Oracle is adding tens of thousands of NVIDIA’s AI-processing graphics chips to its cloud and giving customers access to NVIDIA software tools that can wring even more performance from them, as part of an expanded partnership between the companies announced at Oracle CloudWorld.
Oracle Cloud Infrastructure (OCI) will expand its offering of online accessible computers running NVIDIA’s powerful A100 graphics processors, connected by fast networking and aimed at industries including banking, healthcare, and manufacturing. Oracle also plans to offer the chipmaker’s upcoming H100 “Hopper” GPUs, which can shrink AI model training time from 7 days to 20 hours for some workloads.
Oracle Cloud will also run NVIDIA software called RAPIDS Accelerator that can speed AI jobs running on the company’s GPUs and plans to offer NVIDIA Clara software for medical imaging, genomics research, and drug discovery.
“The writing’s on the wall, the trends are clear: All workloads in the future will be accelerated,” NVIDIA Chief Executive Jensen Huang said in a keynote address October 18 at Oracle CloudWorld in Las Vegas. The company’s chips—whose ability to divide work among many processor cores can speed up tasks such as training and applying neural networks that learn from large amounts of data—currently accelerate about 5% of computing work.
That will increase sharply, said Huang, as business and scientific teams look to lower the cost and power consumption of computing at a time when the traditional way to gain performance—by adding transistors to chips—has slowed. “The next 10 years will be some of the most exciting times in our industry,” he said.
Artificial intelligence systems have advanced to the point where they can compose convincing texts; draft images, art, and video clips based on simple instructions; or even predict the efficacy of drug candidates. Yet the high computing cost and expertise to run such powerful AI systems have kept them out of many organizations’ hands.
That’s starting to change as high-performance cloud computing services broaden access to self-learning, data-hungry AI systems. The newest AI models feature massive amounts of data and complex sets of statistical weights and need fast communication among processing cores that exceed the capabilities of computers in most companies’ own data centers.
AI adoption among global businesses is rising steadily—56% of 1,843 respondents in a 2021 online McKinsey & Co. survey said their organization had adopted AI in at least one function. Yet its business use is most often concentrated in customer service, product development, marketing, and sales, such as automatically answering common queries or identifying customers willing to buy additional goods.
Using OCI, businesses can train models on large amounts of written and spoken language, photos, video, and computer code, and then parlay those learnings into practical tools in business operations and scientific research.
Healthcare and related life sciences will be among the primary beneficiaries, said Huang, who appeared onstage with Oracle CEO Safra Catz.
In addition to expanding Oracle and NVIDIA’s partnership to supply processing power using in training large AI models in the cloud, the companies are teaming up to make software available on OCI so developers can maximize GPUs’ performance and apply them to new scenarios.
Oracle will offer OCI customers access to NVIDIA software tools and pretrained models called AI Enterprise, which lets developers build applications for recognizing speech, making recommendations, and interpreting medical images. OCI will also offer NVIDIA Clara, which are AI applications and development tools for healthcare.
OCI customers will also be able to run NVIDIA tools called RAPIDS Accelerator for Apache Spark, which can speed up data science processing and model training.