跳到正文
Data Center Knowledge· Wylie Wong·· 12 天前AI 評分47

Google 推出面向企業 AI 的 TPU 8tx 和 TPU 8i Max

Google Adds TPU 8tx, 8i Max for Enterprise AI

AI 導讀

Google Cloud 推出 TPU 8tx 和 TPU 8i Max,分別加強敏感 AI 工作負載的安全保護,以及支援大型上下文窗口的 AI 智能體推理。TPU 8tx 加入硬件層級加密,並配搭自訂 AMD CPU 主機;TPU 8i Max 將兩枚 TPU 8i 整合至單一封裝,令運算能力、SRAM 和 HBM 頻寬倍增,並提供 384 GB 高頻寬記憶體。

正文

3 Min Read

TPU 8tx + TPU 8i Max product art by Google Cloud

Google Cloud said its new TPU 8i Max combines two TPU 8i chips into one package, doubling compute, SRAM, and memory bandwidth versus TPU 8i.Image via Google Cloud

To meet growing demand for AI inference, Google Cloud on Monday beefed up its Tensor Processing Unit family with two new chips: a more secure version of its TPU 8t accelerator and a higher-capacity version of its TPU 8i inference chip.

The new processors are designed to address two emerging enterprise requirements: stronger protection for sensitive AI workloads and better performance for agentic AI applications with large context windows, said Mark Lohmeyer, vice president and general manager of AI and computing infrastructure at Google Cloud.

The announcements build on Google Cloud’s eighth-generation TPU chips introduced in April. At that time, the company launched TPU 8t for training, which it said delivers up to 2.7 times better training performance than the prior generation, and TPU 8i for inference, which it said offers 80% better inference price-performance.

TPU 8tx Adds Security; TPU 8i Max Boosts Capacity

One of the new accelerators, TPU 8tx, is built on the same silicon as TPU 8t and adds hardware-level encryption for secure inference and small-scale training. The chip is paired with custom AMD CPU hosts to help protect and isolate sensitive data and model weights, even while inference is actively running, Lohmeyer said during a media briefing.

Related:Huang Resigns: Nvidia's AI Infrastructure Roadmap in Question

Google Cloud also introduced TPU 8i Max, a variant aimed at enterprises running increasingly agentic workloads with large context windows. Lohmeyer said the chip is intended to address memory constraints by effectively “adding more lanes to the highway.”

To create TPU 8i Max, Google Cloud combined two TPU 8i chips into a single package, doubling compute, SRAM, and HBM bandwidth compared with TPU 8i, the company said. TPU 8i Max delivers 384 GB of high-bandwidth memory, Lohmeyer said.

“We’re just seeing tremendous demand for the TPU platform and systems,” Lohmeyer said during the briefing. He added that Google Cloud is also seeing a broadening of AI use cases, including demand for confidential computing and for agentic inference with very large context windows while maintaining strong performance.

Custom TPUs Could Sharpen Google Cloud’s Edge

Dave McCarthy, group vice president at IDC, said Google Cloud’s move toward more specialized custom AI chips could help it further differentiate itself from rival cloud providers.

“The fact that Google was able to launch the 8i and 8t earlier this year, and now they’re coming out with these new ones within the same calendar year, is just showing the pace of innovation,” McCarthy said. “They’re showing their commitment to R&D to build this stuff. And if what they’re saying is true – twice the bandwidth, all these new features – that’s showing they’re potentially outpacing the competition in terms of new product releases.”

Related:AMD Launches Epyc SP7, Helion as AI Rack Demand Rises

McCarthy described confidential computing as an emerging requirement for AI workloads. Cloud providers have traditionally focused on protecting data in transit and at rest, he said, whereas confidential computing is intended to protect data in use.

“If Google can come out with a strong confidential compute story, it kind of gives them a little bit of an edge,” said McCarthy, adding that Nvidia is still developing its strategy in this area.

Google Cloud also said TPU 8i Max allows more silicon to fit within an existing rack footprint, enabling significantly more computing power per square foot and per watt in its data centers.

That could help Google Cloud increase capacity without adding more racks or building more data centers, at least in the near term, potentially easing some of the capacity pressures facing the industry, McCarthy said.

More broadly, McCarthy said Google Cloud’s strategy of continually enhancing its silicon and making TPUs available through other providers could expand the company’s market reach while creating an additional revenue stream.

Related:Nvidia Probes Huang Ties as AI Industry Weighs the Fallout

Oracle recently announced it would make TPU chips available to customers of Oracle Cloud Infrastructure. During the media briefing, Lohmeyer also said Google Cloud would allow select customers to deploy the TPU 8tx in their own data centers.

About the Author

Wylie Wong

Chips and Hardware Writer

Wylie Wong is a journalist and freelance writer specializing in technology, business and sports. He previously worked at CNET, Computerworld and CRN and loves covering and learning about the advances and ever-changing dynamics of the technology industry. On the sports front, Wylie is co-author of Giants: Where Have You Gone, a where-are-they-now book on former San Francisco Giants. He previously launched and wrote a Giants blog for the San Jose Mercury News, and in recent years, has enjoyed writing about the intersection of technology and sports.

來源:Data Center Knowledge · datacenterknowledge.com