Length of Prime95 run

[Benjamin J 751]

I've been reading about Prime95, and the length of time it's recommended to run this. Some mixed responses, though overall it seems people like to run it for 24 hours. I was just wondering what the opinions here are regarding this? So far I've run it for about an hour after each change in the BIOS. Should I do an additional, longer test, to make sure?

Any help is appreciated!

[zmak 331]

I have found the ultimate bench is your flight sim. I have run prime, real bench,OCCT and passed with flying colors only to have weird crashes in the Sim after 1 hour

When overclocking in increments a 24 hour bench will take forever. An hour between BSOD is more than enough and then a final 24 hour if you want to confirm stability but then go to your sim and load setting on medium to high, Pmdg if you have it and launch a 12 hour flight while your sleeping...check in the morning :-)

[Benjamin J 751]

Thanks Zoran, I expected as much. I have my i5 at 4.6GHz now and after an hour of Prime95, it seems stable... Tried going to 4.7GHz but I had to crank up the voltage and I didn't like where my temps were going at that point, as they were going into the low 80s. 4.6GHz kept them around 79C. I guess I'm hitting the limit of my cooling solution... Noctua DH-15S. But, I don't think that an extra 100 or 200MHz will have a gigantic impact on FPS, so...

Following your suggestion, I might do a few hours more of Prime95 this evening, but ultimately I'll install P3D and some addons, and take it for a spin. As you say, it's likely to be the better benchmark :)

[martin-w 3,234]

9 hours ago, Benjamin J said:

I've been reading about Prime95, and the length of time it's recommended to run this. Some mixed responses, though overall it seems people like to run it for 24 hours. I was just wondering what the opinions here are regarding this? So far I've run it for about an hour after each change in the BIOS. Should I do an additional, longer test, to make sure?

Any help is appreciated!

 

 

It's nonsense in my opinion. I've been overclocking and building PC's for quite  a few years and I have never done this. Seems a daft idea to me to stress your CPU to a high level for such an extended period. Why increase degradation unnecessarily.

The other point is that you can run Prime 95 for as many hours as you like and have it pass with flying colours... but then it fails as soon as you run Handbrake, or another one of your favourite applications. Thus, 24 hours of running Prime does not at all ensure absolute stability. The moral of the story is that your CPU needs to be stable for the applications you run. We don't build Prime 95 PC's, we built FSX PC's, and P3D PC's and gaming PC's. For that reason I would absolutely agree with Zoran. Stable in your apps is the measure of success.

The above, is why I prefer RealBench or similar (rather than synthetic stress tests) that runs an  entire gamut of tests. RealBench is designed to stress your CPU more akin to how you use it in daily use. What I will say is that OCCT does sometimes seem to catch unstable overclocks when other stress tests don't.

The other point re Prime 95 is that it will run AVX and absolutely hammer your CPU in ways you don't normally. Expect temps to be way higher than normal. If you insist on running Prime 95, I would say use the older version that doesn't run AVX. Running Prime for 24 hours is pretty much an outdated thinking. 26.6 is the non AVX version. 

Overclock, quick test with Cinebench,  2 to 4 hour extended test with RealBench is all I usually need to do. Have tested troublesome overclocks with OCCT. 

[martin-w 3,234]

39 minutes ago, Benjamin J said:

Tried going to 4.7GHz but I had to crank up the voltage and I didn't like where my temps were going at that point, as they were going into the low 80s. 4.6GHz kept them around 79C. I guess I'm hitting the limit of my cooling solution... Noctua DH-15S. But, I don't think that an extra 100 or 200MHz will have a gigantic impact on FPS, so...

 

 

No, it's not the limit of your cooling solution. Your issue is Prime 95 running AVX and hammering your CPU. Use version 26.6. Or better still run RealBench. 

AVX offset can also be set in the BIOS to clock down when AVX instruction sets are encountered. 

Correct. Overclocking in a well balanced system is linear. 100 - 200 MHz is a very small increase in frame rate. 1 top 2 frames per second at 30 FPS. 

