Intel Kaby Lake Performance: Surprising Jump over Skylake

Battery Life, HEVC Playback, Conclusions

Battery Life Testing

Mobile performance is always a balance between speed and efficiency. High performance in an application or workload is great but users that want a notebook to last all day will find battery life just as important (if not more so).

Our test runs through a selection of pre-determined websites on Wi-Fi and circulates every 30 seconds and scrolling down the pages. Brightness is set to as close to 180 lux as possible (always falls within 175-185 lux) to account for different screens – simply setting all machines to 50% just doesn’t cut it. That is a little higher on the brightness than some other websites use but I find that 180 lux is the brightness level that is most comfortable to me working in a standard lit office space.

Also, we tend to cycle websites faster than most other scripted tests that I find other media using. Every 30 seconds isn’t a setting that shows off super long battery life times but, again, I find it to be more realistic in the long run for real-world consumer usage models. What is important in our battery life tests is to compare the results to each other for relative differences rather than fixed time values (questioning why results don’t match up with other sites or vendors’ claims).

Also, we use Chrome. Sorry tech-world, we are just being real here.

I honestly wasn’t expecting a result in favor of the Kaby Lake machine by this much. At 10.6 hours of useable web browsing, the newer HP Spectre x360 goes nearly 2 hours longer than last year’s model, an improvement of 22%! (Note: yes, this is a relatively "new" unit of last year's model. Should be no battery wear concerns.)

I should note here that the newer HP Spectre x360 has a 57.9 WHr battery while the previous generation has a 56 WHr battery. Obviously that difference (3%) is not what gives the Kaby Lake system the edge.

Based only on battery size and battery life, we can then calculate the power consumption of the entire system during our battery tests. Lower in this case is obviously better as it points to a more efficient design, including processor selection, screen selection, storage, etc. Using this data the Kaby Lake system is using effectively 1 watt less power to get the same browsing workload done when compared to Skylake. 1 watt isn’t much in terms of power consumption but it clearly makes a difference when you are trying stretch the usable time for a notebook.

HEVC/H.265 Video Decode Performance

The primary differentiation point for the graphics and media block of Kaby Lake is the inclusion of an HEVC hardware encode/decode block. This takes the work off of the CPU cores and puts it on dedicated hardware, offloading the CPU and lowering power consumption and utilization.

How much do you ask? To test, I simply recorded some performance monitors from Windows 10 and played back a 4K 24FPS 20mbps 10-bit HEVC (H.265) video file in the Windows Movies & TV app. The result speaks for itself.

Wow! While the new Kaby Lake system was able to playback the movie using less than 10% of the CPU, the Skylake system required ~45% to get the job done. And to be honest, there were some places where I saw dropped frames on Skylake as well – not a perfect viewing experience. It has been a while since we have seen simple video playback hit CPU utilization levels like this.

As you might imagine, this kind of CPU utilization gap will translate to a HUGE amount of battery life advantage for Kaby Lake notebooks while watching similar video. Even upcoming VP9 content being pushed by Google will see this kind of edge on Kaby Lake systems with the dedicated decode block, another capability missing from previous architectures.

Our sample footage, Tears of Steel

Even though HEVC content is still in its infancy, its growing rapidly and the adoption by Google and other third parties will likely accelerate it in 2017.

Closing Thoughts

What started a simple pet project to get comparable Kaby Lake and Skylake systems in house quickly became a riveting dive into the performance advantages of Intel’s newest architecture. I did not expect the gap between these two most recently platforms to be as wide as it was or as ubiquitous through my testing.

For example, looking at something like H.264 video encoding, I was fully expecting a ~5% performance advantage that was in line with the base clock improvements from the specifications we showed you on an earlier page. Instead, we saw well over double digit gains in some areas, proving that clock speed, thermal improvements and technologies like the updated Speed Shift were combining into a “better than the sum of its parts” result. Other noted surprises were the gaming benefits (over 30% faster in Overwatch) and an 18% edge on POV-Ray, a heavily multi-threaded rendering engine.

Additional Kaby Lake advantages in SYSmark and WebXPRT make a bit more sense – with higher clock speeds and a slightly better implementation of Speed Shift getting the CPU to those higher speeds faster, workloads that have a burst pattern are going to be quicker. Application load times, Excel macro runs, quick photo edits; these are all examples of work that should be and are faster on Kaby Lake than on Skylake.

Couple the above with the potential for noticeable and significant battery life gains for mobile devices and I think that the Kaby Lake platform that many of us wrote off as a push by Intel to simply extend the life of current product, should get a second look. Though the potential for Kaby Lake-S parts for desktop consumers is still a question for next year, notebooks using Kaby Lake today can clearly be a step above the previous generation.

Expect more reviews of notebooks using Kaby Lake from PC Perspective soon, including a dive into the new HP Spectre x360 used in this story.