We have recently published an extensive review on the Intel Core i7 5775C, the first Broadwell processor that is available of the shelf. What we have not covered in our previous review is the actual like-for-like performance gain that Broadwell has over the previous generation. We have tested this as well as the gain over earlier generations, covering a total of five generations of Intel processors running on the same clock frequency.
Intel implements their tick-tock strategy when introducing new processors, firstly introducing a new architecture (tock) and after that a new manufacturing process (tick). The latest iteration of this process is the Broadwell chip, which is a processor based on existing architecture but with a new generation of transistors, so a tick. These are normally known for their increase in efficiency and therefore more performance per watt. The processors based on new architecture (tocks) are usually the ones that show a large increase in performance overall, recent examples of these are the Sandy Bridge and Haswell.
It looks like the upcoming Skylake will be the last step in this strategy, as Intel has already announced that after the upcoming tock there will be two ticks in stead of just the one.
|Presler/Cedar Mill||Pentium 4 / D||Tick||65 nm||2006|
|Conroe/Merom||Core 2 Duo/Quad||Tock||65 nm||2006|
|Penryn||Core 2 Duo/Quad||Tick||45 nm||2007|
|Nehalem||Core i||Tock||45 nm||2008|
|Westmere||Core i||Tick||32 nm||2010|
|Sandy Bridge||Core i 2xxx||Tock||32 nm||2011|
|Ivy Bridge||Core i 3xxx||Tick||22 nm||2012|
|Haswell||Core i 4xxx||Tock||22 nm||2013|
|Broadwell||Core i 5xxx||Tick||14 nm||2014
|Skylake||Core i 6xxx||Tock||14 nm||2015|
|Kaby Lake||Core i 7xxx?||Refresh||10 nm||2016|
|Cannonlake||Core i 8xxx?||Tick||10 nm||2017|
It is usually not just the new architecture that is responsible for the increase in performance. Support for new instructions, higher clock speeds, improved Turbo algorythms and more are also a reason for the processors of a new genaration to be faster than their predecessors. This makes it difficult to see in benchmarks what part of the better performance can be directly attributed to the newer architecture and what is just the result of the evolution of the chip.
To find out how this has evolved over the last five generations of Intel processors we conducted a fast IPC test on all of them. We used a processor out of each generation, clocked them at exactly 3.0GHz, disabled the Turbo and HyperThreading settings and used 8 GB DDR3-1600 memory on 8-8-8-24 1T timings with all of them.
We have tested the below processors at the settings mentioned above:
We ran two benchmarks; Cinebench R15 and the Tech Arp x264 video-encoding benchmark. We ran both of these benchmarks multi-threaded (all cores active) as well as single-threaded (one core active).