Google has showcased its new “sixth-generation” Trillium TPUs, and also announced NVIDIA Blackwell AI server availability in early 2025.
Google Looks To Boost Its AI Capabilities Through Adding Cutting-Edge Hardware Into Its Arsenal Such as The Trillium TPU, & NVIDIA Blackwell-Powered Servers For Early 2025
The I/O 2024 was once again filled with the word “AI” throughout the showcase, and we can’t debate this at all, considering how immense artificial intelligence is for Google and its partners. We saw the integration of AI features into mainstream applications such as Gmail and Google Photos, and along with that, the firm gave us a rundown of what to expect with AI models like Gemini and the newly disclosed Veo. To back everything up on a hardware level, Google also announced their newest “Trilium” sixth-generation TPUs.
Expanding on Trillium, Google revealed in a blog post that the new TPUs could achieve a 4.7 times increase in peak compute performance compared to the previous generation, doubling both the HBM capacities and the interconnect bandwidth. This allows Google to operate its AI models and activities in a much-enhanced manner. Moreover, the sixth-generation TPUs are known to be 67% more power efficient, allowing their mass adoption in data centers.
Trillium TPUs have already been announced as integrated into Google’s AI Hypercomputers, and the company has partnered up with the tech firm Hugging Face for “streamlined model training and serving.”
Our partnership with Google Cloud makes it easier for Hugging Face users to fine-tune and run open models on Google Cloud’s AI infrastructure, including TPUs.
We are excited to further accelerate open source AI with the upcoming sixth-generation Trillium TPUs, and we expect open models to continue to deliver optimal performance thanks to the 4.7X increase in performance per chip compared to the previous generation. We will make the performance of Trillium easily available to all AI builders through our new Optimum-TPU library.
– Jeff Boudier, Head of Product, Hugging Face
Apart from developing TPUs, Google also announced that they have included NVIDIA’s Blackwell architecture into their arsenal of “AI weapons,” claiming that it will enable them to take compute performance to a whole new level. It’s important to note that few companies have gained access to Blackwell products, showing how “rare” they are in the markets, at least for now. NVIDIA is indeed adopting an exclusive relationship with clients, preferring one. The Blackwell GPUs will power the Google cloud server solution and will be available by early 2025.
With their most powerful TPUs coupled with the integration of NVIDIA’s Blackwell architecture, Google is certainly going to witness a boost in AI computing performance. They certainly need it, considering the scale of progress the firm has made in integrating generative AI into mainstream consumer applications and progressing rapidly with new and existing AI models.