How to determine if an "OverClock" is stable?

[imikejackson]

I have my Hackintosh up and running very nicely thanks to all the posts in the forums. My CPU is an 8700K and I have it overclocked to 4.8GHz and everything seems to run just fine. The issue is that sometimes when I compile our software the actual compiler seg-faults and dies. I have never seen this happen with the Apple supplied clang with our software. For comparisons I compile our software on my 2019 16" MacBook pro running the same versions of Xcode/Clang/Catalina.

Since this is my first foray into overclocking could it be that I pushed this CPU a bit too far and have some instability in the system? I guess it could be my RAM also? Just trying to figure out some places to start investigating. My temperatures run in the mid 80s (C) when I compile and the compiles will last for about 20 minutes using 100% of the CPU. I saw on some other overclocking forums that there are some benchmarks that could be run to ensure that all is well with the CPU?

The compile failure is not consistent which makes it harder to diagnose. Any suggestions where to start would be great. I could just back off the overclocking back to stock and see if the error ever comes back over the next few weeks I guess. Just hate to loose the overclock.

 

[negatv]

If crashing under load during a compile, certainly wouldn’t call it stable. Easy enough to test- dial it back a little bit and continue testing with what you are doing when it crashes. Might have to bump some voltages or whatnot, or it just is never going to happen at that clock rate.

You won’t be “losing” anything, except maybe the silicon lottery. Its easy to boot up an over clocked machine at a clock rate that makes your proud, but under no load it means nothing. If it’s not 100% stable and reliable, its kind of useless. IMO.

 

[pastrychef]

When I test overclock stability:

• I run Prime95 torture test with small FTFs for 2-3 hours. Watch temperatures because this will really heat up your CPU.
• Then, I run Geekbench 20-30 times and make sure it completes every run without incident.
• Then, I run Handbrake and queue up a few videos and let it convert overnight.

If it does all three without errors, crashes, or freezes, I deem the overclock stable. If there are any issues, I bump up vCore a bit and test again. I try to keep temperatures ~85C or lower.

These are the Prime95 settings I use:

 

[imikejackson]

Thanks @pastrychef. I downloaded the Prime95 and started running it. My temps are hitting 100C so that is not where I want to be. I'm also running this remotely so I cannot physically listen to the fans on the machine. The CPU is air cooled. I may have to bump down the OC to drop the temperatures. I may venture into the office to adjust the BIOS settings then start a long run of Prime95.

 

[pastrychef]

Thanks @pastrychef. I downloaded the Prime95 and started running it. My temps are hitting 100C so that is not where I want to be. I'm also running this remotely so I cannot physically listen to the fans on the machine. The CPU is air cooled. I may have to bump down the OC to drop the temperatures. I may venture into the office to adjust the BIOS settings then start a long run of Prime95.

Air cooling is fine when done right. I use a Noctua NH-D15S CPU cooler. I also delidded my i9-9900K. This resulted in temps that max out at around 85C when running Prime95 with CPU running at 5GHz. Your i7-8700K should be able to run quite a bit cooler than the 9900K.

If you are hitting 100C, the CPU will throttle. You can lower the voltage and/or adjust LLC to get the CPU to run cooler.

As I stated above, I just run Prime95 2-3 hours. You can queue up some videos to convert in Handbrake and let it run overnight.

 

[imikejackson]

I went back and checked the machine. I am also using the same cooler as @pastrychef. I have it in a NZXT H510 case. I adjusted the LLC to level 5 on the ASUS Prime Z370-A Rev 1 board. Tried 5GHZ but it KPs and restarts pretty quick. I bumped the voltage up to 1.37V and put a 1 for the AVX instruction offset. Set the clock back to 4.8GHZ. When doing Prime95 it runs at 4.7GHz and in the high 80s to low 90s. Compiling is about the same speed/temp. I'll let Prime95 go the reset of the afternoon and see if it ever reboots.

