QFT.
This is the reason RMA exists. If a CPU or other part can't do math for a few hours on end, then there is definatly something wrong with it. Always do a stock test to make sure the part you bought is working correctly.
Originally posted by: VirtualLarry
Yes and Yes.Originally posted by: Lothar
Do you test your stock speed system to make sure it's Prime95, Orthos, OCCT, Intel Burn Test stable for 24 hours?
What if you find an error running at stock settings? Do you RMA cpu/motherboard to Newegg/Intel saying it's defective because it wasn't Prime95/OCCT stable for 24 hours running at stock which could lead to silent data corruption?
Originally posted by: Jumpem
I have thousands of family photos, home video, and some financial spreadsheets that are irreplacable. Could these be corrupted by overclocking?
If so, I don't think an extra 5 fps is worth it.
Originally posted by: JimmiG
Ideally, you should have two PC's - one that you use for gaming, overclocking, trying out new beta drivers etc. Then a cool, quiet system running proven, reliable components at their stock speed, with WHQL drivers and only trusted software that you need for productivity, photos etc.
Also, regardless of whether you overclock or not, I do hope you have backups of this irreplaceable data. Otherwise you're guaranteed to loose it all at some point anyway. It's not a question of if the hard drive is going to fail, it's a question of when.
As for corruption, I find that's more likely when you're pushing the FSB or memory too hard, rather than the CPU. An overclocked southbrdige can become flaky, especially when many drives are connected. Add to that mix overclocked RAM sticks that can't hold the data without it getting changed, and you'll get lots of strange crashes and application failures even when CPU usage is pretty low. Of course you can just blame them on Vista like everyone else does, but it's probably a good idea to look into it even if your CPU passes that burn-in test
Originally posted by: VirtualLarry
Yes and Yes.Originally posted by: Lothar
Do you test your stock speed system to make sure it's Prime95, Orthos, OCCT, Intel Burn Test stable for 24 hours?
What if you find an error running at stock settings? Do you RMA cpu/motherboard to Newegg/Intel saying it's defective because it wasn't Prime95/OCCT stable for 24 hours running at stock which could lead to silent data corruption?
Originally posted by: Idontcare
Originally posted by: Lothar
Originally posted by: ecvs85
The thing is that by passing the stress test, you can ensure that the data calculated by the CPU is correct, which is critical in certain situations such as encoding, distributed computing, probably excel etc. One of the reason for stress testing is that by passing the test, the cpu will be stable and it would not be a factor in crashing the computer anymore.
It all depends on the end user, if the computer didn't crash and doesn't run data critical applications then go ahead. But if the machine isn't stable Prime/Orthos wise and still using it to run distributed computing software then you'll be doing more harm than good, bad data hurts those projects.
What defines stability for those programs? The standard "no errors in 24 hours" rule being mentioned all over the internet?
Do you test your stock speed system to make sure it's Prime95, Orthos, OCCT, Intel Burn Test stable for 24 hours?
What if you find an error running at stock settings? Do you RMA cpu/motherboard to Newegg/Intel saying it's defective because it wasn't Prime95/OCCT stable for 24 hours running at stock which could lead to silent data corruption?
Do you increase voltage above the default settings which Intel/AMD engineers already feel is adequate?
Before Intel and AMD release a processor into the retail channels, do they test Prime95 and OCCT for 24 hours ensuring that their processor don't fail a test?
What if the CPU I'm about to order from Newegg/Micro Center passes Intel Burn test, but fails OCCT and Prime 95 running at stock? What should I do?
Also, why stop at 24 hours? Why not 168 hours?
I find the "24 hour" rule being mentioned all over the internet to be completely unnecessary. The most I've ever ran Prime95 for is 18 hours. I'd say about 12 hours is more than enough for my needs.
As is the case with all testing, the tester must first decide what question it is that they want to answer by running the stress test.
If you are running stress tests for arbitrary reasons (some dude on Internet said to run it this long) then you will find the answer to not really address any question you personally might have had.
There are valid reasons for running 24hr tests, that data answers questions that some people want answered. If you don't know the question that is being answered it would be ignorant of you to assume the question itself doesn't exist.
So the question for you is what question do you think you are answering by running an 8hr test, versus a 24hr test, versus a 168hr test? If you know the question you are trying to answer by running your stress test then it is self-evident.
If you aren't really sure why you are running stress tests then the data you are generating by doing so will be rather meaningless and as such you are more likely to conclude the exercise of generating the data is equally meaningless. Your deciding such doesn't actually make it so, just means you are operating within a certain regime of ignorance as to what people intend to accomplish by running longer and longer stress tests.
Those folks are seeking to generate an answer (the data) to a question they have. You might simply not have the desire to answer that question, or you may be unaware that the question exists or that it might be relevant to you.
