Google Cloud and NVIDIA have expanded partnership to announce new AI infrastructure and software for customers to build and deploy massive models for generative AI and speed data science workloads.

In a fireside chat at Google Cloud Next, Google Cloud CEO Thomas Kurian and NVIDIA founder and CEO Jensen Huang discussed how the partnership is bringing end-to-end machine learning services to some of the largest AI customers in the world — including by making it easy to run AI supercomputers with Google Cloud offerings built on NVIDIA technologies. The new hardware and software integrations utilize the same NVIDIA technologies employed over the past two years by Google DeepMind and Google research teams.

“We’re at an inflection point where accelerated computing and generative AI have come together to speed innovation at an unprecedented pace,” Huang said. “Our expanded collaboration with Google Cloud will help developers accelerate their work with infrastructure, software and services that supercharge energy efficiency and reduce costs.”

“Google Cloud has a long history of innovating in AI to foster and speed innovation for our customers,” Kurian said. “Many of Google’s products are built and served on NVIDIA GPUs, and many of our customers are seeking out NVIDIA accelerated computing to power efficient development of LLMs to advance generative AI.”

Google’s framework for building massive large language models (LLMs), PaxML, is now optimized for NVIDIA accelerated computing.

Originally built to span multiple Google TPU accelerator slices, PaxML now enables developers to use NVIDIA H100 and A100 Tensor Core GPUs for advanced and fully configurable experimentation and scale. A GPU-optimized PaxML container is available immediately in the NVIDIA NGC software catalog. In addition, PaxML runs on JAX, which has been optimized for GPUs leveraging the OpenXLA compiler.

Google DeepMind and other Google researchers are among the first to use PaxML with NVIDIA GPUs for exploratory research.

The NVIDIA-optimized container for PaxML will be available immediately on the NVIDIA NGC container registry to researchers, startups and enterprises worldwide that are building the next generation of AI-powered applications.

Additionally, the companies announced Google’s integration of serverless Spark with NVIDIA GPUs through Google’s Dataproc service. This will help data scientists speed Apache Spark workloads to prepare data for AI development.