Achieve stability on your 8700K PC
Achieve stability on your 8700K PC
Hello everyone,
I will first describe the situation in short, and then there will be details. If any additional details are required, please ask, I will provide them. Here we go.
About a month ago I purchased CPU/MB/RAM/GPU/cooler - there was not enough money to buy everything anew in one go, so I chose to get better components and use them temporarily with my case/PSU/HDD. I can't get this PC build stable and it drives me nuts. There were some BSODs, which lead me to further testing. The only thing that reliably gives errors is prime95 Blend.
Now for details:
Specs:
http://speccy.piriform.com/results/jJkoN...o4XXcHsChT
16 GB A-Data XPG Dazzle DDR4 3000 MHz (XMP)
The timings are rough 16-18-18. Mine should be Model Number: AX4U3000316G16-DRD.
Detailed Timings/XMP read:
MSI Z370 Gaming Plus
Gigabyte GTX 1070Ti Gaming 8G
HDD WD Blue 1 GB
Windows 10 x64 PRO FCU (reinstalled anew 10 Jan)
All patched, latest BIOS (some BSODs occurred on older BIOS too)
Pretty much barebones Windows so I can troubleshoot. Only drivers loaded are latest from Intel/Nvidia besides Microsoft's.
Everything got reaseated at least once and cleaned :\
The mobo/RAM/cooler/CPU were purchased in Dec 2017. I replaced an old FX 8350/GTX 970 build, kept some parts to save money. The new build was never fully stable, there's always some sort of issue appearing. BSODs were very infrequent, yet they led me to further testing. Some Windows 10 apps are also sometimes crashing. I've noticed Task Manager crashing, or earlier today when I dragged Opera to the screen edge got it with a dwm.exe crash which manifested in Opera windows blanking on me, only to return to normal a few seconds later. Most BSODs occurred in Assassin's Creed: Origins, but even then there were days I could play without issues, only to get hit with a few crash to desktop from the app without any Event Viewer entry or even a BSOD later on.
Gaming for hours under Driver Verifier (for all drivers but Microsoft) didn't produce any BSODs.
I've ran memtest from boot, 4 times now, one time for 8 passes. Seagate's Long Generic Test. Chkdsk with surface scan for all partitions. Intel Burn Test, OCCT, Aida64, Smalll FFT Prime95 for 6-8 hours. These all came fine. IBT does give errors with lower SA/IO voltages.
It was suggested to me at this point to run Blend, which I didn't consider since memtest came fine after 8 passes and 4 tries. And first time, one worker failed after 5 hrs. I was not that worried as I knew I decreased the default SA/IO voltages that got inflated like nuts by applying the XMP, so I just allowed the motherboard to set the high voltages and let Blend run over night again. This time, after about 9 hours, 5-6 workers failed in short succession.
Prime95 failing with XMP/default motherboard voltages:
One interesting thing about the above screenshot:
- one worker fails @9:00 hrs, another 3 after 7 minutes, and another one after 7 more minutes. Unsure if it means anything.
The previous night, I've applied the RAM timings manually and ran Blend over night again. I am not even 100% sure I did everything right, but the timings look similar in Windows. I've tried running with the default Intel voltages on SA/IO, but it won't even boot with the 3000 MHz XMP without raising them some. For this run of Prime I've let the mobo with the stupid high voltages.
If anyone is curious, SA is 1.32 and IO is 1.264 with XMP. Without XMP they would be set much lower, but still higher than Intel's default. The RAM would be 2400 Mhz without XMP, which is a difference I'm not really prepared to run at, as it's visibly slower in benchmarks and even some games like Assassin' Creed.
So, with manual RAM timings corresponding to the XMP timings and SA/IO voltages left on auto in BIOS Blend runs for 12 hrs now, no workers stopped.
The auto SA/IO voltages seem insanely high however compared to what Intel deems appropriate, and they are even slightly above running the XMP (1.32V vs 1.30V for SA and 1.264V vs 1.250V for IO). Again, I didn't set manually these voltages, the motherboard did. I only set the DRAM voltage to 1.35V as that was specified on the kit. Even at 2400 MHz the motherboard will set the voltages above Intel's.
