I've finally settled down on my 6850K's overclock and wanted to post an update to the final settings along with a few things I learned. And I've also linked to my log on Google docs on the bottom of the post.
Components
- I7 6850K
- Asus Rampage V Edition 10
- Corsair Dominator Platinum DDR4 2800 64GB (4 x 16GB)
- EVGA GTX 780 Ti Classified Hydro-Copper
- Corsair AX1200i
- Samsung 950 Pro nvme SSD
- Samsung 850 Pro SSD
- Corsair 900D Case
- EK XE480 Quad Radiator w/ Corsair ML120 Fans
- EK Supremacy EVO CPU Block
- EK-XRES 140 Revo D5
Final settings on OC:
- CPU Multiplier: 44x
- Cache Multiplier: 36x
- AVX Offset: 0
- CPU Core: Adaptive @ 1.375v
- CPU Input Voltage (VCCIN): 1.92v (Loads at 1.9v) I think I can still fine tune this down.
- LLC: Level 7
- Cache Voltage: Offset +0.14 (Approx. 1.144v at load)
- System Agent Voltage (VCCSA): Offset +0.1 (Approx 1.008v at load)
- VCCIO CPU: 1.1 in bios, but shows 1.056v in software readings.
- DDR Frequency: 3000 (overclocked from 2800)
- DDR Timings: 13-15-15-36 1T (tightened from stock 14-16-16-36 2T)
- DDR: 1.365v
- All other voltages manually set to their stock values
On my specific system, the following are some of the things I've learned after exhausting testing and fine-tuning.
If you're looking for ABSOLUTE system stability, you need to account for combined CPU + GPU loads.I can run the above settings with lower VCORE (1.35), VCCSA (0 offset), VCCIO (stock) if I'm only isolating the stress tests to CPU.
I realized that sometimes OCCT will fail, or crash the system while other times it will run for more than 8 hours without any issues. It would most likely pass when ran overnight while I'm asleep, but other times when I have it running while I'm browsing the web, or watching some videos, it would fail/crash. After some trial and error, I found that a combination of CPU load (OCCT) + GPU usage (watching a video or even using chrome , due to hardware GPU acceleration) would cause my completely stable CPU isolated voltages to become unstable.
Although you'll likely never load the system enough for this to matter, a certain application or game may cause enough load on both components to make your overclock unstable.
Avoid AUTO values when possibleTry to set your voltages to at least their stock values manually. This is especially true for VCCSA, and Cache Voltage, as AUTO values may set these to unreasonable levels. For example, with XMP enabled and VCCSA on auto, my motherboard would set the VCCSA to over 1.3 volts.
System Agent Voltage (VCCSA) is a pain in the ass. Due to the fact that too high of a voltage can cause instability as well, it may be quite annoying to fine-tune this voltage.
For CPU isolated tests, VCCSA didn't seem to affect my overclocks much. But with combined CPU+GPU loads, this voltage along with VCCIO CPU had to be
completely fine-tuned in order for complete stability. Even a small deviation in either of the voltages caused sporadic instability (OCCT passing for 10 hours, then suddenly failing at 2 hours or less on a separate run at same settings)
Variation in test software is necessary for ABSOLUTE stability.I found that while at a certain setting, I was able to pass 8 hours of Realbench, 1000% coverage in HCI memtest, but crash during Google Stressful Application Test (GSAT). I had the assumption that 1000% coverage was the golden standard for memory and cache, but realized that I had to bump up my cache voltage a bit in order to be stable for GSAT.
For absolute stability, I find these combination of tests are ideal.
- OCCT for approximately an hour for a quick initial test while tuning the multipliers and voltages.
- HCI Memtest for approximately an hour.
- Realbench for an hour.
-
Once you're closer to your ideal settings and want to achieve absolute stability:
- 8+ hour run of OCCT with a video playing on your preferred media player in the background along with some browser tabs open.
- 1000% coverage of HCI Memtest
- 8 hour run of GSAT (stressapptest -W -s 28800) Don't forget the -W, which uses more CPU-stressful memory copy.
- 8 hour run of Realbench
- Lastly another run of OCCT for a couple of hours just to seal the deal.
While running any stress test, make sure to check your event viewer > windows logs > system for any WHEA (especially Cache hierarchy) warnings or errors. I found that sometimes I can pass the tests initially, but fail on concurrent runs and find those errors on my log. With some bumps in cache or core voltage, it goes away and I would get closer to stability.
Oh, I've uploaded my log to google docs, where you can
view here. Although each system is unique and will require its own set of tweaks, maybe it will give you a point of reference for your overclock. It's quite messy as it was just for my own reference, but you get the point.