Gentlemen,

I am a new proud owner of an i9 9900k and have spent the last couple of days trying to achieve a good stable overclock. I am by no means an expert in overclocking and am plainly learning along while I do (reading from many overclocking forums etc.)

So far I have achieved 5.0 GHz on all cores with Vcore 1.29 and LLC Level 6. If I go above 1.29 the temps get out of control on Prime95 Small FFTs (non AVX). On the Small FFTs run my temps get to 95C after 15 minutes without crashes, which I am okay with since under any real load they won't ever go that high. X-Plane is always well below 80C for instance.

Since X-Plane does not benefit from more than 4 cores really (see https://youtu.be/idredFBcd70), is there a benefit in trying to clock 4 cores at 5.1 or 5.2 while lowering the clock speed on the other 4 cores? I wouldn't mind going down to the stock 3.6 on half the cores if by having 5.1+ on 4 cores improves X-Plane performance.

Lastly, how do you test for stability with flight simulators in mind? I can run the infinite loop on the Asus Realbench benchmark for 8 hours with no crash. Prime95 Blend (no AVX) however, does not crash but stops workers due to the errors, technically I'm not 100% stable then.

Any comments/discussion is appreciated. Regards,

Jaime

I gave up testing with Prime long ago and just use Realbench. Even that produces more heat than P3d V5. Try my settings if you like. MSI z390 motherboard.

 

 

3 hours ago, DescendDescend said:

I gave up testing with Prime long ago and just use Realbench. Even that produces more heat than P3d V5. Try my settings if you like. MSI z390 motherboard.

 

Thanks Howard, I'll try your settings then!

I believe pursuing Prime95 complete stability only makes sense if the system is built for running Prime95 in the first place! But many OCers disagree.

I am using Prime95 Small FFTs for temperature testing. If it doesn't exceed 95C after 15 minutes I consider it a pass. For stability I will only use Realbench then.

EDIT: Forgot to mention my cooling, I am using a Noctua NH-U14S.

Edited May 28, 20206 yr by Alpha Floor

On 5/27/2020 at 11:29 AM, Alpha Floor said:

So far I have achieved 5.0 GHz on all cores with Vcore 1.29 and LLC Level 6. If I go above 1.29 the temps get out of control on Prime95 Small FFTs (non AVX). On the Small FFTs run my temps get to 95C after 15 minutes without crashes, which I am okay with since under any real load they won't ever go that high. X-Plane is always well below 80C for instance.

Since X-Plane does not benefit from more than 4 cores really (see https://youtu.be/idredFBcd70), is there a benefit in trying to clock 4 cores at 5.1 or 5.2 while lowering the clock speed on the other 4 cores? I wouldn't mind going down to the stock 3.6 on half the cores if by having 5.1+ on 4 cores improves X-Plane performance.

Lastly, how do you test for stability with flight simulators in mind? I can run the infinite loop on the Asus Realbench benchmark for 8 hours with no crash. Prime95 Blend (no AVX) however, does not crash but stops workers due to the errors, technically I'm not 100% stable then.

With my 9900K and NH-D15, I pushed it to no higher than ~80 deg on any core when running P95 without AVX, and then I run P3D at those settings with the AVX offset set to zero, knowing that P3D doesn't hit the CPU nearly as hard as P95.  It's worked well so far with P3D v4.5, which does use a fair amount of AVX in the code.  I have no idea how v5 compares in that regard.  I'm at 5.0 GHz with temps mostly in the 60s while running P3Dv4.5.  I can't recall the core voltage.  If I try P95 with AVX on using the same settings it'll push the CPU right up into throttling.

You can't separately clock cores independently on any Z390 mobo that I know of...the "per core" feature in the BIOS sets a single CPU clock multiplier for all 8 cores based on the number of physical CPUs it sees under load.  So the "Core 6" mult value really means set this mult when six physical cores are loaded, not the speed at which you want to independently run core 6.

6 minutes ago, w6kd said:

With my 9900K and NH-D15, I pushed it to no higher than ~80 deg on any core when running P95 without AVX,

That´s interesting. I hace a NH-U14S and my temps are much higher. When you say P95, do you mean Small FFTs?

Starting to believe I got unlucky with my chip...

10 minutes ago, Alpha Floor said:

That´s interesting. I hace a NH-U14S and my temps are much higher. When you say P95, do you mean Small FFTs?

Starting to believe I got unlucky with my chip...

I forgot to mention...HT is off.  If you're using HT, that might explain it.

2 hours ago, w6kd said:

I forgot to mention...HT is off.  If you're using HT, that might explain it.

So you disabled it on the BIOS?

The reason I was using Process Lasso in the first place was to be able to disable HT for X-Plane ONLY, but not the other processes. If I disable HT in the BIOS, what's the point of the 9900k anyway? Should have gotten a 9700k...

EDIT: Indeed, disabling HT lowered the temps substantially. It seems like I am going to be able to get 5.2 GHz at 1.35 Vcore with LLC6. 

Edited May 29, 20206 yr by Alpha Floor

9900K has larger L3 cache 16Mb Vs 12Mb than 9700k. Thats about it. Definitely HT off.

NH-D15 for me also.

EIST on so that volts low on light load.

 

Edited May 29, 20206 yr by DescendDescend

Folks,

Maybe someone is interested in these numbers I will present. Having OCed the CPU to 5.1 GHz with HT OFF, this is the HWInfo report after a 2.5h session in X-Plane 11.41 in which I flew two VFR flights within the Balearic Islands, with the JustFlight PA28R Turbo Arrow, SpainUHD orthos, Active Sky XP (with stormy weather) and the following settings:

Graphics quality: HDR; Texture quality: maximum; Anti aliasing: 4x SSAA + FXAA; Objects: maxium; Reflections: off; No shadows checked. Resolution: 1440p.