I can understand some boost being needed to run stable at higher RAM frequencies, but these seems excessive. I am unsure if I should run this 24/7 or just fall back to 2400 MHz. If anyone here understands exactly how these things are supposed to work and what is safe or nor for the Z370 platform, please share.
Notes:
- I've already read on RAM voltages and I know they are high for others too. I've seen the "safe" numbers in the guide for KBL, I know 8700K is still a KBL CPU, but I don't know if all those numbers still apply
- the higher voltages have an impact on CPU temp, as expected. I'm reaching max. of 72C (per core) during Blend, averages lower, around 61C
- 2 timers were set wrong by the motherboard, T RRD and T RRD-L, motherboard would set them higher ( so I assume slower) than what is written in XMP. I've manually set those as it's written on the profile now
- the motherboard will spit an "overclock failed" for very minor variations from the "default" timings - including at time for XMP itself. For example I got OC failed with SA 1.20V and IO 1.15V. At 1.21V and 1.16V we're normally into Windows - but don't expect this to be always the case.
- the successful 12 hrs test above, after I rebooted the PC... OC failed.
- at times (like once/day), it's possible that a game crashes to desktop with no errors or warnings in Event Viewer whatsoever, or a Windows app like Task Manager to appcrash. Usually I wouldn't give a damn, but with the unstable PC, i suspect everything :\
And in the end, I am now 1 hr into Blend with manual timings according to XMP, 3000 MHz, SA 1.21V, IO 1.16V (the readings from HWinfo64 are a bit higher). I fully expect this to fail since it failed at 1.3 SA and 1.25 IO (this is what XMP defaults try to set) - but I am willing to give it a shot - maybe just maybe the high voltages were causing instability. I doubt it, since 0.01V lower it was OC failed, but, nothing to lose other than more time. Gonna go to sleep with this profile running and report tomorrow.
Lastly, everything is in warranty, just bought last month. However - sending stuff back that they can't find defective through troubleshooting will result in the same stuff being sent back to me, AND I will have to pay for transport, plus troubleshooting costs. I have already made a significant financial sacrifice with this build which is crazy expensive where I live, I don't mind to RMA, but I really need to find out what first, and be able to tell them how to replicate the defects, so I can guarantee functional gear returning to me. I really don't wanna send all this back only to receive the same crap from the store, and a bill to top it all off too, not to mention being on some shitty PC for a month.
Please excuse the probable writing mistakes, this was a long post because I want to offer all the details necessary, in hopes of getting this solved.
Any assistance is greatly appreciated. Thank you!
PS: gonna go to sleep, then work, so I will be able to respond 12 hrs or so later.
Initially, it's important to note that you're starting from an incorrect assumption. You're not employing Blend to evaluate stability or thermal performance, even though it's meant for memory testing. While it's designed for assessing memory, it isn't being used for that purpose either. Instead, you're relying on one of two versions of Memtest86 for this task. No one with deep experience in Prime95 for evaluating stock or overclocked setups would advise using Blend for thermal or stability checks. If they do, they likely lack the expertise and may have been misled for a long time. What you're doing is using the Small FFT option in Prime95 version 26.6 exclusively. This is the standard method for most baseline stability and thermal limit tests in that version. Prime95 v26.6: http://windows-downloads-center.blogspot...5-266.html For Intel CPU details, you can refer to the comprehensive guide provided by Computronix, the author of the Intel temperature manual, available here: For AMD systems, especially Zen/Ryzen, this resource offers similar but less detailed information. You can also check: http://www.overclockers.com/amd-ryzen-ov...ing-guide/ This is likely the most widely referenced guide for overclocking, with principles applicable across generations and platforms. Another useful source is: http://www.overclock.net/t/91/ultimate-o...king-guide Lastly, "REinstalled" differs from a CLEAN install when it comes to the operating system. You appear reasonably informed, so I assume you're aware of this distinction, though many experienced system administrators overlook key