Hewlett Packard Enterprise (HPE) is offering support for demanding AI applications with a new cloud-based service underpinned by the company’s supercomputers.
The initial service is designed specifically to facilitate large language models (LLM), but the company says it plans to roll out similar AI services for domain- and industry-specific applications in climate modeling, healthcare and life sciences, financial services, manufacturing, and transportation.
All of these will fall under HPE’s GreenLake portfolio, but unlike the rest of GreenLake services, these will be hosted entirely in the cloud. The other services run either entirely on customer premises or in a hybrid on-prem/cloud environment.
The cloud infrastructure resides in a private-cloud data center in Canada, which could raise data-sovereignty concerns. HPE has not responded to a request for comment about that.
The cloud aspect of the new GreenLake for Large Language Models service is provided by HPE’s supercomputing hardware and AI software. The LLM is provided through a partnership with German AI startup Aleph Alpha whose Luminous software is pre-trained and optimised for analysing and processing data at scale. It also supports multiple languages: English, French, German, Italian, and Spanish.
The service is designed for enterprises that run most of their computing on-prem but need bursting capability, said Justin Hotard, executive vice president and general manager for HPC & AI at HPE.
The HPE supercomputers use the Cray programming environment and open-source tools for building and deploying models optimised for integrating unstructured data, specifically the HPE machine-learning development environment. This allows developers to rapidly train and scale models such as generative AI models.
“We have years of experience in scaling performance and tuning supercomputers to support any kind of compute and data intensive workload, and we're committed to making our supercomputers accessible for everyone to capitalise on AI,” he said. “As part of our move into the AI cloud market, we're focused on bringing services to a range of customers so they can build, customise, and deploy their own models.”
Announced at HPE Discover 2023 in Las Vegas this week, GreenLake for Large Language Models is available in North America starting at the end of 2023 and expected in Europe early next year, HPE says. HPE is initially selling the service direct to customers, but over time the company will evaluate potential other routes to market, Hotard said.
HPE is broadening its private-cloud offerings with the introduction of HPE GreenLake for Private Cloud Business Edition (PCB). This is a private cloud that can be deployed on-prem or at a colocation site, and the in-house IT team is responsible for managing the equipment.
That makes it different from HPE GreenLake for Private Cloud Enterprise (PCE), which was launched last year as a fully managed service. PCB is also limited to supporting virtual machines, while PCE also supports containers. In addition to self-managing VMs on-prem, PCB also allows for management within AWS and soon other public cloud providers, HPE says.
PCE has also been extended to let users provision workloads onto AWS, Google Cloud, and Microsoft Azure clouds and eventually support deployment on the Red Hat OpenShift platform.
HPE has expanded its partnership with colocation provider Equinix to allow pre-provisioned HPE private cloud systems (both PCE and PCB) to be deployed into Equinix data centers for customers around the world.
“We recognise that customers sometimes want their solutions self-managed and sometimes want us or other service provider partners to manage their solutions for them and increasingly want to have an option where these solutions are in pre-provisioned and pre-deployed data centers and ready to use,” said Bryan Thompson, vice president of HPE GreenLake Cloud Services Solutions.
Other Discover announcements include that some of HPE's software services will be available via the AWS Marketplace in addition to being sold through GreenLake, including the HPE NonStop Development Environment and HPE Fraud Risk Management. HPE GreenLake for Backup and Recovery has added support for Amazon RDS and Amazon EKS Anywhere.
HPE announced that its OpsRamp AIOps platform is available as a SaaS offering via GreenLake. OpsRamp will continue to be delivered through HPE services and by managed service providers globally as well as integration partners