Intel’s Coffee Lake CPUs aren’t a hard “response” to Ryzen; the CPUs have been taped-out for a while now, but the response appears to align more with the release timeline and pricing. X299 moved forward to compete with Threadripper, and Coffee Lake received a similar treatment. One thing we won’t know, of course, is whether the pricing is some sort of a response to Ryzen. Intel’s i7-8700K premiere CL CPU carries an MSRP target of $360 1ku, a $10-$20 jump over Kaby Lake 7700K CPUs with fewer cores and similar frequencies. Either way, the i7-8700K is here now, and we’ve got a densely packed review covering most aspects of Coffee Lake.
Our Intel i7-8700K review will focus on delidding, liquid metal application, overclocking, gaming & streaming benchmarks vs. Ryzen, power draw, and production benchmarks. Our i5-8600K review will post separately, as this is dense enough as-is.
Deprecating Old Tests
As a foreword, we’ll be deprecating our old CPU benchmarks going forward. Coffee Lake marks a refresh of our CPU test suite, though we are still carrying-forward some now-retired tests for comparative/archival data. These tests will be slowly phased-out as we accumulate more data for the new benchmarks. The idea is to conduct both simultaneously, with the older option acting somewhat like an LTS build as it is gradually retired.
The new testing involves a new suite of games, new testing methods (including 1440p tests), delidding & LM testing (debuted in our X299 coverage), and new production render testing.
Overclocking & Reject Sample
It seems that we received one of the worst samples of the media lot, when it comes to i7-8700K overclocking. We ended up stuck at 4.9-5.0GHz with a Vcore of 1.40-1.42, a far cry from Der8auer’s golden sample unit at 5.2GHz and 1.375Vcore:
We struggled to hold 5.0GHz stable in some production tests (like Blender), and required high voltage to hold 4.9GHz. We haven’t re-attempted overclocking with liquid metal yet, but did delid the CPU with Rockit Cool’s delid kit and used Conductonaut liquid metal.
Some of this comes down to the motherboard. Gigabyte's Ultra Gaming Z370 board seems to output lower voltage than we ask for in BIOS, which makes it difficult to gauge what's actually going on versus what you're telling the board to do.
All of the new (non-legacy) CPU tests were conducted with 3200MHz CL16 memory. A GeIL EVO X kit was used across the board, with some exceptions made for a GSkill RGB Trident Z kit that had been reconfigured to match the GeIL timings. We’ve also moved to a 1080 Ti FTW3 from the 1080 FTW1, making for a stark difference in GPU bottlenecking headroom between the legacy and new tests. Part of this means updating drivers to version 385.69, with Windows now updated to Creator’s Update (game mode disabled) for new CPU benchmarks and reviews. An EVGA SuperNOVA T2 1600W PSU was used as the primary power supply, with a Kraken X62 (max pump + fan speeds) used for the cooler on all benches.
Previous platform and test methodology information can be found in more recent CPU reviews, but we are updating in big ways for CL onward. We’ll detail the test method updates in a separate content piece in the future, as this content is presently being written while the writer is falling asleep for microseconds, hours ahead of embargo.
To answer a few common questions this morning:
- All tests were run BEFORE the delid (stock conditions) except for the liquid metal tests, obviously.
- The 1700 was used for comparison because it is the 1700X & 1800X. They're all the same. Just overclock one to 4GHz and you've now represented all Ryzen 7 CPUs at 4GHz.
Let’s get some quick synthetics out of the way, then move on to the real testing:
NOTE: We observed issues with Gigabyte's F2 EFI revision that forced some tests to run in MCE mode. Fortunately, these do not impact our gaming results, as we used F4 and F5a for those (with MCE disabled by default). The synthetic benchmarks have been mostly recalculated to reflect this issue of Gigabyte's BIOS.