Nvidia was very busy on the conference circuit during November, visiting events hosted by Microsoft, Amazon, and most recently, Hewlett Packard Enterprise (HPE). The not-so-surprising common theme among these events has been how to best support enterprise customers as they increase their development and use of large language models and generative AI applications.
At HPE Discover this week in Barcelona, Spain, HPE and Nvidia announced an expanded strategic collaboration to create a co-engineered, pre-configured AI tuning and inferencing solution so that enterprises can leverage their private data to customize foundation models and production applications for deployment from edge to cloud.
The partners said they aim to deliver “full-stack, out-of-the-box AI solutions” that integrate HPE Machine Learning Development Environment Software, HPE Ezmeral Software, HPE ProLiant Compute and HPE Cray Supercomputers with Nvidia’s AI Enterprise software suite and NeMo large language model (LLM) framework.
The need for such a full-stack approach will be driven by organizations that will demand complete hardware and software solutions to help them drastically reduce the complexity of deploying “computation-intensive” AI applications as they move from the cloud-native architecture era to the “AI-native” era, said Evan Sparks, vice president and general manager of AI solutions and supercomputing cloud at HPE.
“AI requires a fundamentally different architecture, because the workload is fundamentally different from the classic transaction processing and web services… that have become so dominant in computing over the last couple of decades,” he said. “And if you think about this in terms of cloud-native architectures, where we found ourselves over the last decade re-architecting applications to fit into the cloud, we think the next decade is going to require full-stack thinking from hardware all the way up to software layers as organizations lean into the deployment of AI applications.”
Manuvir Das, vice president of enterprise computing at Nvidia, was one of the executives representing Nvidia at HPE Discover, and he added that the expanded collaboration between the two companies and the new solution announced at the event reflect how the market for AI applications is rapidly maturing.
“From Nvidia’s perspective, we have worked on AI for years, but the journey always was that you go to a customer, you try to understand the industry they're in, and then you have to convince them that there's an interesting AI use case for them, and that was most of the conversation,” Das said. “What's happened is now it’s become obvious that there's an AI use case that applies to everybody. It doesn't matter what industry your company is, it doesn't matter what your job function is. Instead of just one use case, there's just hundreds of teams all across the company doing work now on GenAI. So that immediately suggests you need a platform [approach].”
The resulting “enterprise computing solution for generative AI” platform includes the following:
● Purpose-built and optimized for AI: A rack-scale architecture featuring market-leading HPE ProLiant Compute DL380a pre-configured with Nvidia L40S GPUs, BlueField-3 DPUs, and its Spectrum-X Ethernet Networking Platform for hyperscale AI. The solution was sized to fine-tune a 70 billion-parameter Llama 2 model and includes 16 HPE ProLiant DL380a servers and 64 L40S GPUs.
● HPE AI software: HPE Machine Learning Development Environment Software with new generative AI studio capabilities to rapidly prototype and test models, and HPE Ezmeral Software with new GPU-aware capabilities to simplify deployment and accelerate data preparation for AI workloads across the hybrid cloud.
● Nvidia AI software: The company’s AI Enterprise accelerates production AI development and deployment with security, stability, manageability and support. It offers the Nvidia NeMo framework, guardrailing toolkits, data curation tools and pretrained models to streamline enterprise generative AI (GenAI).Sparks also said HPE Services will aso support new AI services to take customers through every step of the journey from GenAI and LLM discovery to implementation. These services are supported by new Global Centers of Excellence for AI and Data now open in Spain, United States, Bulgaria, India and Tunisia.
The latest announcements came after HPE earlier in November announced at the SC23 supercomputing conference a turnkey supercomputing solution powered by Nvidia for large enterprises, research institutions and government organizations to develop and train foundational models–otherwise known as the first phase of the enterprise AI lifecycle. The HPE Discover announcement of a new enterprise computing solution for generative AI is focused on the later phase of that lifecycle–tuning and inferencing. Sparks said the enterprise computing solution for generative AI will be orderable in the first quarter of 2024.