HPE, during the NVIDIA GTC event, unveiled enhancements to its extensive AI-centric lineup, aiming to enhance the implementation of generative AI (GenAI), deep learning, and machine learning (ML) applications. These updates encompass the launch of two fully integrated GenAI solutions developed jointly by HPE and NVIDIA.
HPE stands as a global leader in edge-to-cloud solutions, enabling businesses to expedite results by extracting value from their data across all environments. With a rich history of envisioning the future and driving innovation, the company offers distinctive, open, and intelligent technology services on a subscription basis.
Its portfolio covers Cloud Services, Computing, High Performance Computing & AI, Intelligent Edge, Software, and Storage, ensuring a uniform experience across various cloud platforms and edge devices. This approach aids customers in innovating their business models, enhancing engagement, and improving operational efficiency.
A sneak peek at the upcoming HPE Machine Learning Inference Software, a reference architecture for enterprise retrieval-augmented generation (RAG), and the commitment to creating future products based on the new NVIDIA Blackwell platform. Antonio Neri, the president and CEO of HPE, emphasized the necessity for hybrid solutions to fully realize the potential of GenAI and effectively manage the entire AI lifecycle.
He highlighted the importance of designing solutions that can seamlessly integrate on-premises, colocation facilities, and public cloud environments for training and tuning models, as well as inferencing at the edge. Neri underscored the significance of collaboration between the company and NVIDIA, noting their history of joint innovation. He expressed their commitment to providing AI software and hardware solutions tailored to accelerate the development and deployment of GenAI, from initial concept to full-scale production.
Jensen Huang, the founder and CEO of NVIDIA, highlighted the transformative power of Generative AI in extracting insights from data sourced from connected devices, data centers, and clouds, thereby driving innovation across various industries. He emphasized the expanding partnership with HPE, aiming to empower enterprises to leverage their data effectively.
This collaboration aims to enable enterprises to develop and deploy new AI applications that can revolutionize their operations and drive unprecedented productivity. As unveiled at SC23, HPE has introduced its latest supercomputing solution tailored for generative AI, now open for ordering by organizations in need of a preconfigured, pretested full-stack solution for developing and training large AI models.
This purpose-built solution aims to expedite GenAI and deep learning projects, featuring a turnkey design powered by NVIDIA and capable of accommodating up to 168 NVIDIA GH200 Grace Hopper Superchips. It offers a comprehensive AI/ML software stack, enhancing the model development process for large enterprises, research institutions, and government bodies.
This turnkey solution, inclusive of installation and setup services, targets AI research centers and large enterprises, promising improved time-to-value and accelerated training speeds of 2-3X for various projects like LLMs, recommender systems, and vector databases. Unveiled during the Discover Barcelona 2023 event, HPE introduced its latest enterprise computing solution for generative AI.
This solution is now accessible to customers either directly or through HPE GreenLake, featuring a flexible and scalable pay-per-use model. Developed in collaboration with NVIDIA, this pre-configured fine-tuning and inference solution aims to minimize ramp-up time and costs by providing the necessary compute, storage, software, networking, and consulting services for organizations to develop GenAI applications.
This AI-native full-stack solution empowers businesses with the speed, scale, and control required to customize foundational models using their private data and deploy GenAI applications within a hybrid cloud model. Highlighting a high-performance AI compute cluster and software from both the company and NVIDIA, this solution is well-suited for lightweight fine-tuning of models, RAG, and scale-out inference.
The time required for fine-tuning a 70 billion parameter Llama 2 model using this solution decreases linearly with the node count, taking merely six minutes on a 16-node system. This enhanced speed and performance allow customers to achieve faster time-to-value by enhancing business productivity with AI applications such as virtual assistants, intelligent chatbots, and enterprise search.
The solution, powered by HPE ProLiant DL380a Gen11 servers, is pre-configured with NVIDIA GPUs, the NVIDIA Spectrum-X Ethernet networking platform, and NVIDIA BlueField-3 DPUs. It is further bolstered by HPE’s machine learning platform and analytics software, NVIDIA AI Enterprise 5.0 software featuring the new NVIDIA NIM microservice for optimized inference of generative AI models, along with NVIDIA NeMo Retriever and various other data science and AI libraries.
To tackle the AI skills gap, HPE Services experts will assist enterprises in designing, deploying, and managing the solution, including the application of appropriate model tuning techniques. HPE and NVIDIA have joined forces on software solutions designed to help enterprises advance from AI and ML proofs-of-concept to full-fledged production applications.
The collaboration introduces the Machine Learning Inference Software as a technology preview for its customers, enabling them to deploy ML models rapidly and securely at scale. This new offering integrates seamlessly with NVIDIA NIM, offering foundation models optimized by NVIDIA through pre-built containers.
To address the needs of enterprises seeking to quickly develop and deploy GenAI applications with sensitive data, HPE has created a reference architecture for enterprise RAG. This architecture, available now, is based on the NVIDIA NeMo Retriever microservice architecture and includes a robust data foundation from HPE Ezmeral Data Fabric Software and HPE GreenLake for File Storage. The reference architecture serves as a guide for businesses looking to create tailored chatbots, generators, or copilots.
The solution combines a range of open-source tools and solutions from HPE Ezmeral Unified Analytics Software and HPE’s AI software, including HPE Machine Learning Data Management Software, HPE Machine Learning Development Environment Software, and the new HPE Machine Learning Inference Software, to support data preparation, AI training, and inferencing.
This suite of AI software is available on both HPE’s supercomputing and enterprise computing solutions for generative AI, ensuring a consistent environment for managing GenAI workloads. HPE has plans to create upcoming products using the recently revealed NVIDIA Blackwell platform. This platform includes a second-generation Transformer Engine designed to speed up GenAI workloads. More information about HPE’s future products, which will showcase the NVIDIA GB200 Grace Blackwell Superchip, the HGX B200, and the HGXB100, will be disclosed at a later date.
