[SOLVED] Overclocking my i7 9700k. Is 4 hours of prime95 enough?

Sep 27, 2019
24
0
20
0
prime95 is still running. when should I finish the test?
-i7 9700k /msi mpg z390 gaming pro carbon.
-prime95 v29.8 build 6 (non-avx).
-1.3v 4.9 ghz all core
 

Darkbreeze

Titan
Moderator
No. Not the Realbench benchmark. The Realbench stress test. There are options to do separate tests in Realbench and they do entirely different things. Well, same things, to a degree, but in different ways.

I assure you, there are a bunch of EXTREMELY experienced engineers and world class overclockers out there who whole heartedly disagree with your assessment that Realbench is nothing more than a Cinebench equivalent and that Prime95 is a better metric. Prime95 is superb for testing memory configurations, and to a smaller degree the CPU stability, and is THE best test if you run the Small FFT option, for thermal compliance testing, but it is not a better stability metric than Realbench. "Better" or any other term that suggests one thing is better than another for testing stability is highly misleading because the fact is that depending on WHAT it is you do on a given machine, ALL tests might be not only useful, but recommended, or specific tests might be better suited for ensuring that the kinds of instructions you'll be primarily working with are the focus.

There is no overkill when it comes to stability testing, not really. People who say you can just run IBT for five minutes and OCCT for ten minutes, or any similar nonsense, are fracking idiots and are the same people who, three or four months later, are having to reinstall Windows and download all their game files over again because they corrrupted all those files because they don't "really worry about that because I just game". Yep, you just game alright.
 

Phaaze88

Admirable
Herald
Prime95 would arguably be the best stress test.
If it passes that, it won't really have a problem running anything else.

Cinebench and Realbench are benchmarks, not really stress tests.
Realbench has both benchmark and stress test modes.
You can have an overclock pass Prime95 at 24hs, but still crash during gaming loads or other.
Prime95 makes for a good thermal test, but the load is constant - it does not make for a very good voltage stability test like Realbench stress test, which has highs and lows.
I'd do 8hrs of Realbench first, and if it passes that, then do P95 - I stopped using P95 though due to it being purely synthetic, and knowing I'd never put that kind of load on my cpu normally. It still has its uses.
 

Darkbreeze

Titan
Moderator
Prime95 would arguably be the best stress test.
If it passes that, it won't really have a problem running anything else.

Cinebench and Realbench are benchmarks, not really stress tests.
Prime95 is a good metric for testing thermal compliance and for testing the stability of memory configurations. While it IS a good metric for testing stability, in general, if you choose the Blend mode, or any mode really, and run it long enough, Realbench is a much better option. Running both, Prime95 blend mode for 8 hours (Or a custom configuration found in my memory guide) AND Realbench for 8 hours, in addition to four passes of Memtest86, would be the minimum recommendation I'd offer for anybody TRULY wanting to make sure that the changes they've made are not going to come back and bite them later with corruption from micro-errors.
 

PC Tailor

Dignified
Herald
Prime95 is a good metric for testing thermal compliance and for testing the stability of memory configurations. While it IS a good metric for testing stability, in general, if you choose the Blend mode, or any mode really, and run it long enough, Realbench is a much better option. Running both, Prime95 blend mode for 8 hours (Or a custom configuration found in my memory guide) AND Realbench for 8 hours, in addition to four passes of Memtest86, would be the minimum recommendation I'd offer for anybody TRULY wanting to make sure that the changes they've made are not going to come back and bite them later with corruption from micro-errors.
Thank you, I didn't know the depth of Realbench.
Another learning point for the day :)
 

Darkbreeze

Titan
Moderator
Small FFT is not for testing stability. Small FFT is for testing thermal compliance. Totally different things. For stability using Prime95 you want to run either plain Jane Blend mode, or this, which will really hammer the memory configuration but also the CPU, and pick up a lot of instability that is not found in Memtest86.

Final testing with Prime95

It is highly advisable that you do a final test using Prime95 version 26.6 or the latest version WITH AVX and AVX2 disabled, and run a custom configured Blend test. You can also use the Blend mode option as is, but after a fair amount of personal testing, asking questions from some long time members with engineering level degrees that have forgotten more about memory architectures than you or I will ever know, and gathering opinions from a wide array of memory enthusiasts around the web, I'm pretty confident that the custom option is a lot more likely to find errors with the memory configuration, and faster, if there are any to be found.

Please note as this is rather important, if you prefer, or have problems running version 26.6 because you have a newer platform that doesn't want to play nice with version 26.6, you can use the latest version of Prime95 with the Custom test selected but you will need to make the following change.