differences—such as the fact that a reinstall doesn't recreate the boot partition in most cases. If your system has any drives with Windows installed and those partitions were never fully removed, they might still contain boot partitions that interfere with hardware boot management. If you've only performed a reinstall, update, upgrade, reset, or repair, it's advisable to do a CLEAN install. This involves disconnecting all other drives, deleting every partition from the target drive, and then installing the latest OS ISO into an unpartitioned space. Windows will automatically create partitions, including a new boot partition based on your BIOS settings (for MBR or GPT), and perform necessary formatting. If you haven't done this before, I strongly recommend it. Windows 10 Clean install guide I also wanted to emphasize the importance of having the latest BIOS version—preferably the second most recent one. However, you've already updated. You might consider reverting to the second-to-last version if it includes microcode fixes for Spectre and Meltdown, as Intel has acknowledged issues with those BIOS versions and is working on revisions. Instability could stem from this, even if it's not the sole cause. If possible, aim for a BIOS version prior to January 1st. For now, I won't update my Skylake system's BIOS unless a future release specifically addresses side-channel attack vulnerabilities without impacting SSD or PCIe M.2 performance. Spectre and Meltdown can only compromise security if malicious software is already present; otherwise, as long as you maintain strict security updates, it's safe. For memory testing, many users stick with Memtest86 and Memtest86+. Your approach is sound—leave your memory at default SPD settings until you're confident your CPU overclock is stable and thermally efficient, since poor CPU overclocking can significantly impact memory performance and stability, especially because the memory controller resides within the CPU. Memory can also influence CPU behavior, but typically we prioritize the CPU first. It's often impossible to achieve maximum advertised speed in memory without first stabilizing the CPU through overclocking. Once the CPU is stable, proceed with configuring memory, then thoroughly test it using Memtest for at least seven passes—this process can take several hours.
toyo :
Hello everyone,
I will first describe the situation in short, and then there will be details. If any additional details are required, please ask, I will provide them. Here we go.
About a month ago I purchased CPU/MB/RAM/GPU/cooler - there was not enough money to buy everything anew in one go, so I chose to get better components and use them temporarily with my case/PSU/HDD. I can't get this PC build stable and it drives me nuts. There were some BSODs, which lead me to further testing. The only thing that reliably gives errors is prime95 Blend.
Now for details:
Specs:
http://speccy.piriform.com/results/jJkoN...o4XXcHsChT
16 GB A-Data XPG Dazzle DDR4 3000 MHz (XMP)
The timings are rough 16-18-18. Mine should be Model Number: AX4U3000316G16-DRD.
Detailed Timings/XMP read:
MSI Z370 Gaming Plus
Gigabyte GTX 1070Ti Gaming 8G
HDD WD Blue 1 GB
Windows 10 x64 PRO FCU (reinstalled anew 10 Jan)
All patched, latest BIOS (some BSODs occurred on older BIOS too)
Pretty much barebones Windows so I can troubleshoot. Only drivers loaded are latest from Intel/Nvidia besides Microsoft's.
Everything got reaseated at least once and cleaned :\
The mobo/RAM/cooler/CPU were purchased in Dec 2017. I replaced an old FX 8350/GTX 970 build, kept some parts to save money. The new build was never fully stable, there's always some sort of issue appearing. BSODs were very infrequent, yet they led me to further testing. Some Windows 10 apps are also sometimes crashing. I've noticed Task Manager crashing, or earlier today when I dragged Opera to the screen edge got it with a dwm.exe crash which manifested in Opera windows blanking on me, only to return to normal a few seconds later. Most BSODs occurred in Assassin's Creed: Origins, but even then there were days I could play without issues, only to get hit with a few crash to desktop from the app without any Event Viewer entry or even a BSOD later on.
Gaming for hours under Driver Verifier (for all drivers but Microsoft) didn't produce any BSODs.
I've ran memtest from boot, 4 times now, one time for 8 passes. Seagate's Long Generic Test. Chkdsk with surface scan for all partitions. Intel Burn Test, OCCT, Aida64, Smalll FFT Prime95 for 6-8 hours. These all came fine. IBT does give errors with lower SA/IO voltages.