 

[pastrychef]

I went back and checked the machine. I am also using the same cooler as @pastrychef. I have it in a NZXT H510 case. I adjusted the LLC to level 5 on the ASUS Prime Z370-A Rev 1 board. Tried 5GHZ but it KPs and restarts pretty quick. I bumped the voltage up to 1.37V and put a 1 for the AVX instruction offset. Set the clock back to 4.8GHZ. When doing Prime95 it runs at 4.7GHz and in the high 80s to low 90s. Compiling is about the same speed/temp. I'll let Prime95 go the reset of the afternoon and see if it ever reboots.

That case looks pretty air constrained. You can experiment by running the system with side panel off and see how temps are.

 

[imikejackson]

Turns out the air cooler was installed 90 degrees rotated so I fixed that and in the process re-applied thermal paste (Noctua brand paste). Between those two items my temps running prime95 torture test are in the low 80s. I left all other settings in BIOS the same. The test lasted about an hour then it KP'ed and restarted. I was also doing a compile of our software just to really stress the system. The air exiting the case now "feels" much cooler than it was but the KP is still nagging at me a bit. Something is still funny. I have noticed with this MB that the heat plate that goes on top of the NVMe ssd is extremely hot to the touch. I'm wondering since compiling it IO intensive, if the drive is over heating and causing errors. I have a PCIe x16 card with a much larger heat sink on it for SSDs and I wonder it that would help with temperatures of the SSD.

I loaded up the FakeSMC kexts for the various components (CPU, GPU, ACPI) but one of those turns off the rear fan when loaded so that is a problem. I'm also worried about the core voltage setting in the BIOS being 1.37V. Assuming this is actually the case this seems safe enough but after watching some videos over the weekend (GamersNexus has a nice one) I'm going to bring in a multimeter to try and measure the actual voltage.

Does anyone know how to read a KP dump? I'm curious if one can get a definitive reason as to what caused the KP.

 

[pastrychef]

Turns out the air cooler was installed 90 degrees rotated so I fixed that and in the process re-applied thermal paste (Noctua brand paste). Between those two items my temps running prime95 torture test are in the low 80s. I left all other settings in BIOS the same. The test lasted about an hour then it KP'ed and restarted. I was also doing a compile of our software just to really stress the system. The air exiting the case now "feels" much cooler than it was but the KP is still nagging at me a bit. Something is still funny. I have noticed with this MB that the heat plate that goes on top of the NVMe ssd is extremely hot to the touch. I'm wondering since compiling it IO intensive, if the drive is over heating and causing errors. I have a PCIe x16 card with a much larger heat sink on it for SSDs and I wonder it that would help with temperatures of the SSD.

I loaded up the FakeSMC kexts for the various components (CPU, GPU, ACPI) but one of those turns off the rear fan when loaded so that is a problem. I'm also worried about the core voltage setting in the BIOS being 1.37V. Assuming this is actually the case this seems safe enough but after watching some videos over the weekend (GamersNexus has a nice one) I'm going to bring in a multimeter to try and measure the actual voltage.

Does anyone know how to read a KP dump? I'm curious if one can get a definitive reason as to what caused the KP.

Low 80s is great especially when considering the 1.37v. 1.37v is quite a lot, in my opinion. I run my 9900K at 1.29v at 5GHz.

Try different voltage and LLC combinations.

If your NVMe SSD overheats, in a worse case scenario, it should just throttle. It shouldn't cause a reboot.

Another thing that can cause instability with an overclock is VRM temps.

If your system runs stable without the overclock, then I think it's safe to assume that the overclock is the cause of your kernel panics. That would render the panic logs pretty useless since it was caused by the overclock.

 

[imikejackson]

I have access to a Corsair cwcH100 from another system that would fit in the case if we really thought it would help more than the Noctua Cooler presently installed. I have been running Prime95 for the last 2 hours (while working on the computer) using a heavy workload that should max the temps and I see a steady mid 80s with a spike to 90 occasionally. Your prompts have also lead me to read more articles about the Prime z370-A board, its VRM and other deep technical information. Now that I have the air cooler straightened out a bit I should back off the voltage. Most articles suggest that the 8700K should run at 1.3Volts. I have the current LLC level set to 5. I could bump that to 6 which should help out the VRMs a bit.