The Mobile CPU Core-Count Debate: Analyzing The Real World
by Andrei Frumusanu on September 1, 2015 8:00 AM EST- Posted in
- Smartphones
- CPUs
- Mobile
- SoCs
Modern Combat 5 Playing
The little cluster looks to behave extremely similar to what we saw in Real Racing 3: Three larger threads keep 3 of the cluster's CPU at relatively busy duty-cycles while we see some limited activity on the 4th core.
The big cluster also seems to behave in a similar fashion. One big main thread causes the bulk of the load while we only have occasional small bursts when threads get migrated onto the big cluster. This time we see a more variable load both in terms of requency and rq-depth instead of the flat-line that could be observed in Real Racing 3.
One interesting behaviour caught in this log was how the main big thread got moved around from CPU6 to CPU4 and then again to CPU5 on the 33s mark in the log.
Even though the total rq-depth might be a bit misleading here while it's showing an average of around 2.5, we can see that in the individual per-CPU runqueues we have 4 major threads at work. Again this is a case of using parallelization for the sake of power efficiency instead of performance. The 3 smaller threads on the little cores could have well been handled by a single larger CPU at higher frequency, but it wouldn't have been nearly as power efficient as spreading them onto the smaller cores.
157 Comments
View All Comments
lilmoe - Tuesday, September 1, 2015 - link
"if the interest is high enough":/ Really?
zaza - Saturday, September 5, 2015 - link
Yes Please. It would be nice to see if the same or similar tests works on snaprdagon 810,801 and 615 and Mediatek chips, and intel SoCerchni - Thursday, September 17, 2015 - link
A follow-up with synthetic would be quite interesting.aryonoco - Saturday, September 5, 2015 - link
I just wanted to reiterate the point here an thank the author for this great piece of technical investigative journalism.Andrei, thank you for this work. It is hugely invaluable and insightful.
tipoo - Tuesday, September 1, 2015 - link
Very interesting article. Seems like the mantra of "more cores on mobile are just marketing" was wrong in terms of Android, seems to dip into both four core big and little clusters pretty well. That puts the single thread performance having lagged behind the Apple A series (up until the S6 at least) in a new light, since it can in fact use the full multicore performance.tipoo - Tuesday, September 1, 2015 - link
*That is, barring gaming. More core Android functions do well with multithreading though.jjj - Tuesday, September 1, 2015 - link
In gaming there is a big advantage. By using mostly the small cores you allow for more TDP to go to the GPU. One more relevant thing would console ports in the next couple of years when mobile GPUs will catch up with consoles. The current consoles have 8 small cores and that fits just right with many small cores in Android.retrospooty - Tuesday, September 1, 2015 - link
Not really sure whos "mantra" that was. People that don't understand what the big.little architecture is like some angry Apple fans?tipoo - Tuesday, September 1, 2015 - link
Well sure, whoever they were, but it was a pretty common refrain for every 8 core SoC.soccerballtux - Tuesday, September 1, 2015 - link
for one, it was my mantra. I liked having 4 cores because 2 wasn't enough, but according to my hotplugging times, I only really need 3 for optimal experience most of the time