Techspert Explains: CPU vs. GPU vs. TPU Differences

Back at I/O in May, we unveiled Trillium, the sixth generation of our custom-designed chip called the Tensor Processing Unit, or TPU. Today, we're excited to share that it's now available for Google Cloud customers in preview. TPUs are the magic behind the AI that makes your Google devices and apps super helpful, and Trillium is the most powerful and eco-friendly TPU we've ever made.

So, what's a TPU, and what makes Trillium unique? To get the full picture, it's helpful to know about other types of compute processors like CPUs and GPUs, and what sets them apart. Chelsie Czop, a product manager working on AI infrastructure at Google Cloud, can explain it all. "I collaborate with various teams to ensure our platforms are as efficient as they can be for our customers building AI products," she says. And according to Chelsie, Google's TPUs are a big reason why our AI products are so effective.

Let's dive into the basics! What are CPUs, GPUs, and TPUs?

These are all chips that act as processors for computing tasks. Imagine your brain as a computer doing things like reading or solving math problems. Each of these activities is like a computing task. So, when you use your phone to snap a photo, send a text, or open an app, your phone's brain, or processor, is handling those tasks.

What do these acronyms stand for?

Even though CPUs, GPUs, and TPUs are all processors, they become increasingly specialized. CPU stands for Central Processing Unit. These are versatile chips that can handle a wide variety of tasks. Just like your brain, some tasks might take longer if the CPU isn't specialized for them.

Next up is the GPU, or Graphics Processing Unit. GPUs are the powerhouse for accelerated computing tasks, from rendering graphics to running AI workloads. They're a type of ASIC, or application-specific integrated circuit. Integrated circuits are usually made from silicon, which is why you might hear chips called "silicon" — it's the same thing (and yep, that's where "Silicon Valley" got its name!). In a nutshell, ASICs are designed for one specific purpose.

The TPU, or Tensor Processing Unit, is Google's own ASIC. We designed TPUs from scratch to handle AI-based computing tasks, making them even more specialized than CPUs and GPUs. TPUs are at the core of some of Google's most popular AI services, including Search, YouTube, and DeepMind's large language models.

Got it, so these chips are what make our devices tick. Where would I find CPUs, GPUs, and TPUs?

CPUs and GPUs are found in everyday items: CPUs are in nearly every smartphone and laptop, while GPUs are common in high-end gaming systems and some desktop computers. TPUs, on the other hand, are only found in Google data centers — massive buildings packed with racks of TPUs, working around the clock to keep Google's and our Cloud customers' AI services running globally.

What prompted Google to start developing TPUs?

CPUs were invented in the late 1950s, and GPUs came along in the late '90s. At Google, we started thinking about TPUs about a decade ago. Our speech recognition services were improving rapidly, and we realized that if every user started talking to Google for just three minutes a day, we'd need to double our data center computers. We knew we needed something far more efficient than the off-the-shelf hardware available at the time — and we needed a lot more processing power from each chip. So, we decided to build our own!

And that "T" stands for Tensor, right? Why?

Yep — a "tensor" is the general term for the data structures used in machine learning. There's a lot of math happening behind the scenes to make AI tasks work. With our latest TPU, Trillium, we've boosted the number of calculations it can perform: Trillium has 4.7 times the peak compute performance per chip compared to the previous generation, TPU v5e.

What does that mean, exactly?

It means Trillium can handle all the calculations needed for complex math 4.7 times faster than the last version. Not only is Trillium faster, but it can also manage larger, more complex workloads.

Is there anything else that makes it an improvement over our last-gen TPU?

Another big improvement with Trillium is that it's our most sustainable TPU yet — it's 67% more energy-efficient than our previous TPU. As AI demand keeps growing, the industry needs to scale infrastructure in a sustainable way. Trillium uses less power to do the same work.

Now that customers are starting to use it, what kind of impact do you think Trillium will have?

We're already seeing some amazing things powered by Trillium! Customers are using it for technologies that analyze RNA for various diseases, convert text to video at lightning speeds, and more. And that's just from our initial users — now that Trillium's in preview, we're excited to see what people will do with it.