FPS during the flight: consistently above 50 FPS, I would say average 60 FPS with peaks of up to 100-110 FPS.

I reset HWInfo before starting X-Plane and took the screenshot right after closing X-Plane so that the average values were representative.

On 5/29/2020 at 11:50 AM, DescendDescend said:

9900K has larger L3 cache 16Mb Vs 12Mb than 9700k. Thats about it. Definitely HT off.

NH-D15 for me also.

EIST on so that volts low on light load.

 

Hi Howard,

What is your take on the HWInfo screenshot from above?

I still can't get to pass Asus Realbench stability testing. It detects instability after around 15 minutes into the stress test. Any ideas as to what I could do other than increasing Vcore as that will push up my temps too high?

I'm using Realbench 2.43. Not sure if that is any different to the Asus version. I guess you are just using the stress test part? I'm also using the latest CPUID Hardware Monitor Pro for measurements. Even that was reading an incorrect Vcore a couple of versions ago although it is correct on version 1.38. I checked it with a voltmeter on the motherboard and against MSI's own hardware monitoring tool which was also accurate.

It maybe that your CPU can't run at the voltages I use. Did you try my settings? Not sure what the equivalent is on Asus. Maybe something else is causing instability.

Regarding thermal control I adjusted the fan curves in the EUFI. My single CPU fan on the Noctua is set to run at around 1300+ rpm (roughly max speed) from around 75C upwards. I see yours says 994 rpm max speed. That keeps temperatures in check as long as you have a decent case with good airflow.

42 minutes ago, DescendDescend said:

I'm using Realbench 2.43. Not sure if that is any different to the Asus version. I guess you are just using the stress test part? I'm also using the latest CPUID Hardware Monitor Pro for measurements. Even that was reading an incorrect Vcore a couple of versions ago although it is correct on version 1.38. I checked it with a voltmeter on the motherboard and against MSI's own hardware monitoring tool which was also accurate.

It maybe that your CPU can't run at the voltages I use. Did you try my settings? Not sure what the equivalent is on Asus. Maybe something else is causing instability.

Regarding thermal control I adjusted the fan curves in the EUFI. My single CPU fan on the Noctua is set to run at around 1300+ rpm (roughly max speed) from around 75C upwards. I see yours says 994 rpm max speed. That keeps temperatures in check as long as you have a decent case with good airflow.

Thanks very much, I will try increasing the CPU fan speed! I am using a Noctua NH-U14S with one fan.

I will also try your settings and I will look for Realbench, the non-Asus version.

Two days ago I run Memtest86, the full 4 passes which took almost 6h. Two errors were detected runing Test 8. I went into the BIOS and saw the DRAM Voltage set to 1.35, I changed it to Auto, re-run Memtest86 and now it passed it with no errors. However, Auto seems to set 1.51V on the DRAM, is this not too much? My RAM is Corsair Vengeance DDR4 at 3200 MHz. I am using XMP II profile. What's the difference to XMP I anyway?

You don't need to disable HT if you set up the Affinity Mask properly. By turning HT on with no affinity mask your sim will eat all 16 logical processors which is way over the top since you only need 8 at the most. Even so keeping to HT off, setup is a no brainer because it's less trouble to work out how to do a proper setup.

With 5GHz say you get a nice 30frames per second. With 5.2 that's 5.2 / 5 = 1.04, so 1.04 x 30 = 31.2 which is a nice gain of 1.2fps for the trouble you will cause and the hassle of stability. In my rigs I'm at 4.8 and If I went up to 5.2 I'll get maybe another 2fps. I prefer to set up properly, ignore ultra high GHz and enjoy stability.

32 minutes ago, SteveW said:

You don't need to disable HT if you set up the Affinity Mask properly. By turning HT on with no affinity mask your sim will eat all 16 logical processors which is way over the top since you only need 8 at the most. Even so keeping to HT off, setup is a no brainer because it's less trouble to work out how to do a proper setup.

With 5GHz say you get a nice 30frames per second. With 5.2 that's 5.2 / 5 = 1.04, so 1.04 x 30 = 31.2 which is a nice gain of 1.2fps for the trouble you will cause and the hassle of stability. In my rigs I'm at 4.8 and If I went up to 5.2 I'll get maybe another 2fps. I prefer to set up properly, ignore ultra high GHz and enjoy stability.

Thanks Steve,

The other argument in favour of HT ON is that my PC is not dedicated 100% to X-Plane, I'm doing other things on it and HT on is nice to have for productivity etc.

How do I set Affinity Mask?

I used to use Process Lasso but gave it up after a while as I thought that it didn't really help at all and I had some suspicions that it was preventing X-Plane from loading scenery correctly during flight.

OK. That's the same with me. If I convert a movie, run Paint3D or something like that, I want all the power I can get which is all Logical Processors enabled. The movie program can then convert as many frames as possible in the shortest time.

But with these simulators they work differently. They need a good high frequency on the first core then the other cores will run in parallel to read in the data and assemble the scene, run the system resources and run the addon exe apps. With HT enabled they run the parallel side and the exe apps faster and more efficiently.

With HT enabled you should use the [jobscheduler] section for the Prepar3D Affinity Mask. Use Proc Lasso or a batch file for corralling the addon exe processes near the later cores.

As a starting point I would say go for AM=21845=01,01,01,01,01,01,01,01 which is 8 cores running with one LP per core.

With Proc Lasso dedicate 10,10,10,10,00,00,00,00 to the exe apps.

Bring the OC down a little to accommodate the extra throughput since heat=work.

 

Those guys with all the fans running full pelt? Their PCs push a lot of air but they'll never got off the ground. 😊