Trillium TPU, built to power the future of AI

To deliver the next frontier of models and enable you to do the same, we’re excited to announce Trillium, our sixth-generation TPU, the most performant and most energy-efficient TPU to date.

More than a decade ago, Google recognized the need for a first-of-its-kind chip for machine learning. In 2013, we began work on the world’s first purpose-built AI accelerator, TPU v1, followed by the first Cloud TPU in 2017. Without TPUs, many of Google’s most popular services — such as real-time voice search, photo object recognition, and interactive language translation, along with the state-of-the-art foundation models such as Gemini, Imagen, and Gemma — would not be possible.

Trillium TPUs achieve an impressive 4.7X increase in peak compute performance per chip compared to TPU v5e. We doubled the High Bandwidth Memory (HBM) capacity and bandwidth, and also doubled the Interchip Interconnect (ICI) bandwidth over TPU v5e. Additionally, Trillium is equipped with third-generation SparseCore, a specialized accelerator for processing ultra-large embeddings common in advanced ranking and recommendation workloads. Trillium TPUs make it possible to train the next wave of foundation models faster and serve those models with reduced latency and lower cost. Critically, our sixth-generation TPUs are also our most sustainable: Trillium TPUs are over 67% more energy-efficient than TPU v5e.

[returnedinformation]

What about the successor of Google Coral? I'm asking for us mortals

[arirajuns]

Thanks for the notebookllm. Its very useful.

[Atom]

Will there be special programs/offers for university research institutes?

[joe_hoeller_chicago]

TPUs although very useful and a big gain in speed for training large AI models, are only on GCP. So if you’re not on that cloud, you can’t use them.

[sharex]

The narration feels like it's been made by a vlogger who just started yesterday in his basement.

[EarthCreature.]

Not 1 solar panel on Google's building. Shame on them

[andreamanninfiaschi]

All of this power, all of this technology...for watching dogs go surfing.

[jjgerald]

TPUs. Mga gawa ko yan, gawa namin yan.

Google is mine.

[cystarkman]

That facility reminds of other earlier and also larger platforms for intelligence.

[Unineil]

I want a tpu on my pi that interacts more with local 1-3b llms.

[dineshlamarumba]

The future of ai harsware technology should be such that we xan also develop micro intelligent robots, how will gpu, tpu in any way fit such micro robots?

Will it be used in Project Nimbus in the West Bank?

[julioconradomarinardila]

Excelente herramienta para LA Google cloud

[mcd]

Looking good! Accelerate is the way….to 1 Trillium FLOPS!

[matt]

i think AI is out and too hyped, i am betting bigger on the metaverse and organoid intelligence

[syntaxed]

If only the realtime translation was good - It and subtitles are horrible, no one uses it but hey thanks for forcing me to verify login/location every 5minutes

[sahilx]

No more cpus or gpus for humans anymore. 😢
Everything is for AI, from now on. 🥺

[GobalKrishnanV]

I have doubt, human brain not have skill all topic, even for single topic no one be excellent. Then how we say want we doing is correct?😅😊❤❤

[HaiLeQuang]

There is no point to build AI chip if your AI product sucks. Move fast or you would be the Nokia of 2020s years.