It was suggested to me at this point to run Blend, which I didn't consider since memtest came fine after 8 passes and 4 tries. And first time, one worker failed after 5 hrs. I was not that worried as I knew I decreased the default SA/IO voltages that got inflated like nuts by applying the XMP, so I just allowed the motherboard to set the high voltages and let Blend run over night again. This time, after about 9 hours, 5-6 workers failed in short succession.
Prime95 failing with XMP/default motherboard voltages:
One interesting thing about the above screenshot:
- one worker fails @9:00 hrs, another 3 after 7 minutes, and another one after 7 more minutes. Unsure if it means anything.
The previous night, I've applied the RAM timings manually and ran Blend over night again. I am not even 100% sure I did everything right, but the timings look similar in Windows. I've tried running with the default Intel voltages on SA/IO, but it won't even boot with the 3000 MHz XMP without raising them some. For this run of Prime I've let the mobo with the stupid high voltages.
If anyone is curious, SA is 1.32 and IO is 1.264 with XMP. Without XMP they would be set much lower, but still higher than Intel's default. The RAM would be 2400 Mhz without XMP, which is a difference I'm not really prepared to run at, as it's visibly slower in benchmarks and even some games like Assassin' Creed.
So, with manual RAM timings corresponding to the XMP timings and SA/IO voltages left on auto in BIOS Blend runs for 12 hrs now, no workers stopped.
The auto SA/IO voltages seem insanely high however compared to what Intel deems appropriate, and they are even slightly above running the XMP (1.32V vs 1.30V for SA and 1.264V vs 1.250V for IO). Again, I didn't set manually these voltages, the motherboard did. I only set the DRAM voltage to 1.35V as that was specified on the kit. Even at 2400 MHz the motherboard will set the voltages above Intel's.
I can understand some boost being needed to run stable at higher RAM frequencies, but these seems excessive. I am unsure if I should run this 24/7 or just fall back to 2400 MHz. If anyone here understands exactly how these things are supposed to work and what is safe or nor for the Z370 platform, please share.
Notes:
- I've already read on RAM voltages and I know they are high for others too. I've seen the "safe" numbers in the guide for KBL, I know 8700K is still a KBL CPU, but I don't know if all those numbers still apply
- the higher voltages have an impact on CPU temp, as expected. I'm reaching max. of 72C (per core) during Blend, averages lower, around 61C
- 2 timers were set wrong by the motherboard, T RRD and T RRD-L, motherboard would set them higher ( so I assume slower) than what is written in XMP. I've manually set those as it's written on the profile now
- the motherboard will spit an "overclock failed" for very minor variations from the "default" timings - including at time for XMP itself. For example I got OC failed with SA 1.20V and IO 1.15V. At 1.21V and 1.16V we're normally into Windows - but don't expect this to be always the case.
- the successful 12 hrs test above, after I rebooted the PC... OC failed.
- at times (like once/day), it's possible that a game crashes to desktop with no errors or warnings in Event Viewer whatsoever, or a Windows app like Task Manager to appcrash. Usually I wouldn't give a damn, but with the unstable PC, i suspect everything :\
And in the end, I am now 1 hr into Blend with manual timings according to XMP, 3000 MHz, SA 1.21V, IO 1.16V (the readings from HWinfo64 are a bit higher). I fully expect this to fail since it failed at 1.3 SA and 1.25 IO (this is what XMP defaults try to set) - but I am willing to give it a shot - maybe just maybe the high voltages were causing instability. I doubt it, since 0.01V lower it was OC failed, but, nothing to lose other than more time. Gonna go to sleep with this profile running and report tomorrow.
Lastly, everything is in warranty, just bought last month. However - sending stuff back that they can't find defective through troubleshooting will result in the same stuff being sent back to me, AND I will have to pay for transport, plus troubleshooting costs. I have already made a significant financial sacrifice with this build which is crazy expensive where I live, I don't mind to RMA, but I really need to find out what first, and be able to tell them how to replicate the defects, so I can guarantee functional gear returning to me. I really don't wanna send all this back only to receive the same crap from the store, and a bill to top it all off too, not to mention being on some shitty PC for a month.
