When Intel (NASDAQ:INTC) started talking more openly about its 14-nanometer technology and the first processor family based on that technology (known as Broadwell), the company said Broadwell delivered on the order of two times the performance per watt as Haswell (although I should caution that Intel was referring to the Core M product lines in this comparison, not necessarily all of the Broadwell chips).
Although Broadwell parts have been in the wild for some time, it has been hard to get a good read on the parts' power efficiency relative to Haswell parts due to confounding factors in the performance and power tests.
That said, with the launch of Broadwell parts for socketed desktops, review sites have been able to more directly compare the performance-per-watt of Haswell and Broadwell. So let's see if Intel's claims ultimately held up.
Not quite a doubling, but a respectable boost nonetheless
Website Legit Reviews, which tested the Broadwell-based i7-5775C, ran performance and power tests of Intel's Haswell and Broadwell parts while running HandBrake, which is described as a "multiplatform video transcoder." Video transcoding is a particularly CPU-heavy task, making it great for getting a sense of the performance-per-watt benefit that the Broadwell CPU and the 14-nanometer process bring.
According to the results, the 5775C delivered 0.449 units of performance per watt of power consumed. The prior-generation Haswell 4770K delivered 0.297 and the Haswell Refresh 4790K came in at 0.273. If we compare the 5775C to the 4770K (the more efficient of the high-end Haswell chips), the Broadwell chip is 51% more efficient than the prior-generation Haswell chip.
Intel says process technology improvements tend to deliver a 1.6 times boost in performance-per-watt. Given that Broadwell is more or less a slightly tweaked Haswell, the efficiency improvements here are likely mostly attributable to the 14-nanometer process.
This bodes well for Skylake
Later this year, Intel is expected to launch its next-generation processor family known as Skylake. Skylake will also be built on Intel's 14-nanometer technology; however, according to the Intel Developer Forum website, it has been "completely redesigned to bring new [intellectual properties] and integrations, great performance and reduced power consumption."
For desktops, where Intel didn't really bring Broadwell out broadly, the boost in going from 22-nanometer Haswell to second-generation 14-nanometer Skylake should be quite impressive.
In mobile, although CPU architecture rework should improve efficiency over 14-nanometer Broadwell, I think the real excitement is elsewhere on the chip. The graphics/media engine should be substantially improved, and Intel is apparently integrating many traditionally mobile-focused intellectual properties into the low-power variants of the chip.
In short, Broadwell has shown the 14-nanometer process can substantially improve efficiency over the 22-nanometer process. With Skylake, Intel is likely to take even greater advantage of the fundamental improvements the transistors bring with broad architectural improvements.
This is also a great thing for Intel's data center group
Early next year, Intel is expected to move its bread-and-butter Xeon E5 server processor line to Broadwell and the 14-nanometer manufacturing technology. Since performance-per-watt is so highly valued in data center applications (given that energy costs are significant portions of the cost to operate a data center), the move to 14-nanometer could lead to a substantial improvement in the value proposition of Intel's upcoming server chips.