Time to see how well the new Kaby Lake generation responds to increased bandwidth. Since we didn't see any gains in Instructions Per Clock we assume it will scale in the same fashion memory wise as the Skylake series. Take note that the Integrated Memory Controller of the Kaby Lake series is capable of handling +4000Mhz easily.

To conduct our tests we are varying the memory speed from 2133MHz up to 3733MHz, the max our Gigabyte Z270X Gaming7 allowed us to do with BIOS F4g. Newer BIOSes or other boards might push our G.Skill B-Die kit close to the 4000mark, though that is not the concept behind this test, nor what this Gigabyte Aorus motherboard was designed for. The idea is to spot if a high end tweakable memory kit paired with the i7-7700K brings any significant performance advantages.

We maintained the memory timings at C15-15-15-35 2T between the below speeds:

• 2133MHz
• 2666MHz
• 3000MHz
• 3200MHz
• 3600MHz
• 3733MHz

Why keep the timings fixed you might ask? Otherwise it will be comparing apples with oranges.. Memory performance is always a mixture of the used memory ICs, their rated speed and timings. If we compare the popular Hynix MFR IC based 3200C15 memory with a B-Die based 3600C16 kit, there won't be much of a performance difference, even though that 3600 kit certainly looks faster on paper. By just adjusting the memory speed we exclude the timings variable; secondly by using the same memory ICs we also rule out any performance hits there.

The AIDA64 bandwidth tests shows the solid scaling when selecting a higher memory divider, especially going from 2133 to 3000MHz results in a big boost. From speeds beyond 3600 the performance increase becomes less and less prominent. We can spot similar behavior on the below benchmarks: take for example the Cinebench R15 benchmark below, which in general loves to be backed up by the memory and a high performance subsystem, so scales with any added bandwidth you can throw at it. Note the scaling has lost in most benchmarks its momentum with system memory speeds once over 3600MHz.

But not every application needs tons of bandwidth to perform. We are throwing the Blender benchmark into the benchmark suite, this to prepare for another important upcoming CPU release, so we get a feel for the way it renders the scene. A very minor performance improvement is observed, though nothing earth shattering. The same accounts for the two game benchmarks. Memory speed is not that important to gain any noticeable Frames Per Second in your favorite games. If you are a gamer look elsewhere, a decent main stream memory kit will be just fine, no need to cash out!