At Oracle’s annual Oracle Open World summit in San Francisco last month, the Redwood Shores company revealed significant enhancements to Oracle Cloud Infrastructure (OCI), its bare metal compute service for enterprises. Following on the heels of those announcements, it today launched new OCI Generation 2 instances it described as tailor-made for “performance-sensitive” high performance compute (HPC) workloads such as machine learning and engineering simulations.

They’re powered by Intel’s Xeon processor lineup and Mellanox’s high-performance network interface controllers, and they slot under Oracle’s new clustered network offering, which makes available to customers low-latency, high-bandwidth RDMA networks with direct memory access to machine memory.

“In the past three years, Oracle has made significant progress in developing and improving high performance bare-metal offerings capable of providing on-premise customers with the options they need to move their HPC workloads to the cloud,” Clay Magouyrk, senior vice president of Oracle Cloud Infrastructure, said. “Our growing collaboration with industry-leading organizations helps ensure that Oracle Cloud Infrastructure continues to expand and offer the best performance at the lowest cost for the workloads that customers really need the cloud for.”

With clustered network setups, Oracle explained, customers can run workloads like DNA sequencing, car crash simulations, oil reservoir explorations, and more in the cloud cost-efficiently. If they’re already using Oracle Database, it’s a plug-and-play affair; cloud-stored datasets feed seamlessly into clustered network storage, where they can be used to train artificial intelligence (AI) models running on processors, high-end graphics cards, or a bespoke combination of silicon.

“With this new offering,” Oracle wrote, “[we] now provides large organizations with a complete set of solutions for any HPC workload, enabling businesses to capitalize on the benefits of modern cloud computing while enjoying performance comparable to on-premises compute clusters at a cost that makes sense for their business.”

The new Oracle Cloud Infrastructure instances join two others announced in October.

Early last month, Oracle debuted support for Nvidia’s HGX-2 platform, which sandwiches together 16 interconnected Nvidia Tesla V100 GPUs (with 0.5TB of memory) for a combined two petaflops of computing power. GPU-accelerated deep learning and HTPC containers in Oracle Cloud Infrastructure are available from the Nvidia GPU Cloud container registry, which notably supports Nvidia’s open source RAPIDS end-to-end data science pipeline.

More recently, Oracle and AMD jointly unveiled the first AMD Epyc processor-based instances on Oracle Cloud Infrastructure. Oracle claims that the Compute Standard E2 platform, which is generally available in bare metal and one, two, four, and eight-core virtual machine shapes, is the most cost-effective instance (3 cents per core hour) available on any public cloud.

Separately, the new instances leverage Oracle Cloud Infrastructure Gen 2’s spotlight features, including security enhancements, an improved autonomous database, and new integrated experiences for Oracle Cloud Marketplace, Oracle’s online store for Oracle Cloud business apps and professional services.

A few include a new Key Management Service (KMS) that affords admins greater control over data encryption, plus a Cloud Access Security Broker (CASB) that monitors and enforces secure configurations; distributed denial of service (DDoS) shielding; and a Web Application Firewall (WAF) designed to protect against attacks on the web. Also in tow (in preview) is Oracle Autonomous Transaction Processing, an autonomous database cloud service that allows teams to deploy and manage projects in discrete cloud-like environments on Oracle Cloud.

By the end of 2019, Oracle Cloud Infrastructure will expand to data centers in Toronto, Chicago, Sao Paulo, Zurich, Middle East, Mumbai, Seoul, Tokyo, and Osaka, the company revealed at Oracle Open World, up from the four regions in which it’s available currently.

In September, Oracle reported $6.61 billion in revenue from its cloud services and license support segment, up 3.2 percent from the prior year. The company, citing research from Deloitte, expects customers to spend more than $57.6 billion on machine learning by 2021, up from $12 billion in 2017.