3 ms·

The only metric in which the A9 is really better than the 820 is single core performance. Multi-core and GPU metrics are about the same.

acdha 10y ago
Single core performance is still the single most important metric for most users, especially since things like web browsers tend bottleneck on single thread performance and browsing is one of the most common things people use phones for, especially given how commonly native apps use embedded web views. Measuring browser performance is hard because there's a lot of different functionality but notice how routinely we see reviews where the device the Android flagship phones are trying to beat is the iPhone 6 or even 2013's 5S: https://cdn.arstechnica.net/wp-content/uploads/2016/09/iPhone-7-redo.002-980x720.png https://cdn.arstechnica.net/wp-content/uploads/2016/09/iPhon... https://cdn.arstechnica.net/wp-content/uploads/2016/09/iPhone-7-redo.001-980x720.png https://cdn.arstechnica.net/wp-content/uploads/2016/09/iPhon... (from https://arstechnica.com/apple/2016/09/iphone-7-and-7-plus-review-great-annual-upgrades-with-one-major-catch/#h14 https://arstechnica.com/apple/2016/09/iphone-7-and-7-plus-re...) Some of that reflects the considerable amount of work which Apple has put into Mobile Safari but a lot of that is going to come down to single thread performance.
- chetanahuja 10y ago
  "Single core performance is still the single most important metric for most users," This is an empty statement backed by no data. The only evidence you provide to the superiority of single core perf is browser performance. Except, 1) browser usage is now a mostly insignificant part of time spent on mobile (http://flurrymobile.tumblr.com/post/127638842745/seven-years-into-the-mobile-revolution-content-is http://flurrymobile.tumblr.com/post/127638842745/seven-years... ) and 2) The bottleneck in mobile apps which depend on the network (including browsing the web) is rarely the CPU, it's the network.
  - acdha 10y ago
    That's why I mentioned embedded web views: a lot of time recorded as native apps involves embedded web-views and these days that means things like JavaScript or layout performance matter more than might be immediately obvious. On iOS, the least involved way to see this is in things like news apps where Safari content blockers also block advertisements in the app. The other side of this is that we're not really talking about which app has the most time in the foreground so much as which app causes the user to wait the most. Much of that time will be network I/O which is a real challenge but also not relevant to this discussion about CPU performance. Fundamentally, all I'm trying to say is that Amdahl's law still applies until we're at the point where the user is never waiting on computation. Developers have been getting better at multithreading but uneven CPU usage is still common enough that I'd favor fewer faster cores over more slower cores.
x0x0 10y ago
which is the most important metric. see, eg, industry-wide struggles to effectively write parallel software.

acdha 10y ago

Single core performance is still the single most important metric for most users, especially since things like web browsers tend bottleneck on single thread performance and browsing is one of the most common things people use phones for, especially given how commonly native apps use embedded web views. Measuring browser performance is hard because there's a lot of different functionality but notice how routinely we see reviews where the device the Android flagship phones are trying to beat is the iPhone 6 or even 2013's 5S: https://cdn.arstechnica.net/wp-content/uploads/2016/09/iPhone-7-redo.002-980x720.png https://cdn.arstechnica.net/wp-content/uploads/2016/09/iPhon... https://cdn.arstechnica.net/wp-content/uploads/2016/09/iPhone-7-redo.001-980x720.png https://cdn.arstechnica.net/wp-content/uploads/2016/09/iPhon... (from https://arstechnica.com/apple/2016/09/iphone-7-and-7-plus-review-great-annual-upgrades-with-one-major-catch/#h14 https://arstechnica.com/apple/2016/09/iphone-7-and-7-plus-re...) Some of that reflects the considerable amount of work which Apple has put into Mobile Safari but a lot of that is going to come down to single thread performance.

chetanahuja 10y ago
"Single core performance is still the single most important metric for most users," This is an empty statement backed by no data. The only evidence you provide to the superiority of single core perf is browser performance. Except, 1) browser usage is now a mostly insignificant part of time spent on mobile (http://flurrymobile.tumblr.com/post/127638842745/seven-years-into-the-mobile-revolution-content-is http://flurrymobile.tumblr.com/post/127638842745/seven-years... ) and 2) The bottleneck in mobile apps which depend on the network (including browsing the web) is rarely the CPU, it's the network.
- acdha 10y ago
  That's why I mentioned embedded web views: a lot of time recorded as native apps involves embedded web-views and these days that means things like JavaScript or layout performance matter more than might be immediately obvious. On iOS, the least involved way to see this is in things like news apps where Safari content blockers also block advertisements in the app. The other side of this is that we're not really talking about which app has the most time in the foreground so much as which app causes the user to wait the most. Much of that time will be network I/O which is a real challenge but also not relevant to this discussion about CPU performance. Fundamentally, all I'm trying to say is that Amdahl's law still applies until we're at the point where the user is never waiting on computation. Developers have been getting better at multithreading but uneven CPU usage is still common enough that I'd favor fewer faster cores over more slower cores.

x0x0 10y ago

which is the most important metric. see, eg, industry-wide struggles to effectively write parallel software.