Oracle Corp.’s Cloud Infrastructure business is adding new bare-metal compute instances for high-performance computing workloads such as artificial intelligence and engineering simulations.

The bare-metal Oracle Cloud Infrastructure instances — meaning they ship without additional software, making them lean and thus potentially faster — announced Monday are an addition to the company’s Clustered Network. That’s a remote direct memory access network used to provide a cluster of computers with lower latencies and higher bandwidth.

Oracle reckons its new instances are capable enough that they will tempt “countless enterprises” that have so far been unable to migrate their existing HPC workloads to the cloud. Reasons given include the expense of running such operations in the cloud and a lack of fast enough hardware, both of which Oracle said its new offering can address.

“Organizations have struggled to find a cloud that can support new workload requirements while realizing cost efficiencies and flexibility that meet business goals and overall technology vision,” Oracle said in a release.

The company presented a list of new use cases for HPC cloud workloads that its new instances should enable, including car crash simulations, DNA sequencing and reservoir simulation for oil exploration, among others. For example, customers can now use data from the Oracle Database alongside powerful graphics processing units to run neural net AI training in the cloud.

The new instances are powered by Intel Corp.’s top-of-the-line Xeon processors and Mellanox Technologies Inc.’s network interface controllers.

“Having a supercomputer in your cloud infrastructure is quickly becoming a table stake for infrastructure-as-a-service providers,” said Holger Mueller, principal analyst and vice president of Constellation Research Inc. “With Oracle adding supercomputer capabilities to Oracle Cloud, it makes it more attractive to CxOs who want to power their next-generation applications in the cloud. Increasingly, these use cases require supercomputer capabilities across a variety of compute and storage resources, so getting the mix right is key.”

Photo: Håkan Dahlström/Flickr