In the bottom of the Torture test selection popup menu there will be some options for disabling AVX. I recommend that you do so, not because we are doing thermal testing and require a steady state workload (Which AVX wouldn't affect anyhow, as Computronix explained to me), but because the last thing you need during memory testing is having to worry about CPU temperatures, and you will, with AVX enabled.

So, uncheck the option for AVX2. That will un-gray the option for AVX, and uncheck that box as well.

Now open Prime95.

Click on "Custom". Input a value of 512k in the minimum FFT size field. Leave the maximum FFT size field at 4096k. In the "Memory to use" field you should take a look at your current memory allocation in either HWinfo or system resource monitor. Whatever "free" memory is available, input approximately 75% of that amount. So if you currently have 16GB of installed memory, and approximately 3GB are in use or reserved leaving somewhere in the neighborhood of 13GB free, then enter something close to 75% of that amount.

So if you have 13GB free, or something reasonably close to that, then 75% of THAT would be 9.75GB, which, when multiplies times 1024 will roughly equal about 9984MB. You can average things out by simply selecting the closest multiple of 1024 to that amount just to keep it simple, so we'll say 10 x 1024= 10240mb and enter that amount in the field for "Memory to use (MB)". We are still well within the 13GB of unused memory BUT we have left enough memory unused so that if Windows decides to load some other process or background program, or an already loaded one suddenly needs more, we won't run into a situation where the system errors out due to lack of memory because we've dedicated it all to testing.

I've experienced false errors and system freezes during this test from over allocating memory, so stick to the method above and you should be ok.


Moving right along, do not change the time to run each FFT size.Leave that set to 15 minutes.

Click run and run the Custom test for 8 hours. If it passed Memtest86 and it passes 8 hours of the Custom test, the memory is 100% stable, or as close to it as you are ever likely to get but a lot of experts in the area of memory configuration suggest that running the extended Windows memory diagnostic test is also a pretty good idea too.

If you get errors, (and you will want to run HWinfo alongside Prime95 so you can periodically monitor each thread as Prime will not stop running just because one worker drops out, so you need to watch HWinfo to see if there are any threads not showing 100% usage which means one of the workers errored and was dropped) then you need to either change the timings, change the DRAM voltage or change the DRAM termination voltage, which should be approximately half of the full DRAM voltage.

There are also other bios settings that can affect the memory configuration AND stability, such as the SOC, VCCIO and system agent voltages, so if you have problems with stability at higher clock speeds you might want to look at increasing those slightly. Usually, for Intel at least, something in the neighborhood of 1.1v on both those is pretty safe. There are a substantial number of guides out there covering those two settings, but most of them are found within CPU overclocking guides so look there in guides relevant to your platform.

As a further measure of assurance that your WHOLE configuration is stable, you can download and run Realbench for 8 hours. If the system freezes or fails when running Realbench with your full memory amount set, try running it again but select only half your amount of installed memory.
 
What would be the impact on you if your pc should shut down unexpectedly?

No doubt that would upset you, but normally, no harm will come.
You just restart.
Just like a household power failure.

In real usage, you will not be pushing loads nearly as heavy as the stress tests.
Moreover, your actual usage of instructions may well be very different from what your stress test is using.

Consider a much shorter testing period. Say an hour.
The ultimate test is if YOUR workload is stable.

Heresy??
 

Darkbreeze

Titan
Moderator
Yes, heresy.

Rest assured, there will shortly be a rather angry gathering of firebrand and pitchfork wielding enthusiast villagers arriving shortly at your doorstep to drag you to the town square for a draw and quartering session. Or perhaps just a public stoning.

24-48 hours for non-critical scientific or folding type scenarios is probably not essential and is overkill, but testing a configuration for 8 hours is not excessive. If a person feels it is, then they should simply not overclock and leave things at the stock configuration. If you're going to overclock, especially if you're reducing voltage, then you should put in the work. An hours worth of testing on one utility or application may not even be representative of the types of workloads you may encounter whether you intend to or not. Windows and many applications are likely to use instruction sets that you haven't the foggiest idea about. It would be a lot better to KNOW that you've done your due diligence to a reasonable degree than to assume that just because you "just game" it's not a factor.

It is. There are a lot more things going on than "just gaming" on any system even if it's only the drivers, applications and engines that are running and being supported by the operating system. I'm sure there aren't many people that would be happy about having a system that never crashes while playing games but after a few weeks of accumulated microerrors begins throwing mystery errors and unexplained behaviors at you because your saved game files are entirely corrupted due to a minor instability that was incrementally putting zeros where there should have been ones and a couple of ones where there should have been a zero.

