A custom built eight-core processing unit that can run three trillion operations each second powers Google's camera — and it has even more planned.
We've been using the Pixel 2 and it's bigger sibling the Pixel 2 XL for a bit. While getting things ready for a review and testing all the things there was one consistent discussion, and it centered around the photos the Pixel 2's cameras were churning out. What we were seeing from both the 12.2MP rear camera and 8MP front-facing was just so much better than any other phone we've ever used. And we've used a lot of them.
Fast forward a few days and we found out why. Google has designed and is using a custom imaging-focused SoC (system on chip) in the Pixel 2, and it's called Pixel Visual Core.
We don't have all the details. Google isn't ready to share them and maybe isn't even aware of just what this custom chip is capable of yet. What we do know is that the Pixel Visual Core is built around a Google-designed eight-core Image Processing Unit. This IPU can run three trillion operations each second while running from the tiny battery inside a mobile phone. Right now, Google's HDR+ routines are the only processes using this IPU, and it runs fives times faster while using less than one-tenth of the energy than it would if it ran through the standard image processor in the Snapdragon 835.
Google says this is possible because of how well the software and hardware have been matched with each other. The software on the Pixel 2 controls "many more" details of the hardware than you would find in a typical processor to software arrangement. By handing off control to software, the hardware can become a lot more simple and efficient.
Google is a software company first and foremost. It's no wonder that its first custom mobile SoC leverages software the way other companies use hardware.
Of course, this means the software then becomes more and more complex. Rather than use standard methods of writing code, building it into a finished product and then trying to manage everything after all the work is finished, Google has turned to machine learning coding languages. Using Halide for the actual image processing and TensorFlow for the machine learning components themselves, Google has built its own software compiler that can optimize the finished production code into software built specifically for the hardware involved.
Right now the only thing using the Pixel Visual Core is the camera's HDR+ technology. We've witnessed first-hand what it is capable of doing and we really like what we see. This is different than anything we've seen before. This is what comes next.
HDR+ is only the beginning.
Google says we should expect to see the Android 8.1 developer preview in the coming weeks and Pixel Visual Core will be opened up as a developer option. The goal is to give all third-party apps access through the Android Camera API. This will give every developer a way to use Google's HDR+ and the Pixel Visual Core, and we expect to see some really big things.
For the one last thing we always love to hear about, Google says that we should remember the Pixel Visual Core is programmable and they are already building the next set of applications that can harness its power. As Google adds more abilities to its new SoC, the Pixel 2 and 2 XL will continue to get better and be able to do more. New imaging and machine learning applications are coming throughout the life of the Pixel 2, and we're ready for them.
Post a Comment