Google Cloud Joins with NVIDIA to Advance AI
Google Cloud’s Next '23 conference in San Francisco this week has featured a slew of generative AI announcements, including a partnership with NVIDIA that builds on that vendor’s H100 chips to buttress Google’s own supercomputing resources.
“As generative AI evolves, so do infrastructure requirements,” said Google Cloud CEO Thomas Kurian in a keynote address yesterday. “Our ultrascale, high reliable AI supercomputing systems combine TPU and GPU accelerators, high-performance AI-optimized storage, scalable networking, offloads, and water cooling.”
The reference to TPUs (tensor processing units) working in tandem with GPUs (graphics processing units from NVIDIA) represents Google’s efforts to join with its largest competitor in supercomputing chips rather than struggle for market position against the maker of the H100 behemoth. It also shows that NVIDIA is more open than has been previously supposed to sharing its much-sought-after components with a vendor that hopes to compete against its offerings. Hence, Google Cloud’s new, upgraded TPU, the Cloud TPU v5e, available in preview, was offered in the same breath with NVIDIA GPU instances at Next this week.
“Many people ask me, what’s the relationship between TPUs, GPUs, how do you think about it,” said CEO Kurian. “Very simply put, as AI evolves, the needs on the hardware architecture and software stack evolve from training to inferencing to new capabilities like embeddings, and we want to offer customers the broadest, most optimized choice of accelerators. We’re actually offering 13 different types of accelerators in GCP.”
For example, Google Cloud has also hitched its supercomputing software wagon to NVIDIA’s star in the form of new A3 VMs, which virtualize supercomputing instances based on NVIDIA’s H100s. And in a cameo appearance at the keynote, NVIDIA CEO Jensen Huang said that Google and NVIDIA will work together on the next generation of NVIDIA supercomputer, the DGX GH200, which is based on NVIDIA’s Grace Hopper chip.
Google Cloud CEO Thomas Kurian meets NVIDIA CEO Jensen Huang at Next. Source: Google Cloud
Google Cloud had a slew of additional AI announcements, including improvements to its Vertex AI platform, which offers access to over 100 large language models (LLMs), the backbone of generative AI applications. Included are foundation models from third parties as well as open-source models. There also are upgrades to Google Cloud’s own models, such as vertical models Sec-PaLM 2 and Med-PaLM 2, aimed at cybersecurity and healthcare, respectively.
Google Cloud also announced the general availability of Duet AI, a chatbot and code assistant for use in Workspace and Google Cloud as well as by developers.
GCP Adds Multicloud Networking
Stuck in among the generative AI avalanche was a multicloud networking platform called Cross-Cloud Network. This is a complex but intriguing upgrade to the GCP global network that incorporates high bandwidth interconnects, security buttressed with Palo Alto Networks technologies, and on-ramps to Alibaba Cloud, Amazon Web Services, Microsoft Azure, and Oracle Cloud Infrastructure. From the Google Cloud press release: “As more organizations leverage hybrid and multicloud environments, Cross-Cloud Network helps ease operational requirements and lets them focus on running the business versus the network.”
“Cross-Cloud Network aims to simplify and speed up connectivity, scalability and security as we leverage emerging capabilities such as AI and machine learning on Google Cloud,” stated Ken Kirchoff, senior director of enterprise architecture at Priceline, in the press release.
All told, Google Cloud is pulling out all the stops to attract customers to deploy GCP’s many features and functions in an effort to ride the AI wave into the future – a strategy backed by parent company Alphabet. “We truly believe we are embarking on a golden age of innovation,” said Alphabet CEO Sundar Pichai before introducing Kurian’s keynote. “We all know we are in the midst of one of the most profound technology shifts in history.”
Futuriom Take: Google Cloud is betting big on AI and its accelerated computing infrastructure. Its alliance with NVIDIA is a smart move that should advance the effort.
