HPE offers up HPC for AI training with GreenLake for Large Language Models

HPE has launched a new artificial intelligence (AI) offering for its GreenLake as a service platform at its annual conference, HPE Discover.

Dubbed HPE GreenLake for Large Language Models (LLMs), the high-performance computing (HPC)-powered offering allows customers to privately train and deploy large-scale AI projects, which has previously been a challenge. 

Access to supercomputers outside of very specific industries is limited and often expensive, and unless you’re an AI company, using it for AI training hasn’t typically been a priority.

Justin Hotard, executive vice president and general manager for HPC and AI at HPE, told a pre-announcement press conference: “Supercomputers today are costly and complex to adopt and manage. Supercomputers also require unique data center capabilities for power and cooling. And until now, supercomputers have not been available on demand in a consumption model.”

HPE GreenLake for LLMs hopes to solve this, the company said. It will give organizations the opportunity to train their own LLMs using HPE’s Cray XD supercomputers and Nvidia H100 GPUs on demand, without needing to invest in additional infrastructure or skills.

HPE GreenLake for LLMS: Leaning on German engineering

Underpinning the GreenLake for LLMs proposition is technology developed by German startup Aleph Alpha. 

The Heidelberg-based company’s flagship Luminous software is a pre-trained LLM optimized for analyzing and processing data at scale.

Explaining the decision to work with Aleph Alpha, Hotard said Luminous was trained using HPE supercomputers and existing HPE software.

“[Luminous] has already been implemented by various organizations in healthcare and financial services, and in the legal profession as a digital assistant,” he added. 

“By enabling on-demand access to Luminous, enterprises can also take advantage of one of its multiple languages, which include English, French, German, Italian, and Spanish.”

HPE has also taken into account data privacy and data sovereignty concerns while developing the product. 

GreenLake for LLMs only uses the customer’s data for their language model training, eliminating fears about exposing trade secrets, for example.

The company foresees a relatively broad potential customer base. Hotard told ITPro that it will be primarily useful to enterprises, including larger public sector bodies, but that “there are certain startups that can take advantage of this to accelerate their offerings and services” – notably those in the AI space.

A true private cloud offering

Unusually for an HPE GreenLake product, this is not a hybrid or on-premises managed product, but is hosted entirely in a remote private cloud housed in a facility in Canada. 

This means that, for now at least, customers won’t be bringing any of the associated hardware into their data centers.

The company is also hopping aboard the eco-friendly data center train for this project, with the company planning for this initial instance to be 100% carbon neutral.

Hotard said: “The objective [for] all of our cloud deployments … is to provide 100% carbon neutral offering to our customers. The first one is using nearly 100% renewable power. And one of the benefits of liquid cooling is you can actually use the … wastewater, the heat comes out of the heat of water and reuse it.

“We have examples of that in other supercomputing installations and we're leveraging those learnings and expertise for this cloud deployment as well.”

Anyone getting excited to try HPE GreenLake for LLMs for themselves may wish to temper their expectations, though. 

While the firm is accepting orders now, you will effectively be joining a waiting list. Additional availability is expected by the end of the year, but will be limited to North America. 

Availability in Europe is expected to follow early next year, with a question mark over rollout in the rest of the world. Pricing is also currently unavailable.