Please excuse the probable writing mistakes, this was a long post because I want to offer all the details necessary, in hopes of getting this solved.
Any assistance is greatly appreciated. Thank you!
PS: gonna go to sleep, then work, so I will be able to respond 12 hrs or so later.
Please list all hardware, PSU ect.
Are all bios setting stock, defaults or auto? Any cpu oc?
Download,
https://www.hwinfo.com/download.php
post a screenie sensor only. after running stress test.
Edge up Kaby lake guide is the best we have for now, I haven't seen in any posts much variance as far as safe voltages.
72c in prime blend is nothing to worry about, what are gaming temps like?
Have you tested each stick of ram individually?
You may want to turn off XMP, and input ram timings manually,
Um, you really DIDN'T bother to even read through his post, did you?
Let me address a few points. Kaby Lake guide is NOT the best there is for now. There are tons of Coffee lake overclocking tutorials.
72 degrees in Prime95 Small FFT with a major overclock is not a problem really, but in Blend mode, it is, because Blend mode is not a steady state stress/load utility and there are usually significant periods where the CPU is doing nothing at all in Blend mode, while waiting for RAM or virtual memory (Because Prime almost always overallocates memory services in Blend mode that are not available and need to be made up through virtual memory) services to complete, so it should definitely not be that high.
Plus, 72 degrees at stock settings would be MAJORLY unacceptable while under a steady state load as the thermal state should be nowhere near that kind of temp at a stock configuration. More like mid fifties to low sixties, maximum, if that. 4.6Ghz or higher full time OC, I'm ok with that temp. Stock 8700k configuration with a max ALL core turbo frequency of 4.3Ghz, I'm not liking that temperature at all and am looking to see why I'm lacking in the cooler department or whether I'm overboard on voltage.
He also clearly spelled out that he's definitely made manual adjustments to the memory timing, voltage and some of the sub settings.
He also linked to his entire configuration specs. Either read the post next time or pass on throwing out silly statements that are not based on the information provided by the OP.
Hey there, thanks for your responses.
A few points to note:
- The operating system was installed neatly. I removed the previous Windows partitions (Recovery/EFI System/etc.), formatted them, and then installed Windows on the new partition. I can't really format my other two partitions on the same disk, D and E. All my data remains there, though some isn't essential. I don’t have enough online storage to securely back everything up, so I’m hoping writing a few hundred GB of DVDs won’t become necessary.
- The disk is GPT-formatted. The OS is running in UEFI mode, and Safe Boot is enabled. Testing with them off a while ago didn’t show any significant changes.
- This is the final BIOS available, which includes the Meltdown/Spectre fixes. The issue is that earlier BSODs happened with the older version.
- Regarding temperatures:
Coffee Lake is extremely hot. My readings look consistent with what others report online, except for those who haven’t disassembled the CPU. That’s a risk I’m not willing to take, since it would void my warranty and the processor needs to last me a long time. The blend temperature is actually quite stressful, though not always so. Sometimes it appears to run AVX, other times it doesn’t, because my CPU fluctuates between 4300 MHz and as low as 3.7GHz. Power limits are consistently reached, and the CPU power consumption averages around 94.501W after eight hours of blending, which is essentially its TDP and what this CPU can achieve at default settings. Also, Blend puts a steady amount of pressure on the CPU, especially in the latest version of Prime. The CPU never drops below full utilization. I’m running a stress test for Blend to see if it makes a difference.
There’s no doubt I could have better temperatures, but please keep in mind I’m operating at stock settings and using a budget cooler that I barely managed to include. Will I consider buying a better one later? Yes, eventually, but first we need an SSD and a PSU in the next few months. Then another case, and possibly a larger Noctua cooler—we’ll find out then. For now, this cooler should be sufficient for keeping my stock 8700K running normally. Still, there’s no room for a bigger one at the moment.
http://www.zalman.com/contents/products/...tml?no=128
I might upgrade the CPU later, but it’s not worth it right now given these conditions. My main goal is to check if anything is faulty.
