Google is notching up its chipmaking sport as the corporate seeks to cut back its dependence on chips made by different firms for its cloud computing enterprise. Google on Monday unveiled two new variants of its tensor processing items (TPUs), in-house developed chips meant for AI workloads, that broaden its chip portfolio to extra use instances: confidential inference and extra reminiscence capability for big workloads.
The brand new items, TPU 8tx and TPU 8i Max, are a part of the corporate’s eighth technology of chips launched in April this 12 months, and construct on what these chips provided, which was extra inference capability and assist for big coaching necessities.
TPU 8tx, designed with AMD, relies on the TPU 8t chip, and is geared particularly for confidentially coaching and operating AI fashions, Google mentioned in a weblog submit. The chip “builds safety instantly into the {hardware} layer, conserving delicate knowledge and mannequin weights protected and remoted even whereas in energetic use,” the corporate wrote.
The chip additionally has a customized inferencing community that lets it run reasoning alongside coaching, so it may be used to distill and prepare small fashions on the fly. That capability, in response to Google, might show helpful for instances the place an organization would possibly wish to prepare a mannequin on its proprietary knowledge with out having to maneuver the info between gadgets.
In the meantime, TPU 8i Max mainly combines two of the inference-focused TPU 8i chips the corporate launched in April, doubling the reminiscence capability, processing energy and bandwidth, however in a single package deal. The result’s a chip that lets AI fashions deal with a bigger vary of data throughout queries.
The chip launch comes as hyperscalers like Microsoft, Amazon, and even AI labs like OpenAI and Anthropic are engaged on their very own traces of AI-specific semiconductors and accompanying software program to be able to cut back their reliance on the dominance of Nvidia.
The information additionally comes at a very weak second for Nvidia, whose CEO Jensen Huang over the weekend admitted to having an affair with somebody alleged to be a Chinese language spy.
Whenever you buy by way of hyperlinks in our articles, we could earn a small fee. This doesn’t have an effect on our editorial independence.
