When it bought Cray back in May 2019, HP Enterprise hinted at offering HPC systems as a service. Now it is delivering on that with the introduction of HPE GreenLake cloud services for HPC.
HPE has made a lot of headway with its GreenLake program, the pay-per-use model created in response to the popularity of cloud service providers. It lets customers pay as if they are buying a cloud service, but it’s provisioned using infrastructure deployed at customer sites or in colocation facilities. Up to now it’s been used in standard IT applications, like app and Web serving or databases.
The high-performance computing service is designed for enterprises to support demanding compute and data-intensive workloads such as AI and ML. The new offering removes the complexity and cost associated with traditional HPC deployments by delivering fully managed, pre-bundled services based on purpose-built HPC systems, software, storage, and networking that come in small, medium, and large options.
Customers order it through a self-service, point-and-click portal to choose the configuration they need for their workloads. The company claims it can speed up deployment of HPC projects by up to 75% and reducing capital expenditures by up to 40%.
Enterprises can deploy the services in data-center environments, whether on-premises in their own enterprise or in a colocation facility, and gain fully managed services that allow them to pay only for what they use.
This is a big deal because HPC has a very high financial barrier to entry: hardware for servers, networking, storage, and cooling, plus special power supplies. A claimed 40% reduction in expenses makes it available to a much wider market, HPE says.
Remember, GreenLake (and Dell On Demand and Lenovo TruScale Infrastructure Services) are the alternative to up-front hardware acquisition. Instead of buying outright, the customer pays a monthly usage fee—a lease—and has the option of the vendor managing the hardware. When the lease is up, instead of customers having to dispose of the hardware, the vendor takes it back.
HPE is going all out to support GreenLake. In 2019 it announced plans for all HPE hardware to be available through GreenLake by 2022. HPC seems an odd choice because unlike traditional IT apps, it runs full tilt constantly. In a metered scenario that can add up fast.
But Peter Ungaro, senior vice president and general manager of High Performance Computing & Mission Critical Solutions at HPE, says there are two drivers here that help to save money.
“First is that smaller HPC and AI systems don’t typically run at full utilization as the demand for them is less than the large systems,” he said via email “Also, customers may want to prototype, say an AI application, and then that application usage will scale over time as they move into production and have larger data volumes.”
“Secondly, we also have customers that need excess ‘burst’ capacity from time to time that wouldn’t normally be utilized on a day-to-day basis. Both of these types of customers will benefit greatly from the TCO of this solution,” he said.
Also, because it’s a managed service, customers who lack the skills, facilities or budget to use advanced HPC or AI systems otherwise can take advantage of this technology just like larger corporations or government institutions can, he added.
Commercial uses might involve variable workloads where HPC, AI, and ML techniques help core business processes, as in healthcare diagnostics, fraud detection in financial services, or computational fluid dynamics, he said.
While hardware is at the center of GreenLake, it involves a lot of software. HPE GreenLake Central is a platform for customers to manage and optimize their HPC services. HPE Self-service dashboard enables users to run and manage HPC clusters on their own, without disrupting workloads, through a point-and-click interface, and HPE Consumption Analytics provides analytics of usage and cost based on metering.
For the HPC hardware, HPE also offers HPC, AI & App Services, which standardizes and packages HPC workloads into containers, making it easier to modernize, transfer, and access data.
HPE plans to expand HPE GreenLake cloud services for HPC to additional technologies, which includes Cray-based compute, software, storage, and networking in the future.
Initial pre-bundled offerings for HPE GreenLake cloud services for HPC will be generally available globally in the spring of 2021.