load on cpu and memory in IBT then how the heck is a game going to stress it in a comparable way. cpus are way different than gpus as they almost never get pushed hard in actual games unless all you have is a dual core cpu.

 

[Racan]

So if our main purpose of stability testing is to make sure our hardware can run our software at an accelerated speed, why SHOULDN'T we use our software to test it with since that's what we'll be using on a day to day basis...

And how would you know that the overclock is truly stable? Your CPU could still be unstable and produce corrupt data even if your favorite software doesn't outright crash.

 

[BallaTheFeared]

IMO...

You should test with what you're doing, or simply put, just use it.

Use the BBCodes to make adjustments, check the event viewer for minor errors.

I've done this since Sandy Bridge, and outside of finding appropriate clocks for my i3-540 when I first got it I haven't had any crashes with it. Now my finicky 7950's, that's a perfect example since no amount of single program stability testing is ever going to matter due to their nature.

 

[John Connor]

I wonder if Super PI Mod would be useful for benchmarking? http://www.techpowerup.com/downloads/366/super-pi-mod-v1-5/

 

[JimmiG]

which brings up a different question: if games do not stress the CPU so much, isn't it sufficient that the CPU is stable during games and stays comfortably below the thermal limit even if it hits max thermal limit in OCCT or crashes in Prime95?

why prepare for Armageddon against the seasonal Flu?

Peace of mind. If I get a BSOD 1 month later in a new game, I don't know whether there's an issue with the drivers/game, or a result of my overclock. With proper stress tests, I can feel reasonably confident that the issue is not related to the overclock.

 

[Carfax83]

if a cpu can make it through 100% full load on cpu and memory in IBT then how the heck is a game going to stress it in a comparable way. cpus are way different than gpus as they almost never get pushed hard in actual games unless all you have is a dual core cpu.

A CPU can make it through 100% full load on CPU and memory in IBT, and when you think all is good, crash or freeze the system when you're doing something as innocuous as surfing the internet or loading/playing your favorite game.

There is no silver bullet for this, but by playing games at least you will know whether your overclock is GAME stable, which is what truly matters.

And how would you know that the overclock is truly stable? Your CPU could still be unstable and produce corrupt data even if your favorite software doesn't outright crash.

There are always signs. After I built my new 3930K rig, I was experimenting with voltages to see how low I could get it and still be stable at my desired clock speed, which was 4.4ghz.

I started at 1.31v, and kept working my way down. I tried 1.285v and was completely stable through repeated play throughs of the Welcome to the Jungle level in Crysis 3; a level that is very stressful for CPUs due to all of the physics and animations for the thousands of blades of grass.. Also played Far Cry 3 for a bit by driving around the island.

So I lowered it again to 1.28v this time, and when I did my Crysis 3 play through, I noticed that the gameplay wasn't as smooth as it was before. It was laggier, but the game DID NOT crash or freeze.

Still, I took this as a sign that 1.28v was too low to sustain the 4.4ghz clock speed and the CPU was probably down clocking momentarily during gameplay. So in the end, I decided that 1.285v was my final voltage, and I haven't had any problems since then.

So you see, there are signs my friend. You just need to keep your eyes peeled.....like this

 

[BrightCandle]

Until we get a program that can actually validate all the instructions are working we can't really stability test properly. What we do today is horribly flawed as a testing approach for the purpose of determine reliable computation. Just testing float or integer or mixed workloads isn't going to get us the sort of test we really need. The program that produces the most heat certainly doesn't make it the best validation tool.

 

[toyota]

A CPU can make it through 100% full load on CPU and memory in IBT, and when you think all is good, crash or freeze the system when you're doing something as innocuous as surfing the internet or loading/playing your favorite game.

There is no silver bullet for this, but by playing games at least you will know whether your overclock is GAME stable, which is what truly matters.



There are always signs. After I built my new 3930K rig, I was experimenting with voltages to see how low I could get it and still be stable at my desired clock speed, which was 4.4ghz.