 

Edited March 13, 2018 by martin-w

[Benjamin J 751]

Thanks for your comprehensive contribution, as always! Your points make a lot of sense. I'll try what you describe: use Cinebench and RealBench. I'll give that older version of prime95 a whirl as well, just to see how the computer performs differently under it. The 24hr Prime95 always seemed a bit daft to me, but being a novice I wanted to make sure...

RE temps higher than normal with Prime95: definitely so, indeed. When I was just running Steam, browsing, and monitoring temps as I was using the computer, I noticed that at idle the various cores are typically around 30C, and generally not going over 50-60C when starting up a program or some such.

[lodestar 186]

2 hours ago, Benjamin J said:

I noticed that at idle the various cores are typically around 30C, and generally not going over 50-60C when starting up a program or some such.

same with my noctua. The highest temp i ever saw was 64c with P3Dv4(2700K@4.8GHz). I agree with advices above, i also recommend RealBench and no 24h tests. I would add only this - when you find stable Vcore for your OC(no BSOD), increase Vcore for just 2-3 notches(2-3 clicks on minimal increment). 

After this, if you get BSOD(ususaly 124), increase vcore until stable. Before this, ensure there is no dust, and keep in mind some GPU drivers can cause BSOD, i had this with a couple of them. Just roll back GPU driver if you get this. 

I would recommend you max Vcore arround 1.4 - 1.425, but this is up to you. Mine is at 1.4 for 6 years now, i would go for 1.425 or 1.45 but i need to add too much more Vcore to get 4.9GHz with my CPU, so i decided 4.8GHz and 1.4 Vcore looks realy nice.

start here www.overclock.net    Find there overclocking guide, you can use skylake guide becouse they are pretty much same. Check average Vcore for your OC. 

BTW, most of people have not heard of this https://click.intel.com/tuningplan/   If you want to be on the safe side, you can buy it for your CPU for $40.00. Intel will replace your CPU if something happens while overclocking. They usually don't ask the questions, but you should not go for more then 1.5 - 1.52 Vcore - i'm not sure which is the max for your CPU according to Intel, was 1.52 for mine. I've read somewhere 1.5 for your's

Edited March 13, 2018 by lodestar

[Benjamin J 751]

Thanks for the additional comments Zeljko, this is some solid info to go on, and I'm happy to see broader agreement in the thread. That's usually a good sign. I've got my processor at a vcore of 1.25 and the clock at 4.6GHz - sounds like I will have a lot of additional room to play with. Before I'll go ahead I'll do a benchmark with RealBench, as you suggest.

I see the BSOD mentioned several times now as a marker for an unstable computer. I gotta ask: is this something I might encounter during the Realbench on Cinebench tests? Or would it start to pop up randomly when doing anything else as well? I guess, in a broader light, I'm wondering what the read-out for these tests exactly is.

[lodestar 186]

2 hours ago, Benjamin J said:

Thanks for the additional comments Zeljko, this is some solid info to go on, and I'm happy to see broader agreement in the thread. That's usually a good sign. I've got my processor at a vcore of 1.25 and the clock at 4.6GHz - sounds like I will have a lot of additional room to play with. Before I'll go ahead I'll do a benchmark with RealBench, as you suggest.

I see the BSOD mentioned several times now as a marker for an unstable computer. I gotta ask: is this something I might encounter during the Realbench on Cinebench tests? Or would it start to pop up randomly when doing anything else as well? I guess, in a broader light, I'm wondering what the read-out for these tests exactly is.

1.25 Vcore looks realy low to me for 4.6 GHz. And yes, BSOD should happen during the Realbench or Cinebench if OC is unstable. You should add Vcore if this happens. Realbench or Cinebench will stress your CPU like everyday usagge is going to stress it at max. I find Realbench pretty same as MSFS based sims, usagge and temps are same. If you get random BSOD's while not testing, your OC is probably unstable, if not dust or GPU driver, like i already said. I belive your Vcore is too low for 4.6GHz, find some Vcore charts for your CPU - i'll try to find some for you tomorow, i'm tired now after work(but after a quick look, it seems to be arround 1.36 - 1.37). 