Originally posted by: Zensal
Originally posted by: VirtualLarry
Yes and Yes.Originally posted by: Lothar
Do you test your stock speed system to make sure it's Prime95, Orthos, OCCT, Intel Burn Test stable for 24 hours?
What if you find an error running at stock settings? Do you RMA cpu/motherboard to Newegg/Intel saying it's defective because it wasn't Prime95/OCCT stable for 24 hours running at stock which could lead to silent data corruption?
QFT.
This is the reason RMA exists. If a CPU or other part can't do math for a few hours on end, then there is definatly something wrong with it. Always do a stock test to make sure the part you bought is working correctly.
Which is complete BS. Unfortunately, both ATI and NV cards have sub-standard engineering margin, especially as far as temps and power go, so they attempt to prevent what is otherwise perfectly-normal software from running on their GPUs.Originally posted by: Lothar
Both ATI and Nvidia have labeled OCCT as dangerous for a reason and throttle it when their GPUs detect it.
The point of a lengthy test is, the longer you run it, the longer your confidence interval is. One other reason for a 24hr test is that it subjects the test system to both thermal and AC cycles during the day and night, subjecting it to all possible variances.Originally posted by: Lothar
Also, keep in mind I stopped prematurely at 18 hours. It doesn't mean my computer wouldn't have been 24 hours Prime95 stable if I didn't stop it prematurely.
Some people say you discover 99% of the errors within the first 12-16 hours of running Prime95. I haven't seen any basis for that evidence either.
Yes, I personally test for 24 hours, but as I noted in my thread (which I could probably dredge up), it wasn't good enough. My rig was rebooting, mysteriously, after several weeks at a time. I have since reduced the overclock on that machine from 3.2/400x8 to 2.8/350x8, and now it runs continuously without issue.Originally posted by: Lothar
Based on your posts I've seen in Storage and this section as well, you're very intelligent so maybe you can provide me with your reasons.
Do you personally test for 24 hours?
What questions are you trying to answer that can't be answered by running a 12, 16, 18, 24, 48, or a 168 hour test?
Originally posted by: VirtualLarry
Which is complete BS. Unfortunately, both ATI and NV cards have sub-standard engineering margin, especially as far as temps and power go, so they attempt to prevent what is otherwise perfectly-normal software from running on their GPUs.Originally posted by: Lothar
Both ATI and Nvidia have labeled OCCT as dangerous for a reason and throttle it when their GPUs detect it.
Imagine if Intel throttled their CPUs down, if someone ran a NOP-loop or something. (Ok, ok, for the technical folks, I know that wouldn't comprise a stressful load, but just imagine for a minute that it did). People would be getting their pitchforks ready.
Originally posted by: munky
I disagree. The Furmark gpu stress test creates an unrealistic load on the gpu, with the sole intent of stressing it as much as possible; the kind of load that would never be encountered in actual games. If my video card runs every game out there without overheating, then I don't care if it can't run some synthetic stress test. CPU's are more general-purpose than gpu's, so they need to handle just about any code a developer can write. But even so, if there was an overheating issue in some stress test that would never be replicated in actual apps, I wouldn't care much about it either.
Originally posted by: VirtualLarry
Originally posted by: munky
I disagree. The Furmark gpu stress test creates an unrealistic load on the gpu, with the sole intent of stressing it as much as possible; the kind of load that would never be encountered in actual games. If my video card runs every game out there without overheating, then I don't care if it can't run some synthetic stress test. CPU's are more general-purpose than gpu's, so they need to handle just about any code a developer can write. But even so, if there was an overheating issue in some stress test that would never be replicated in actual apps, I wouldn't care much about it either.
I disagree, more and more, GPUs are being used for GPGPU usage, which means that they are used for processing much like a heavily-parallel CPU. That means, that they should be stable, no matter WHAT load the software places on them. You're wrong about Furmark and OCCT being "unrealistic". When processing GPGPU data, the GPU is placed under similar loads. So the bottom line is that AMD and NV don't design for the worst-case, loads, only average, and that is going to hurt them once GPGPU usage increases in the market, as those users are going to want a STABLE platform for processing. The sad fact is that not too many of those cards on the market today can handle that type of processing load 24/7.
Originally posted by: VirtualLarry
Originally posted by: munky
I disagree. The Furmark gpu stress test creates an unrealistic load on the gpu, with the sole intent of stressing it as much as possible; the kind of load that would never be encountered in actual games. If my video card runs every game out there without overheating, then I don't care if it can't run some synthetic stress test. CPU's are more general-purpose than gpu's, so they need to handle just about any code a developer can write. But even so, if there was an overheating issue in some stress test that would never be replicated in actual apps, I wouldn't care much about it either.
