WEBVTT 00:00:00.000 --> 00:00:03.280 After many years of the same handful of companies 00:00:03.280 --> 00:00:08.600 making the processors in our computers and phones, we're finally seeing some major changes 00:00:08.600 --> 00:00:12.200 in rapid succession. Amazon rolled out servers based on their own 00:00:12.200 --> 00:00:16.080 Graviton processors. Apple, of course, announced their groundbreaking 00:00:16.080 --> 00:00:20.240 Apple Silicon for Macs and iDevices designed entirely in-house. 00:00:20.240 --> 00:00:24.720 And now Google has entered the fray with its new Tensor processors. 00:00:24.720 --> 00:00:29.360 And although it looks like they'll be manufactured by Samsung, Google actually had a major role 00:00:29.360 --> 00:00:32.520 in designing the chip. They had their fingers in there. 00:00:32.520 --> 00:00:37.560 But unlike Google's Tensor Processing Unit, which is an AI-focused card designed for servers, 00:00:37.560 --> 00:00:41.720 Tensor is a mobile-focused system on a chip, or SOC, 00:00:41.720 --> 00:00:47.160 meaning it combines a CPU and other computing elements, such as RAM, a modem, and audio processing, 00:00:47.160 --> 00:00:52.920 together for better efficiency. Chips like these are the mainstays in modern smartphones, 00:00:52.920 --> 00:00:57.960 but only a few companies design their own. Apple, with their A-Series, is one example, 00:00:58.000 --> 00:01:03.640 while the majority of Android-powered gadgets use off-the-shelf SOCs from MediaTek or Qualcomm, 00:01:03.640 --> 00:01:07.600 with Samsung's in-house Exynos chips being a notable exception. 00:01:07.600 --> 00:01:11.640 Even Google's own Pixel lineup has used Qualcomm chips until now, 00:01:11.640 --> 00:01:16.560 but the end is near. And going forward, they'll be featuring Tensor chips. 00:01:16.560 --> 00:01:22.280 But what is a Tensor chip and why exactly is Google moving chip design in-house 00:01:22.280 --> 00:01:26.240 when it might be easier just to outsource it? Well, when there's an app for that 00:01:26.280 --> 00:01:29.480 it was a popular catchphrase 10 years ago. 00:01:29.480 --> 00:01:32.640 Oh no. I don't think most of the people using it anticipated 00:01:32.640 --> 00:01:37.500 how much we would come to expect from our phones. Everything from DSLR look-alike photos 00:01:37.500 --> 00:01:42.080 to high-fidelity gaming to instant AR-powered translation. 00:01:42.080 --> 00:01:46.080 Is this the future? But expect all of these things we do. 00:01:46.080 --> 00:01:50.040 And as the software that enables these features becomes more and more complex, 00:01:50.040 --> 00:01:53.600 some of the familiar hardware inside our phones hasn't kept up. 00:01:53.600 --> 00:01:59.200 So instead of using more general-purpose processors from an outside company, Google developed Tensor 00:01:59.200 --> 00:02:03.840 so the hardware in its Pixel devices would be a perfect fit for Android, in their words. 00:02:03.840 --> 00:02:08.640 Which, of course, Google also develops. Now, although we don't have tons of Tensor specifications 00:02:08.640 --> 00:02:12.380 to throw at you right now, it's pretty clear that much of Tensor's silicon 00:02:12.380 --> 00:02:17.600 is going to be dedicated towards image processing, which isn't too surprising considering the Pixel lineup 00:02:17.600 --> 00:02:21.240 has long featured strong camera performance as a selling point. 00:02:21.240 --> 00:02:25.600 And because modern phones rely heavily on AI to make its images look good, 00:02:25.640 --> 00:02:29.920 especially important because the sensors inside mobile cameras aren't all that large. 00:02:29.920 --> 00:02:34.440 One of Google's primary focuses for Tensor is for it to deliver speedy performance 00:02:34.440 --> 00:02:39.520 with AI-intensive tasks, such as improving grainy or blurry nighttime photos. 00:02:39.520 --> 00:02:44.800 The enhanced AI capabilities of Tensor can take images from the multiple lenses on Pixel 00:02:44.800 --> 00:02:47.880 and run several different AI algorithms at once 00:02:47.880 --> 00:02:53.160 to improve picture quality and sharpness, even in tough situations, including fast motion. 00:02:53.160 --> 00:02:57.120 Or when there's so much light, the picture is in danger of being blown out. 00:02:57.120 --> 00:03:02.960 I was born in the light. The emphasis on AI should also help with voice input. 00:03:02.960 --> 00:03:07.160 This means your phone should be able to better tell the difference between which word is a command 00:03:07.160 --> 00:03:11.280 versus which word is a part of your message, provide better auto-generated captions 00:03:11.280 --> 00:03:15.280 if you're watching a video in a foreign language, and deliver better experiences with apps 00:03:15.280 --> 00:03:19.320 that use any form of AI, such as those that detect objects. 00:03:19.320 --> 00:03:24.680 This also means that smartphones with a Tensor chip won't have to rely as much on cloud AI processing, 00:03:24.720 --> 00:03:30.240 freeing up bandwidth, and possibly allowing some apps to function without needing a strong connection to the cloud. 00:03:30.240 --> 00:03:35.960 Because you don't need Google to see everything you're doing. But how long will you need to wait to get your hands on it? 00:03:35.960 --> 00:03:40.640 Well, Google doesn't wanna just sit on this thing forever, especially as it's already behind Apple 00:03:40.640 --> 00:03:44.080 in terms of vertically integrating SOCs into its own phones. 00:03:44.080 --> 00:03:47.440 And the Pixel 6, which will be the first device to feature the Tensor chip, 00:03:47.440 --> 00:03:51.720 should be available in fall 2021. Google has left open the possibility 00:03:51.720 --> 00:03:55.960 it'll show up in other devices at some point, but let's see how well their Tensor-based phones 00:03:55.960 --> 00:03:59.480 take off first. If they do well, maybe we'll see Google's stores 00:03:59.480 --> 00:04:02.760 popping up at malls, where everything is painted white. 00:04:02.760 --> 00:04:06.120 Thanks for watching, guys. Like or dislike the video, whatever you wanna do. 00:04:06.120 --> 00:04:09.400 Check out our other videos, and comment below with video suggestions. 00:04:09.400 --> 00:04:15.280 And don't forget to subscribe and follow. Or else you won't get any Tensor chips anywhere.