Hewlett Packard Enterprise on Wednesday introduced machine learning platforms it said can enable users to develop and train AI models faster and at a greater scale.
HPE’s Machine Learning Development System integrates machine learning software platform, GPU and CPU chips and accelerators and networking to enable users to reduce the time they spend building and training machine learning models from weeks and months to days, according to HPE.
HPE Swarm Learning, meanwhile, is a machine learning framework for the edge or distributed sites.
Since most of the training of AI models occurs at a central location, developers and IT professionals have to move large volumes of data back and forth from repositories and this constant data exchange is further slowed by data privacy and data ownership requirements, according to HPE.
Swarm Learning enables users to train and harness models at the edge, rather than at the server level. The HPE AI system also enables users to use distributed data and build machine learning models, while preserving data governance and privacy, the vendor said.
The old and the new
Machine Learning Development System and Swarm Learning have been in the works for a while, said Frederic Van Haren, an analyst at the Evaluator Group.
The vendor started discussing its plans for HPE Swarm, in particular, in 2020, he said. Now, the product’s focus on the edge and blockchain security could help it gain traction with organizations looking for those components.
HPE’s Machine Learning Development System stems from its acquisition of determined AI in 2021, but “I don’t see the added value by HPE,” Van Haren said.
Determined AI provided an open-source deep learning platform on which data scientist can train models and share GPU resources.
Machine Learning Development System enables HPE to combine determined AI software with the HPE hardware, the vendor said.
The system includes a full software stack including determined AI tools now called the HPE Machine Learning Development Environment, plus the vendor’s Docker container technology, HPE Performance Cluster Manager and the Red Hat Enterprise Linux Operating System.
While both offerings could gain attention from government entities, labs, and large organizations, Van Haren said he does not see the vendor getting much interest from enterprises.
“HPE is known for selling hardware, not software or solution stacks,” he said.
Late to the market
HPE’s products may not be competitive in the machine learning market, said Andy Thurai, an analyst at Constellation research.
“HPE is late to the market with already many players crowding this market,” he said.
While the HPE Apollo 6500 Gen10 system is powerful enough to support faster machine learning model building and training, and the distributed training and automated hyperparameter tuning enable developers to train models more efficiently, the Machine Learning Development System “can only compete against private [clouds] which is a somewhat limited target space,” Thurai said.
“Most enterprises have adopted one of the public clouds for their model training already,” he said, referring to AWS, Azure or Google. “It will be an uphill battle to try to bring it back in-house.”
However, Swarm Learning may have some potential with enterprises, Thurai said.
“While the current ML model training and consumption needs to change to use this technology, there is a potential for it,” he said.
The AI tool enables enterprises to avoid the extensive process of redacting, cleansing, protecting and managing data before moving it back and forth to a particular location.
“The distributed ML model framework can be particularly useful in situations where edge locations can have volumes of datasets. Otherwise, this will be useless,” he said.
The Machine Learning Development System is priced on a cluster basis, according to HPE. The vendor has yet to make pricing information available for Swarm Learning. Both AI systems are available now.