When Google launched the Pixel phones back in 2016, many saw this as the company’s effort to set new standards for what an Android phone should be. Google may have been able to change the standards for what phone cameras can do, but the company surely hasn’t changed smartphones…yet. Enter, Google Tensor, a chip that’s manufactured by Google and will replace Qualcomm SoCs that have powered Pixel phones for the last five years. What do we know about the Tensor chips so far? Not much, but we can make some educated guesses.
What is Google Tensor?
As mentioned above, Google Tensor is a system-on-chip, which means it will be the brains of the Pixel 6 devices. It’s an octa-core chipset built on Samsung’s 5 nanometer fabrication process using processor architectures from ARM. Being the developer of Android, this gives Google an undue advantage of sorts on the smartphone space, since the company can tailor Android to make the new Tensor chip perform better than others. But why is a new chip different?
Here’s why…
You might argue that Google has no background in hardware, and the company’s history with Pixel phones so far doesn’t really give us confidence, except that while Google is new to building new phones, it’s no stranger to building new chips. The Google Tensor is named after something you may have heard of before — the Tensor Processing Unit (TPU), a chip that is quite popular in the data center space.
“Tensor Processing Units (TPUs) are Google’s custom-developed application-specific integrated circuits (ASICs) used to accelerate machine learning (ML) workloads. Cloud TPUs allow you to access TPUs from Compute Engine, Google Kubernetes Engine and AI Platform.”
More than half of the AI and ML algorithms that our phones process today actually happens on the cloud. That’s why the Google Assistant or Alexa show a buffering animation when you give them a command. They send your voice command to the cloud, where a processor that’s significantly more powerful than the one on your phone processes the command and tells the phone what to do.
Claim number one…
Which brings us to the first claim that Google is making with the new Google Tensor chip — that it will do more AI processing “on-device”.
When you put two and two together, that’s more than believable right now. TPUs were designed to process AI and ML and possibly replace the powerful GPUs that take care of AI processing at data centers today. Here’s an excerpt from a GPU vs TPU comparison done by Bangalore-based IT firm Mphasis:
“The results indicate that speedups (from TPUs) by a factor of more than 15 are possible, but they appear to come at a cost. First, v3-8 TPU’s pricing is 5.5 times greater than for the P100, which by itself sounds alarming; however, calculating the amount of training per dollar yields a monetary savings of more than 64% since the TPU is so much faster.”
Essentially, Google knows what kind of power is required to process AI algorithms. Which means that in many instances, the Tensor chip won’t need to send data to the cloud. It already knows how to process the commands.
Keep a check on your expectations though, because there’s no way that a chip inside the tiny smartphone can match what a data center chip does. Simply put, phones just cannot provide the same thermal headroom and performance required for data center level performance, at least not till we design the chips that ran Iron Man’s AI in Marvel’s Avengers.
No comments:
Post a Comment