Google ‘completely comfortable’ with Tensor not winning standards and vertical integration benefiting the Pixel

The Tensor G2 is this week’s topic on Made by Google Podcast, and Episode 4 provides an overview of the Pixel’s approach to chips, as well as benchmarks.

Monica Gupta is Senior Director of Product Management at Google Silicon Teams and strives to “focus on what [Google] need to[s] five years from now” to get their chips. The interview confirmed how this in-house approach to chips involved the silicon team talking with Google AI researchers “to find out exactly where machine learning models are headed in five years.”

I don’t make decisions based on where machine learning is today, and I can say that because I work at Google. Same with the software that our software team does. I know where the software team wants to take user experiences five years from now. This is the benefit of not being a silicon supplier trader, but an in-house silicon supplier. So these swap decisions are very tricky, but I think it gets a little easier when you’re vertically integrated.

Meanwhile, there was a section on the Tensor team’s outlook on benchmarks and what prioritizes Pixel instead:

I think Classic Standards served a purpose at one point, but I think the industry has evolved since then. And if you look at what Google is trying to do by pushing AI innovations into the smartphone, because we feel that’s an approach that will provide useful experiences like some of the ones I just mentioned, the classic benchmarks were composed at a time when AI and phones didn’t exist until.

They may tell some stories, but we don’t feel like they’re telling the whole story. And so what we measure for us is the actual software workloads that we’re running on our chip, and then with each generation of tensors we strive to make it better, whether it’s better quality, better performance, or less power.

Google says it’s “absolutely comfortable” not winning or looking great in terms of benchmarks because it prioritizes the end-user experiences that result from the segment decisions you make.

Like on the Pixel 6 and Pixel 7, you can see all the amazing innovations that we’ve been up to, and many of them were like the first on the Pixel. So we are very comfortable with this approach

This approach reflects what Google values ​​(i.e., AI-powered features) – even though there are AI standards – but it will be interesting to see if their philosophy changes as the company’s silicon matures. For example, there is a ready-made nature of the first two generations that could eventually turn out to be dedicated cores, such as Apple, with more experience in this area.

When asked about the silicon roadmap, Gupta reiterated an opinion from last year about how the ultimate goal is to get Tensor to support ambient computing:

…the overarching vision for us and the Tensor family revolves around ambient computing. Ambient computing means technology makes your life easier. I think we have a lot of evidence for this that we talked about today, whether it makes photography easier, whether it makes phone calls and how you use your phone, like your daily tasks, easier.

I’d say we’re building on that vision of ambient computing and figuring out how to do things that are incredibly complex and precise on the chip in an energy efficient way that will unleash some ambient computing experiences.

More on Pixel:

FTC: We use affiliate links to earn income. more.

Check out 9to5Google on YouTube for more news:

#Google #completely #comfortable #Tensor #winning #standards #vertical #integration #benefiting #Pixel

Leave a Comment

Your email address will not be published. Required fields are marked *