I just finished stability testing over the long weekend of my recent i5-8600k build and believe I have settled on the sweet spot for speed, temps, voltage, and overall system responsiveness. This was my first build in almost 7 years, and probably the easiest overclocking situation yet. Here is the build:
Intel i5-8600k + Noctua NH-U9S cooler
Asrock Z370 Fatality K6 Gaming (Bios 1.30)
2 x 8GB GSkill Trident Z DDR4-3200 CL14
EVGA GTX 1070Ti Black Edition
500 GB Samsung 850 EVO SSD
EVGA Supernova 750W G3 PSU
Windows 10 Pro 64
Phanteks Enthoo Pro M Case
First off, for anyone looking for a great case, I highly endorse the Phanteks Enthoo Pro M. A pleasure to work in and among the best cable management options I've ever seen. Great airflow, clean and neat build.
My Goal
My target was a 4.8 - 5ghz, fully stable, 9+ hour Prime95 overclock with peak temps not to exceed the low 80s, without having to delid. The internet seems to be loaded with plenty of claims of people hitting 5ghz on these chips with relative ease. However, after going through the overclock process myself, I wonder how many people actually included a Prime 95 test (newer versions that use AVX) and were able to keep temps at acceptable levels for an extended run. These chips get hot!
My final overclock setting is 4.8ghz @ 1.275v with a 2 AVX offset. Not bad at all, IMO.
Bios Settings - Asrock Z370 Fatality Gaming K6
The Asrock Fatality K6 gaming mobo was fantastic to build with and the Bios is well thought out and configured. After flashing to the latest 1.3 version, loading initial setup defaults, loading XMP setting for the DRAM, Enabling Multi-core enhancement (so all cores turbo the same), setting boot priorties, installing Windows, and then playing with the PC for several days of general use, I began the overclock. Since this board comes with Optimized OC settings as part of the bios, I went to the "OC Tweaker" tab and loaded the template for "4.7ghz." This immediately does several things: it changes your Long and Short Duration power limits to 4095, changes Load Line Calibration to Level 1, sets the CPU Ratio to 47, changes CPU/Cache voltage to Fixed Mode, and presets a (very high!) 1.37vcore.
I immediately knew to lower vcore to 1.30v for the first go round. I also disabled Spread Spectrum, and lowered the AVX offset to 3 from 6.
Overclocking and Testing
After some brief stability testing @ 4.7ghz, it then became a trial and error process of maximizing clock speed while running the lowest possible vcore. It's probably also good that I unknowingly used an outdated version of Prime 95 without AVX while testing 4.7ghz @ 1.30v as the AVX version would have sent temps through the roof @ 1.30v!
With Prime version 29.4, I next tested 4.8ghz @ 1.29v, then 4.9ghz @ 1.29v, then lowered vcore to 1.285v, 1.28v, and 1.275v. While there were no Prime 95 errors, there was some wonky Windows 10 behavior at 1.28v and 1.275v.. such as the Edge browser not opening and an overall sluggish response. At 4.9ghz and 1.285v, Edge worked, but I noticed some system stuttering during 3D Mark testing that seemed not normal. I was also not comfortable with peak temps which reached 87 - 91C under small FFTs.
So I dropped back to 4.8ghz at the same 1.285v, but lowered the AVX offset to 2 from 3. Everything now was super smooth. I was able to web surf, play music, and open files during Prime 95 testing with no issues. The system felt peppy, and peak temps under Small FFTs reached 79 - 84C on the assorted 6 cores -- much better. The final step was raising the Uncore (Cache Ratio) to 44 from 40 and rerunning Prime 95, which created no issues. 4.8ghz @ 1.275v with a 2 AVX offset seems to be where this particular chip wants to be.
Some Benchmarks
4.9ghz @ 1.285v, 3 AVX offset, 44 uncore ratio
Prime 95 v29.4 Peak temps Small FFTs: 87 - 91C
Cinebench: 1212 CPU, 186.9 Open GL
3D Mark TimeSpy: 6875
3D Mark Firestrike: 16,791
4.8ghz @ 1.275v, 2 AVX offset, 44 uncore ratio
Prime 95 v29.4 Peak temps Small FFTs: 79 - 84C
Cinebench: 1196 CPU, 187.4 Open GL
3D Mark TimeSpy: 6862
3D Mark Firestrike: 16,934
CPU peak temps during Cinebench and 3DMark testing in both cases were 61 - 63C, and 62 - 65C, respectively. CPU temps also never got above 65C during 2 straight hours of Star Wars Battlefront 2 multiplayer @ 1440p, Ultra settings. For those wondering, my EVGA GTX 1070Ti is clocked at 1900 mhz core and 2003 mhz memory.
Conclusion
As can be seen, dropping back to 4.8ghz using 10 less millivolts lowered temps significantly while stabilizing the system. Cinebench and Timespy score differences are negligible while Firestrike actually improved at the lower setting. Gaming was smooth and enjoyable and CPU temps outside of the punishing Prime 95 test peak in the mid 60Cs which seems quite reasonable. I've read some comments on the internet that Prime 95 is not representative of real world use, and thus can be ignored in pursuit of the 5ghz goal and the higher voltage needed to achieve it. Well, I guess I'm old school then, and respectfully disagree. Stability and longevity are just as important, and besides, I'm too old and too busy these days to try and squeeze every last drop out of a CPU. I'll leave that task to my teenage son if he catches the bug someday...
