I posted this originally in a different thread, but I think it deserves its own thread. [EDIT] See this post for results of the m3, i5, and i7, with the m3 tested on two different table tops: https://forums.macrumors.com/thread...stained-cpu-load.2073415/page-2#post-25271927 Summary: All machines pretty close to each other in performance, but the i7 is indeed the fastest and the m3 on a wood table is the slowest. However, the m3 on a granite table did better, presumably because the granite wicked away heat better. [/EDIT] I ran Cinebench R15.038 twenty-five times in a row on my 2017 MacBook Core m3-7Y32. This took about 65 minutes. So essentially it's over an hour of sustained, load aside from the second or so pause between runs. The numbers below are the Cinebench 64-bit multi-core CPU scores, and the speed of the first run is roughly what one would get from an early 2015 Core i5-5250U MacBook Air. Ambient room temperature was 23°C. The table top was a granite breakfast bar. The MacBook was running off battery power, with screen dimming turned off. WiFi and Bluetooth were turned off, and Sophos anti-virus was deactivated. Out of all these things, it seems only the anti-virus has a real impact on performance. OS is macOS 10.13 High Sierra. 16 GB RAM and 256 GB SSD. Note that part way through the 11th run, I moved the computer over a bit because I was rearranging stuff on the counter. Low and behold, the score went up, and continued to go up the next run too. I noticed the part of the table where the computer had been was a little bit warm (but not hot), so it seems moving the computer over to a cooler spot helped performance. So, right after run 22, I intentionally did the same thing again, and again the score went up. Below are the raw scores for future reference in case someone wants to put it into an Excel spreadsheet, but the most valid scores are the first 10 (and then from 12-22) for the reasons stated above: 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 You'll note that not even once did the score drop below 250. It may have dropped a bit below 250 had I not moved the computer, but extrapolating that would mean it would taken about half an hour of sustained load to go below 250, which is only 5-6% less than max speed. (Given that 25 runs is about 65 minutes, that means 10 runs is around 25 minutes of sustained load, and 12 runs is about half an hour. I think it would have taken 12 runs to dip below a score of 250.) More importantly, compare this to Notebookcheck's 2017 MacBook Core m3 results. I didn't get anything like their weird results: Perhaps they got a lesser m3, or else it was hotter in their room, or else their table retained more heat. Their m3 seemed to periodically throttle really hard after sustained high workload, but it's curious because it's very different behaviour from their 2016 models (below). Furthermore, the 2017 Core m3 beats all of the 2016 m3, m5, and m7 MacBooks from Notebookcheck's 2016 m3/m5/m7 MacBook face-off: This 2017 m3 also beats Eldar Gezalov's 2017 Core i7 MacBook under the same test: I'm not sure why Eldar's Core i7 is slower than my Core m3. Either it is throttling more or else he has more background processes running. If it's throttling more, then it's a repeat of Notebookcheck's 2016 results, where the m7 ran slower than the m5.