I disagree, more and more, GPUs are being used for GPGPU usage, which means that they are used for processing much like a heavily-parallel CPU. That means, that they should be stable, no matter WHAT load the software places on them. You're wrong about Furmark and OCCT being "unrealistic". When processing GPGPU data, the GPU is placed under similar loads. So the bottom line is that AMD and NV don't design for the worst-case, loads, only average, and that is going to hurt them once GPGPU usage increases in the market, as those users are going to want a STABLE platform for processing. The sad fact is that not too many of those cards on the market today can handle that type of processing load 24/7.
That's the key, varying degrees of load. Games don't put a huge 100% load on a GPU. Scenes have varying complexity. GPGPU apps are different. Just like Prime95 is a different app than Microsoft Word, and puts a lot more load on your CPU. (Don't forget, Prime95 is a working app too, not just a stress-test program.)Originally posted by: munky
And how are GPGPU apps more stressful than a high tech game like Crysis? Virtually all modern games involve complex shaders, with varying degrees of load on the gpu.
Did you know that a few months ago, after "optimization" of the F@H core code that runs on GPUs, people were complaining because their GPUs were getting too hot and erroring, due to the load, so the F@H people put in voluntary throttling, to allow the GPUs to run cooler. Face it, video cards as built today are not designed to run with 100% 24/7 load.Originally posted by: munky
People have been running F@H on gpu's for some time now, and I haven't heard anything about vendors throttling the gpu or the OCP kicking in when the app is running, as opposed to Furmark.
Originally posted by: VirtualLarry
That's the key, varying degrees of load. Games don't put a huge 100% load on a GPU. Scenes have varying complexity. GPGPU apps are different. Just like Prime95 is a different app than Microsoft Word, and puts a lot more load on your CPU. (Don't forget, Prime95 is a working app too, not just a stress-test program.)Originally posted by: munky
And how are GPGPU apps more stressful than a high tech game like Crysis? Virtually all modern games involve complex shaders, with varying degrees of load on the gpu.
Did you know that a few months ago, after "optimization" of the F@H core code that runs on GPUs, people were complaining because their GPUs were getting too hot and erroring, due to the load, so the F@H people put in voluntary throttling, to allow the GPUs to run cooler. Face it, video cards as built today are not designed to run with 100% 24/7 load.Originally posted by: munky
People have been running F@H on gpu's for some time now, and I haven't heard anything about vendors throttling the gpu or the OCP kicking in when the app is running, as opposed to Furmark.
Btw, F@H has burned out at least one of my video cards, an Asus 9600GSO that was factory overclocked.
Originally posted by: munky
Originally posted by: VirtualLarry
That's the key, varying degrees of load. Games don't put a huge 100% load on a GPU. Scenes have varying complexity. GPGPU apps are different. Just like Prime95 is a different app than Microsoft Word, and puts a lot more load on your CPU. (Don't forget, Prime95 is a working app too, not just a stress-test program.)Originally posted by: munky
And how are GPGPU apps more stressful than a high tech game like Crysis? Virtually all modern games involve complex shaders, with varying degrees of load on the gpu.
Did you know that a few months ago, after "optimization" of the F@H core code that runs on GPUs, people were complaining because their GPUs were getting too hot and erroring, due to the load, so the F@H people put in voluntary throttling, to allow the GPUs to run cooler. Face it, video cards as built today are not designed to run with 100% 24/7 load.Originally posted by: munky
People have been running F@H on gpu's for some time now, and I haven't heard anything about vendors throttling the gpu or the OCP kicking in when the app is running, as opposed to Furmark.
Btw, F@H has burned out at least one of my video cards, an Asus 9600GSO that was factory overclocked.
And if they were designed for 100% loads 24/7, they would most likely cost more. As a gamer, I sure as hell don't want to pay extra for a card because some other people are gonna be using it for F@H 24/7. Let them get their own gpgpu-approved version.
+1 Your CPU is not even guaranteed 100% problem free (could very well be 99.999999999%) even with Prime testing, let alone without.Originally posted by: Rubycon
Would you drive your car with the wheel lugnuts hand tight?
You're playing with fire here. Anything you do with your computer cannot be trusted. From corrupted pictures to media files that won't play back - the possibilities of this biting you HARD are limitless.
Originally posted by: munky
I disagree. The Furmark gpu stress test creates an unrealistic load on the gpu, with the sole intent of stressing it as much as possible; the kind of load that would never be encountered in actual games. If my video card runs every game out there without overheating, then I don't care if it can't run some synthetic stress test. CPU's are more general-purpose than gpu's, so they need to handle just about any code a developer can write. But even so, if there was an overheating issue in some stress test that would never be replicated in actual apps, I wouldn't care much about it either.