In case you missed it, we spent four hours live overclocking an Intel i7-8086K just a couple days ago. The OC effort was watched by about 2300 people concurrently, spanning all four hours, and was one of our most successful streams to-date. The viewership was beaten only, and unsurprisingly, by our #RIPLTT stream’s 5000 concurrent viewers.
As for the testing, it was all 8086K overclocking in Firestrike Physics, with some additional memory overclocking in the final two hours. Components used were varied, depending on what was happening at any given time, and the final frequency was high. We closed at 5.35GHz, running a 101 BCLK with 53x all-core multiplier. Some additional testing was done in effort to push individual cores to 54x, but we couldn’t get it stable. Despite our ultimate core limitations at just under 5.4GHz, the CPU itself – barring the IMC – is the best-binned 8700K we’ve had hands-on with yet. Our 8086K (which is a binned 8700K) managed to hold 5.1GHz at roughly 1.3V with relative stability in Firestrike, only running into exponential increases in voltage requirement upon pushing 53x multipliers. We even attempted 1.5V for a 5.4GHz overclock, but just couldn’t stabilize. Our plan is to return in the future with a bigger or more exotic cooling solution atop the die. Our X62 did admirably, and the delid with liquid metal (Thermal Grizzly Conductonaut) kept thermals in check, but lower is still better.
The volt-frequency curve was impressive overall, allowing significantly lower voltages at a given clock than either of our 8700K CPUs. Our 8700Ks typically run 4.9-5.0GHz at about 1.4V (in the best case) or 1.42-1.45V (in the case of the potato chip 8700K). This CPU managed 5.0GHz at 1.3V, a significant power and thermal reduction.
Other parts used include the ASUS ROG Maximus X motherboard, Corsair AX1600i PSU, and an EVGA GTX 1080 Ti FTW3. We used three kits of RAM: Corsair Vengeance LPX 3200 CL16, GSkill Trident Z Black 3600, and Corsair Vengeance 4600 CL19. We already knew that the Trident Z kit could hold at least 4000MHz, as it was our kit selected for the previous #RIPLTT stream. We also suspected the Vengeance 4600 kit could hold 3800-4000 with tight timings.
Unfortunately, as we continued swapping kits, it became evident that we had an IMC limitation on the CPU. This is a matter of binning – we got a good core, but our IMC seemed troubled, even when pushing voltages that we wouldn’t recommend for 24/7 use. The IMC limitation stuck us at around 3600MHz max, regardless of kit, though we were able to trim down to CL14 in some tests. Other timings tightened included:
- TRRD_S: 4
- TRRD_L: 5
- TFAW: 16
- TCWL: 14
- TREFI: 65000
TRFC: 240 - TRAS: 28 (lowest possible on Coffee Lake)
In the future, we will likely also tune maxmem in Windows. We configured IO and SA voltages to 1.3V, with vDIMM between 1.5 and 1.85, depending on the test. 1.85 isn’t recommended for long-term use, but is fine for an overclock benchmark. The memory can take it without issue, but the IMC is another story.
Our goal was 24,000 points, which proved just out of reach. We settled at 23,745, which was good enough for us.
The stream was tremendous fun, and we’d like to thank everybody who joined. We love the fan interaction during the stream, and it’s always fun to show the real results of overclocking – often bluescreens, with occasional rewards that make it worthwhile.
Find the stream archive above.
Editorial: Steve Burke
Video: Andrew Coleman