Edited March 14, 2018 by lodestar

[G-RFRY 1,706]

It`s worth checking out this guy you'll know why higher end motherboards are more expensive note the VRM  and why some boards are not stable on high OC.

 https://www.youtube.com/watch?v=CoUtA7DKXhU

After several home builds the motherboard is very important to me using ASUS A1 Suite software after running for 4months re-run the software it pushed me to 5ghz not delided it tried to go to 5.1 but the failed my system is setup to run extreme in P3Dv4 linked to my exe, in A1 software, the rest of the time half my case fans stop not needed.  

[Benjamin J 751]

13 hours ago, lodestar said:

1.25 Vcore looks realy low to me for 4.6 GHz. And yes, BSOD should happen during the Realbench or Cinebench if OC is unstable. You should add Vcore if this happens. Realbench or Cinebench will stress your CPU like everyday usagge is going to stress it at max. I find Realbench pretty same as MSFS based sims, usagge and temps are same. If you get random BSOD's while not testing, your OC is probably unstable, if not dust or GPU driver, like i already said. I belive your Vcore is too low for 4.6GHz, find some Vcore charts for your CPU - i'll try to find some for you tomorow, i'm tired now after work(but after a quick look, it seems to be arround 1.36 - 1.37). 

I've done Prime95 testing, as well as several RealBench tests and found that it's pretty stable at 1.25V, no BSOD or anything... I've seen some other posts on forums claiming similar GHz ranges with Vcores similar to mine, so it seems pretty usual to me? Found one thread by a guy who had significant trouble though, so I can only assume that that specific person had a 'bad' CPU, as in he 'lost the silicon lottery'.

Thanks for your explanation about when to expect the BSOD, that helps. The RealBench test didn't drive my temps even over 60C, so I feel more comfortable pushing clock and Vcore higher. And as you all said, RealBench seems to do a test that's overall much more realistic. Though note I'v only done the 'benchmark' at this time, not the 'stress test', nor have I used Cinebench yet. I'll try it sometime today.

The one thing that eludes me about RealBench, however, is the final score it returns... I have no idea how to interpret this. I'm getting a 90.000-100.000, and have no idea if this is good or not. I've read a little about it and found some wildly varying numbers for other people, but it seems to really depend on whether you close all kinds of background processes and apps, of which I closed none. I literally booted up the computer and did the RealBench benchmark without bothering closing anything.

[Benjamin J 751]

Update: up to 4.8GHz with vCore at 1.3V. Stable through CineBench, Realbench benchmarking as well as a 15 minute Realbench stresstest, with temps peaking at 74C. I suppose I might as well try to push the clock a little further...? Or should I submit the computer to a longer Realbench stress test first? Say, 2-4 hours, as Martin previously suggested?

Still no idea what that Realbench score is supposed to mean, but I take it that me not closing any background programs might mess with the benchmarking, making the score rather meaningless...

[lodestar 186]

don't worry about score. Which version of RealBench you have? This is the latest http://www.majorgeeks.com/files/details/asus_realbench.html  It's not yet officially released, so you can't get it from asus. A lot of people have reported low scores with previous versions. It's not realy important, but i recommend you to run benchmark too, after you get your CPU stable, and compare results with other people with similar systems. Sometimes, it can show you if something is not configured properly, if you get low score. Check this for windows configuration, it's realy the best https://www.tweakhound.com/  I'm using it since windows xp, you have there installation and tuning guides for all versions of windows. 

 Now, i did a quick research after work, and you should run RealBench stress test(not benchmark)for 2h. This is a recommendation by someone from ASUS. I didn't asked, do you overclock manualy or using MB utilities for auto OC? If you wan't, you can post your settings, maybe i can help. BTW, you are right about Vcore, coffee lake needs lower voltages compared to earlier generations. I'll try to find some links, comparison charts and guides and i'll post them later. For now, from what i've read, your temps are a bit high, that's normal if yor CPU is not delidded(first thing i would do BTW, but only using something like this https://rockitcool.myshopify.com/products/re-lid-kit - be aware about intel tuning plan and and warranty, delidding immediately voids the warranty). Your temps are actually ok, but too much for my taste and i think your CPU can do better. 

