Google’s new Trillium AI chip delivers 4x speed and supports Gemini 2.0

Google Just revealed Trilliuma sixth-generation AI accelerator chip, which delivers performance improvements that could fundamentally change the economics of AI development while pushing the boundaries of what is possible in machine learning.

The custom processor that supports the training that Google recently announced Gemini 2.0 The AI ​​model delivers four times the training performance of its predecessor while using much less energy. This achievement comes at a crucial moment, as technology companies race to build increasingly sophisticated artificial intelligence systems that require massive computational resources.

“TPUs support 100% of the training and inference in Gemini 2.0,” Google CEO Sundar Pichai explained in an article. Latest ad Highlighting the chip’s central role in the company’s AI strategy. The scale of the deployment is unprecedented: Google has connected more than 100,000 Trillium chips into a single network fabric, creating what amounts to one of the most powerful AI supercomputers in the world.

How Trillium’s 4x performance boost is transforming AI development

Trillium specifications represent significant advances across multiple dimensions. The chip delivers a 4.7x increase in peak compute performance per chip over its predecessor, while doubling high-bandwidth memory capacity and chip interconnect bandwidth. Perhaps most importantly, it delivers a 67% increase in energy efficiency – a critical metric as data centers grapple with the massive power demands of AI training.

“When training the Llama-2-70B model, our tests showed that Trillium achieves near-linear scaling from a 4-chip Trillium-256 pod to a 36-chip Trillium-256 pod with 99% scaling efficiency.” Marc Lemire, Vice President of Compute and AI Infrastructure at Google Cloud. This level of scaling efficiency is particularly notable given the challenges typically associated with distributed computing at this scale.

The Economics of Innovation: Why Trillium is changing the game for AI startups

The implications for Trillium’s business extend beyond initial performance metrics. Google claims the chip delivers up to a 2.5x improvement in training performance per dollar compared to the previous generation, which could reshape the economics of AI development.

This cost-effectiveness can be especially important for organizations and startups developing large language models. AI21 Labs, one of Trillium’s early customers, has already reported significant improvements. “The advances in scale, speed and cost-effectiveness are significant,” he noted. Barack LenzCTO at AI21 Labs, in the announcement.

Rising to new heights: Google’s 100,000-chip AI supernetwork

Google’s deployment of Trillium within its AI supercomputer architecture demonstrates the company’s integrated approach to AI infrastructure. The system combines more than 100,000 Trillium chips with a Jupiter network fabric capable of delivering 13 petabits per second of chunked bandwidth – making it possible to scale a single distributed training task across hundreds of thousands of accelerators.

“Flash usage growth has exceeded 900% which is incredible,” Logan Kilpatrick, a product manager on Google’s AI Studio team, noted during the developer conference, highlighting the rapidly growing demand for AI computing resources.

Beyond Nvidia: Google’s bold move in the AI ​​chip wars

The release of Trillium has intensified competition in artificial intelligence devices Nvidia It dominated with its GPU-based solutions. While Nvidia’s chips remain the industry standard for many AI applications, Google’s custom silicon approach could offer advantages for specific workloads, especially in training very large models.

Industry analysts point out that Google’s huge investment in developing custom chips reflects a strategic bet on the growing importance of AI infrastructure. The company’s decision to make Trillium available to cloud customers signals a desire to compete more aggressively in the cloud AI market, where it faces strong competition from… Microsoft Azure and Amazon Web Services.

Fueling the future: What Trillium means for tomorrow’s AI

The implications for Trillium’s capabilities extend beyond direct performance gains. The chip’s ability to efficiently handle mixed workloads — from training large models to running inference for production applications — points to a future in which AI computing becomes easier and more cost-effective.

For the broader tech industry, Trillium’s release signals that the race for AI hardware supremacy is entering a new phase. As companies push the boundaries of what is possible with AI, the ability to design and deploy specialized devices at scale could become an increasingly crucial competitive advantage.

“We are still in the early stages of what is possible with AI,” Google DeepMind CEO Demis Hassabis wrote at the company. Blog post. “Having the right infrastructure – both hardware and software – will be critical as we continue to push the boundaries of what AI can do.”

As the industry moves toward more sophisticated AI models that can operate autonomously and reason across multiple types of information, demand for the underlying hardware will increase. With Trillium, Google has demonstrated that it intends to stay at the forefront of this evolution, by investing in the infrastructure that will support the next generation of AI advancement.