Happens all the time. Most people just don't realize that THAT is what is going on, and clear everything out and reinstall periodically, assuming it's "just a bug or flaw in Windows" or something equally astute.
 
Sep 27, 2019
24
0
20
0
Small FFT is not for testing stability. Small FFT is for testing thermal compliance. Totally different things. For stability using Prime95 you want to run either plain Jane Blend mode, or this, which will really hammer the memory configuration but also the CPU, and pick up a lot of instability that is not found in Memtest86.

Final testing with Prime95

It is highly advisable that you do a final test using Prime95 version 26.6 or the latest version WITH AVX and AVX2 disabled, and run a custom configured Blend test. You can also use the Blend mode option as is, but after a fair amount of personal testing, asking questions from some long time members with engineering level degrees that have forgotten more about memory architectures than you or I will ever know, and gathering opinions from a wide array of memory enthusiasts around the web, I'm pretty confident that the custom option is a lot more likely to find errors with the memory configuration, and faster, if there are any to be found.

Please note as this is rather important, if you prefer, or have problems running version 26.6 because you have a newer platform that doesn't want to play nice with version 26.6, you can use the latest version of Prime95 with the Custom test selected but you will need to make the following change.

In the bottom of the Torture test selection popup menu there will be some options for disabling AVX. I recommend that you do so, not because we are doing thermal testing and require a steady state workload (Which AVX wouldn't affect anyhow, as Computronix explained to me), but because the last thing you need during memory testing is having to worry about CPU temperatures, and you will, with AVX enabled.

So, uncheck the option for AVX2. That will un-gray the option for AVX, and uncheck that box as well.

Now open Prime95.

Click on "Custom". Input a value of 512k in the minimum FFT size field. Leave the maximum FFT size field at 4096k. In the "Memory to use" field you should take a look at your current memory allocation in either HWinfo or system resource monitor. Whatever "free" memory is available, input approximately 75% of that amount. So if you currently have 16GB of installed memory, and approximately 3GB are in use or reserved leaving somewhere in the neighborhood of 13GB free, then enter something close to 75% of that amount.

So if you have 13GB free, or something reasonably close to that, then 75% of THAT would be 9.75GB, which, when multiplies times 1024 will roughly equal about 9984MB. You can average things out by simply selecting the closest multiple of 1024 to that amount just to keep it simple, so we'll say 10 x 1024= 10240mb and enter that amount in the field for "Memory to use (MB)". We are still well within the 13GB of unused memory BUT we have left enough memory unused so that if Windows decides to load some other process or background program, or an already loaded one suddenly needs more, we won't run into a situation where the system errors out due to lack of memory because we've dedicated it all to testing.

I've experienced false errors and system freezes during this test from over allocating memory, so stick to the method above and you should be ok.


Moving right along, do not change the time to run each FFT size.Leave that set to 15 minutes.

Click run and run the Custom test for 8 hours. If it passed Memtest86 and it passes 8 hours of the Custom test, the memory is 100% stable, or as close to it as you are ever likely to get but a lot of experts in the area of memory configuration suggest that running the extended Windows memory diagnostic test is also a pretty good idea too.

If you get errors, (and you will want to run HWinfo alongside Prime95 so you can periodically monitor each thread as Prime will not stop running just because one worker drops out, so you need to watch HWinfo to see if there are any threads not showing 100% usage which means one of the workers errored and was dropped) then you need to either change the timings, change the DRAM voltage or change the DRAM termination voltage, which should be approximately half of the full DRAM voltage.

There are also other bios settings that can affect the memory configuration AND stability, such as the SOC, VCCIO and system agent voltages, so if you have problems with stability at higher clock speeds you might want to look at increasing those slightly. Usually, for Intel at least, something in the neighborhood of 1.1v on both those is pretty safe. There are a substantial number of guides out there covering those two settings, but most of them are found within CPU overclocking guides so look there in guides relevant to your platform.

As a further measure of assurance that your WHOLE configuration is stable, you can download and run Realbench for 8 hours. If the system freezes or fails when running Realbench with your full memory amount set, try running it again but select only half your amount of installed memory.
asus realbench system test passed.(8hr)
ı need to go now prime95 custom ?
i have 12 gb free ram(totally16 gb) can u tell me settings ?
 

ASK THE COMMUNITY

TRENDING THREADS