I started at 1.31v, and kept working my way down. I tried 1.285v and was completely stable through repeated play throughs of the Welcome to the Jungle level in Crysis 3; a level that is very stressful for CPUs due to all of the physics and animations for the thousands of blades of grass.. Also played Far Cry 3 for a bit by driving around the island.

So I lowered it again to 1.28v this time, and when I did my Crysis 3 play through, I noticed that the gameplay wasn't as smooth as it was before. It was laggier, but the game DID NOT crash or freeze.

Still, I took this as a sign that 1.28v was too low to sustain the 4.4ghz clock speed and the CPU was probably down clocking momentarily during gameplay. So in the end, I decided that 1.285v was my final voltage, and I haven't had any problems since then.

So you see, there are signs my friend. You just need to keep your eyes peeled.....like this

yes I know that. I am talking about just checking to make sure its stable under load. my system would randomly crash doing nothing with an oc done in the regular way but was fine in games and IBT. the only way to oc with this mobo and be stable is to oc with raising turbo.

 

[Kallogan]

Totally agree with the OP. I had an i3 530 overclocked to near 4 ghz that passed every stress tests but as soon as i was launching a game. Bam, BSOD ! I had to add voltage or lower clocks a bit to make it completely stable for gaming.

 

[Idontcare]

The reason we "test" our overclock is to gain some idea of its "time to failure" (i.e. how long before it becomes unstable).

Is it stable for 5 minutes? 5 hours? 5 days? 5 weeks?

The reason we add the word "stress" in front of the word "test" in creating a more rigorous "stability-test" is because we don't want to wait 5 days or 5 weeks to develop enough data for us to comfortably determine the typical time-to-failure for our overclock.

Playing games is not a "stress test", it is simply a "test". There are no acceleration effects entering into the picture in the engineering sense.

The OP's issue is not with stress testing per se, rather their issue is with the fact that the suite of available stress testers on the consumer scene are very limited in terms of the scope of what they test.

Our modern CPUs have nearly 2000 instructions in the ISA, any one of which can become unstable while overclocking.

A stress test program like Prime95 or IBT might stress-test say 200 of those 2000 instructions, at best, easily leaving 90% of the ISA completely untested (let alone not stress-tested) for stability.

Games will use different instructions, as well as test the stability in other components in the system that are not part of the processing core itself.

So it should not be of any surprise that a system which passes Prime95 can still be unstable with a game, or vice versa for that matter.

The bigger concern people should have when overclocking and not having a way to detect errors and instability in the way that a bonafide stress-tester will detect them is that they are leaving their system completely wide open for silent data corruption to occur.

Photo data corruption; in this case, a result of a failed data recovery from a hard disk drive

Your OS can become unstable, you start getting more video-driver related crashes to the desktop while playing games or doing other stuff as simple as browsing the web. Its a real pain because you don't really have a stable OC but now you have to redo your entire install.

Stability is a multi-faceted issue because it involves the interaction of multiple components in the compute topology. Relying on just one program, or one particular suite of programs, is a recipe for guaranteed data corruption and eventual disaster.

It doesn't make any one approach or program superior to the others, you absolutely must do "all of the above".

Intel and AMD do this too, and there is a reason why they do it despite the associated costs of validating each and every CPU's instruction in the ISA.

If you think about it, it is a silly sort of ignorance-fueled-arrogance on our part for us to think we can simply get away with declaring our overclocked CPU "stable" simply by running a freeware program that at best is checking and validating maybe 10% of the ISA (be it from testing with Prime95 or a couple of games).

 

[Carfax83]

If you think about it, it is a silly sort of ignorance-fueled-arrogance on our part for us to think we can simply get away with declaring our overclocked CPU "stable" simply by running a freeware program that at best is checking and validating maybe 10% of the ISA (be it from testing with Prime95 or a couple of games).

That's a good explanation Idontcare. :thumbsup: People put too much stock in these stress test programs..

This blogger wrote an interesting piece about stress testing, and claims that the best way to stress test your hardware is to run both CPU AND GPU stress tests at the same time.

So instead of running Prime95 and then FurMark afterwards, you run them both at the same time. Perhaps this has merit, but I personally will stick with gaming. Running endless benchmarks and tests is boring as hell to me :thumbsdown: