4 ms·

> The current entry level iPhone; iPhone SE 2020 with A13 is now faster in Single Threaded performance than ALL current shipping Android, including Flagship And

by antpls 7y ago

> The current entry level iPhone; iPhone SE 2020 with A13 is now faster in Single Threaded performance than ALL current shipping Android, including Flagship Android. I stopped to look at Single Thread performance long time ago when comparing phones. I now look for AI coprocessor performances, new sensor types and wireless charging performance

WoodenChair 7y ago
> I stopped to look at Single Thread performance long time ago when comparing phones. I now look for AI coprocessor performances, new sensor types and wireless charging performance The amount of daily time the average person spends doing tasks affected by single core performance versus the amount of time people spend doing AI related tasks or using exotic sensors makes you an outlier.
- antpls 7y ago
  Do you have any source showing global apps usage to support your claims? Just by guessing, I would say whatsapp/FB/instagram/Snapchat/youtube are the most browsed apps. All those apps use Neural Networks in the background, for text translation, automatic subtitles, picture and video filters, encoding, decoding, upscaling video, ... Neural networks will run constantly in the background to analyze phone sensors data and adjust the device energy usage, infer location, etc Even games will use them massively (see the RTX application from Nvdia) A performant single threaded CPU can probably run those AI applications fine, but an AI coprocessor (or neural processor or DSP or whatever the marketing named it) runs those applications by using 10 times less energy (easily) I recommend to everyone to take a look at http://ai-benchmark.com/ http://ai-benchmark.com/ . The improvement in this domain is massive, like an order magnitude bigger over the last 2 years, to the contrary of CPU and single threaded performance Edit : oh I also forgot mandatory usage of NN for facial recognition and pose recognition. Regarding sensor, the new radar in the Pixel cannot work without neural network : https://ai.googleblog.com/2020/03/soli-radar-based-perception-and.html?m=1 https://ai.googleblog.com/2020/03/soli-radar-based-perceptio...

by antpls 7y ago

WoodenChair 7y ago

> I stopped to look at Single Thread performance long time ago when comparing phones. I now look for AI coprocessor performances, new sensor types and wireless charging performance The amount of daily time the average person spends doing tasks affected by single core performance versus the amount of time people spend doing AI related tasks or using exotic sensors makes you an outlier.

antpls 7y ago
Do you have any source showing global apps usage to support your claims? Just by guessing, I would say whatsapp/FB/instagram/Snapchat/youtube are the most browsed apps. All those apps use Neural Networks in the background, for text translation, automatic subtitles, picture and video filters, encoding, decoding, upscaling video, ... Neural networks will run constantly in the background to analyze phone sensors data and adjust the device energy usage, infer location, etc Even games will use them massively (see the RTX application from Nvdia) A performant single threaded CPU can probably run those AI applications fine, but an AI coprocessor (or neural processor or DSP or whatever the marketing named it) runs those applications by using 10 times less energy (easily) I recommend to everyone to take a look at http://ai-benchmark.com/ http://ai-benchmark.com/ . The improvement in this domain is massive, like an order magnitude bigger over the last 2 years, to the contrary of CPU and single threaded performance Edit : oh I also forgot mandatory usage of NN for facial recognition and pose recognition. Regarding sensor, the new radar in the Pixel cannot work without neural network : https://ai.googleblog.com/2020/03/soli-radar-based-perception-and.html?m=1 https://ai.googleblog.com/2020/03/soli-radar-based-perceptio...