Thanks for reading. Hopefully this review was helpful to some of you considering similar hardware for your build. Happy overclocking!
Intel i5-8600k + Noctua NH-U9S cooler
Asrock Z370 Fatality K6 Gaming (Bios 1.30)
2 x 8GB GSkill Trident Z DDR4-3200 CL14
EVGA GTX 1070Ti Black Edition
500 GB Samsung 850 EVO SSD
EVGA Supernova 750W G3 PSU
Windows 10 Pro 64
Phanteks Enthoo Pro M Case
First off, for anyone looking for a great case, I highly endorse the Phanteks Enthoo Pro M. A pleasure to work in and among the best cable management options I've ever seen. Great airflow, clean and neat build.
My Goal
My target was a 4.8 - 5ghz, fully stable, 9+ hour Prime95 overclock with peak temps not to exceed the low 80s, without having to delid. The internet seems to be loaded with plenty of claims of people hitting 5ghz on these chips with relative ease. However, after going through the overclock process myself, I wonder how many people actually included a Prime 95 test (newer versions that use AVX) and were able to keep temps at acceptable levels for an extended run. These chips get hot!
My final overclock setting is 4.8ghz @ 1.275v with a 2 AVX offset. Not bad at all, IMO.
Bios Settings - Asrock Z370 Fatality Gaming K6
The Asrock Fatality K6 gaming mobo was fantastic to build with and the Bios is well thought out and configured. After flashing to the latest 1.3 version, loading initial setup defaults, loading XMP setting for the DRAM, Enabling Multi-core enhancement (so all cores turbo the same), setting boot priorties, installing Windows, and then playing with the PC for several days of general use, I began the overclock. Since this board comes with Optimized OC settings as part of the bios, I went to the "OC Tweaker" tab and loaded the template for "4.7ghz." This immediately does several things: it changes your Long and Short Duration power limits to 4095, changes Load Line Calibration to Level 1, sets the CPU Ratio to 47, changes CPU/Cache voltage to Fixed Mode, and presets a (very high!) 1.37vcore.
I immediately knew to lower vcore to 1.30v for the first go round. I also disabled Spread Spectrum, and lowered the AVX offset to 3 from 6.
Overclocking and Testing
After some brief stability testing @ 4.7ghz, it then became a trial and error process of maximizing clock speed while running the lowest possible vcore. It's probably also good that I unknowingly used an outdated version of Prime 95 without AVX while testing 4.7ghz @ 1.30v as the AVX version would have sent temps through the roof @ 1.30v!
With Prime version 29.4, I next tested 4.8ghz @ 1.29v, then 4.9ghz @ 1.29v, then lowered vcore to 1.285v, 1.28v, and 1.275v. While there were no Prime 95 errors, there was some wonky Windows 10 behavior at 1.28v and 1.275v.. such as the Edge browser not opening and an overall sluggish response. At 4.9ghz and 1.285v, Edge worked, but I noticed some system stuttering during 3D Mark testing that seemed not normal. I was also not comfortable with peak temps which reached 87 - 91C under small FFTs.
So I dropped back to 4.8ghz at the same 1.285v, but lowered the AVX offset to 2 from 3. Everything now was super smooth. I was able to web surf, play music, and open files during Prime 95 testing with no issues. The system felt peppy, and peak temps under Small FFTs reached 79 - 84C on the assorted 6 cores -- much better. The final step was raising the Uncore (Cache Ratio) to 44 from 40 and rerunning Prime 95, which created no issues. 4.8ghz @ 1.275v with a 2 AVX offset seems to be where this particular chip wants to be.
Some Benchmarks
4.9ghz @ 1.285v, 3 AVX offset, 44 uncore ratio
Prime 95 v29.4 Peak temps Small FFTs: 87 - 91C
Cinebench: 1212 CPU, 186.9 Open GL
3D Mark TimeSpy: 6875
3D Mark Firestrike: 16,791
4.8ghz @ 1.275v, 2 AVX offset, 44 uncore ratio
Prime 95 v29.4 Peak temps Small FFTs: 79 - 84C
Cinebench: 1196 CPU, 187.4 Open GL
3D Mark TimeSpy: 6862
3D Mark Firestrike: 16,934
CPU peak temps during Cinebench and 3DMark testing in both cases were 61 - 63C, and 62 - 65C, respectively. CPU temps also never got above 65C during 2 straight hours of Star Wars Battlefront 2 multiplayer @ 1440p, Ultra settings. For those wondering, my EVGA GTX 1070Ti is clocked at 1900 mhz core and 2003 mhz memory.
Conclusion
As can be seen, dropping back to 4.8ghz using 10 less millivolts lowered temps significantly while stabilizing the system. Cinebench and Timespy score differences are negligible while Firestrike actually improved at the lower setting. Gaming was smooth and enjoyable and CPU temps outside of the punishing Prime 95 test peak in the mid 60Cs which seems quite reasonable. I've read some comments on the internet that Prime 95 is not representative of real world use, and thus can be ignored in pursuit of the 5ghz goal and the higher voltage needed to achieve it. Well, I guess I'm old school then, and respectfully disagree. Stability and longevity are just as important, and besides, I'm too old and too busy these days to try and squeeze every last drop out of a CPU. I'll leave that task to my teenage son if he catches the bug someday...
Thanks for reading. Hopefully this review was helpful to some of you considering similar hardware for your build. Happy overclocking!