I'll try to find some usefull stuff for you and i'll post that later. I'm interested in this, becouse i'm saving for a new PC, and i need to decide about components. If you have time, please post your CPU vs GPU usagge with P3Dv4 and P3D settings. Before i buy new PC, i need to buy new GPU, and after a lot of calculations and reading - 1060 or 1070/1070ti would be the best match for my CPU, so i'm interested about performance with your 1070 

[martin-w 3,234]

18 hours ago, Benjamin J said:

Update: up to 4.8GHz with vCore at 1.3V. Stable through CineBench, Realbench benchmarking as well as a 15 minute Realbench stresstest, with temps peaking at 74C. I suppose I might as well try to push the clock a little further...? Or should I submit the computer to a longer Realbench stress test first? Say, 2-4 hours, as Martin previously suggested?

Still no idea what that Realbench score is supposed to mean, but I take it that me not closing any background programs might mess with the benchmarking, making the score rather meaningless...

 

You have the i5 don't you? So no HT. In which case your voltage at 4.8 looks about average to me. 74C is fine. make sure your fan curve is set properly of course, so your fans ramp up when they need to. Your CPU temp is also influenced by your ambient temp don't forget. 

15 minutes stress test is fine while you're finding your limit. Once you reach your limit, the max frequency and minimum voltage, up the voltage a fraction as a margin of error. When you reach the frequency you are happy with and the thermals area good, run RealBench for 2 hours. I run 2 hours minimum, and sometimes up to 4 hours. 

RealBench score, like any benchmark score is for comparison. Don't worry about it. It's the stress test that's important for you, to ensure your rig is stable. After your OC is done and stable, you can run the benchmark if you like, and compare to others scores. 

Once you're happy with your OC, save it as a profile in the UEFI. You can then switch back to optimised defaults if you like, try the benchmark, and then try your OC and see the results. Whatever you fancy. :)

 

My 8700K is a particularly bad overclocker, but with the NH-D15S and with HT off, and running the Asus 5 GHz OC profile, I get...

RealBench for 15 mins.  

Volts manually set to to 1.35. (LLC took it to 1.376)

77 degrees. Ambient was about 20 C. 

Passed!

Interestingly  there was no difference in temp despite dropping the voltage compared to the Asus profile voltage. 

Edited March 15, 2018 by martin-w

[Benjamin J 751]

Zeljko, thanks for taking the time and trouble to help out like that, I greatly appreciate it! The RealBench verison I have is the one that can be currently downloaded form the ROG website - when I get home I can let you know what version it actually is. I can have a look at tweakhound too, thanks for the link! I did not delid, though I read about it. As a first time builder, I decided not to get too crazy ;)

Martin, thanks to you too! You've been an endless source of good advice as I've gone through the process of building this rig and  configuring it. Much appreciated!

Glad to hear that the score is unimportant... :) Stuff like that ends up stressing me out.

It's good to hear that my current settings are considered "normal" and temps are fine. Indeed this is the i5-8600k. I upped the clock to 4.9GHz, still at Vcore 1.3, and subjected it to a 2hr Realbench stresstest, which it passed with temps not going over 76C. I reckon the average temps were probably more along the lines of 70C though. In the meantime I installed Prepar3D as well, but haven't really installed any addons yet. Just took it for a very quick spin which, unsurprisingly, worked well. So, Zeljko, I can't give you those settings and usages quite yet, but down the line I can definitely shoot you a message with those statistics.

Given how my computer is handling the current settings, I'm going for a final clock of 5GHz, and I'd really rather not go over a Vcore of 1.3, seeing as my temps are edging towards 80C. So, following your both advice, if I can hit 5GHz stably at a Vcore of 1.3, I'll start backing off on the Vcore until I hit the minimum possible, doing the 15min Realbench stress tests between each change, until I hit an issue. Then up the Vcore again by a notch or two and do the 4hr Realbench stress test. then, I guess, the proof of the pudding is in the eating: I'll install some demanding P3D addons, and take it for a spin!

One more thing: What about GPU OC'ing?