Understanding this is crucial, because this PC needs to last years. If something isn’t working properly, I should return it under warranty.
Blend was recommended only after a Small FFT test; memtest passed fine.
I ran memtest multiple times—from USB boot, all four runs were error-free. Three times I used default settings, and the fourth time I increased to 8 passes instead of the standard 4. No issues detected.
I also performed SmallFFT for about ten hours with no errors.
@toyo, darkbreeze provided great guidance, I recommend you follow it. If you're curious about why I mentioned running 26.6 and Small FFTs while still using blend, it's because we all in the overclocking community put in effort to assist others at our own cost. We hope those seeking help will take your advice seriously, share the results, and let us know if it works for you. Good luck if you do!
Darkbreeze :
Um, you really DIDN'T bother to even read through his post, did you?
Let me address a few points. Kaby Lake guide is NOT the best there is for now. There are tons of Coffee lake overclocking tutorials.
72 degrees in Prime95 Small FFT with a major overclock is not a problem really, but in Blend mode, it is, because Blend mode is not a steady state stress/load utility and there are usually significant periods where the CPU is doing nothing at all in Blend mode, while waiting for RAM or virtual memory (Because Prime almost always overallocates memory services in Blend mode that are not available and need to be made up through virtual memory) services to complete, so it should definitely not be that high.
Plus, 72 degrees at stock settings would be MAJORLY unacceptable while under a steady state load as the thermal state should be nowhere near that kind of temp at a stock configuration. More like mid fifties to low sixties, maximum, if that. 4.6Ghz or higher full time OC, I'm ok with that temp. Stock 8700k configuration with a max ALL core turbo frequency of 4.3Ghz, I'm not liking that temperature at all and am looking to see why I'm lacking in the cooler department or whether I'm overboard on voltage.
He also clearly spelled out that he's definitely made manual adjustments to the memory timing, voltage and some of the sub settings.
He also linked to his entire configuration specs. Either read the post next time or pass on throwing out silly statements that are not based on the information provided by the OP.
And maybe you should as well, where in the original post did OP say ANYTHING ABOUT OVERCLOCKING!
Did you see a PSU mentioned I sure as hell didn't. Or how about case, cooler. These were not listed in the link, Gee great one I guess those are not important.
Having said that adjustments were made dosen't tell us what they were, now does it.
You go on and on about about prime blend but seem to have missed all the other tests that op ran "Intel Burn Test, OCCT, Aida64, Smalll FFT Prime95 for 6-8 hours. These all came fine. IBT does give errors with lower SA/IO voltages."
Your arrogant. There are some great mods on this forum you are not one of them. Toms would be better of without you.
4Ryan6 :
@toyo, darkbreeze provided you with great guidance, I recommend you heed it!
I never doubted the value of the response or said I wouldn’t apply its advice. As I was already working on Blend, it made sense to let it complete so I could observe the outcomes. (12 hours went smoothly).
I still need to confirm a few points.
The advice suggests skipping AVX testing and sticking with an older Prime95 version. It also advises disregarding the errors I encountered during Blend with the newest Prime95. This approach would likely reduce CPU strain compared to using the latest Prime with AVX.
Is this accurate?
Because I find it hard to overlook errors in calculation software, if Prime95 doesn’t work—no matter the version or whether it supports AVX—I’d consider my PC unstable.
If any part of this reasoning is off, please let me know.
I included your message in the comment. Please let me know if you need anything else.
4Ryan6 added a note to the post reminding everyone to review it again. Thank you for the feedback. I plan to run SmallFFTs on the older Prime later, and I’ll share the results in about 10-12 hours tomorrow. I still don’t see any explanations about why I should overlook Blend errors. You can ask people for their input, and your advice is really valued, though it needs to be logical. Ignoring AVX and related issues isn’t logical from my perspective.