Kaby Lake Core m3-7Y32 MacBook does pretty well under sustained load

[Eug]

Notebookcheck had done an article on the 2017 Core m3 MacBook and got strange results in their repeated Cinebench R15 test, where they run Cinebench repeatedly for an hour. Scores in their test gradually decreased over time but then around the 15 minute mark on the 8th run the score dropped like a rock - CPU throttled hard.

https://www.notebookcheck.net/Apple-MacBook-12-2017-Laptop-Review.230656.0.html

I wasn't too happy seeing that behaviour, so I decided to test this out myself. I did 25 sequential runs of Cinebench over the course of 65 minutes or so. Ultimate geekery I know.

I am happy to report that my results didn't follow Notebook check's at all. Here are my results:

264
261
259
258
258
258
257
255
255
253
254
257
257
257
256
255
255
254
252
252
251
250
254
258
258

Note though that only the first 10 runs are truly accurate IMO. Part of the way through the 11th run I had moved the computer over a little bit and all of a sudden the scores improved again for several runs. I attribute that to the fact that new spot on the table was cooler. To confirm this, after the 22nd run I moved the computer again and yep, scores went back up again.

Extrapolating from the curve, it would have taken about 12 runs to get below 250, from the initial first score of 264. That means it would have taken about half an hour of sustained load to drop around 5-6% in performance, in this fanless computer, which by the way has a TDP-up config. (Base clock of this chip is usually 1.1 GHz, but it's 1.2 GHz in the MacBook.)

It also scored and behaved much better than all of the m3, m5, and m7 MacBooks from 2016 in Notebookcheck's face-off:

https://www.notebookcheck.net/Face-Off-Apple-MacBook-12-Core-m3-Core-m5-and-Core-m7.172046.0.html

BTW, in their 2016 test, the m7 does worse than the m5.

Needless to say, I am pleased with the fanless Kaby Lake Core m3.

 

[Bouowmx]

Notebookcheck also tested Intel Kaby Lake-R using this Cinebench R15 performance over time. This test is useless to me; I just want to know one number: sustained frequency because of power limit.

 

[Eug]

Notebookcheck also tested Intel Kaby Lake-R using this Cinebench R15 performance over time. This test is useless to me; I just want to know one number: sustained frequency because of power limit.

Intel Power Gadget reports this CPU throttles at 2.5-2.6 GHz with YES testing. However, on the Mac, the application hasn’t been updated since 2016 and technically doesn’t officially support Kaby Lake so I can’t be completely sure if what it’s reporting is accurate. It may be accurate though since IIRC it will max out in some usage for a split second higher than 2.6 GHz but under 3 GHz. (Its Turbo Boost max spec is 3.0 GHz so that does make sense.)

 

[Eug]

Someone was good enough to post their i7 results, so I've compared them in a graph. Please note that only the first 10 runs are accurate. On run 11 for the m3, I moved the computer to a different spot on the counter, and that must have cooled it a bit as the performance went up. Also, for the i7, the owner got a phone call during run 11, so the score went way down.

But overall, the i7's advantage over the m3 isn't large, and the performance curves over time seem to be converging, presumably as residual heat increases. IOW, it seems they throttle to similar speeds.

Here are the scores with the graph zoomed in a bit to emphasize the differences (which aren't large).

This convergence makse sense since with 4-thread yes testing, both the i7 and m3 will throttle quickly to 2.5 to 2.6 GHz (assuming Intel Power Gadget is reporting things correctly).



P.S. Here is all of my data graphed:

As mentioned, at run 11, I moved the computer to a different spot on the table, and performance improved, likely due to the fact that part of the table hadn't been heated up by the computer. To confirm this, right after run 22, I moved it again intentionally to a cooler part of the table, and again, speed improved.