At last week’s Hewlett Packard Enterprise (HPE) Discover, the company announced its entry into the AI cloud market by expanding its HPE GreenLake portfolio. In 2019, HPE acquired supercomputing vendor Cray Inc. At the time, HPE CEO Antonio Neri said, “The answers to some of society’s most pressing challenges are buried in vast amounts of data.”
However, supercomputing was primarily reserved for academic researchers and well-funded companies. As businesses begin to embrace AI, the need for more advanced computing infrastructure has skyrocketed. Cloud computing services promise unlimited bandwidth, but they are not always the most efficient or cost-effective way to train large-scale models. A new HPE AI cloud service called HPE GreenLake for Large Language Models (LLM) enables organizations to train, tune, and deploy AI models at scale while ensuring security and compliance for responsible and sustainable AI deployment. can promote.
HPE GreenLake for LLM runs on HPE Cray XD supercomputers. This new his GreenLake product eliminates the need for customers to purchase and manage a potentially expensive and complex supercomputer. It leverages the HPE Cray programming environment and HPE’s AI and ML software suite to provide comprehensive tools for optimizing HPC and AI applications, training large-scale models, and managing data.
HPE has partnered with German AI startup Aleph Alpha to provide a ready-to-use Large Language Model (LLM) as part of an on-demand, multi-tenant supercomputing cloud service. HPE GreenLake for LLM provides access to Aleph Alpha’s Luminous, a 13 billion parameter pre-trained large-scale language model available in multiple languages.
HPE GreenLake for LLM enables companies to use their own data to train, tune, and deploy large-scale AI models using HPE AI software and supercomputers. By using company databases, articles, and industry-specific data, companies can base their models on factual data and relevant user context, eliminating problems such as hallucinations. Hallucinations (also called confabulations or delusions) in artificial intelligence are things that the AI responds to with confidence that are inaccurate or do not appear to be justified by the training data.
HPE hasn’t announced specific support for OpenAI, but it’s clear that OpenAI is deeply involved with Microsoft. HPE products offer another option in the market for businesses looking for alternatives to Microsoft and OpenAI.
What does that mean for businesses?
One of the common challenges with training LLMs is that lack of computing power can cause LLMs to not run or hang in action. Insufficient computing power slows down the time it takes to train a model, and restarts increase the cost of the model. Enterprises need a way to run HPC-focused machine learning applications without building and managing on-premises supercomputers.
As mentioned above, another problem with using open-source LLMs today is that LLMs are general-purpose models trained on large corpora of internet data and are subject to potential inaccuracies and specific Lack of industry data. Supercomputing for LLMs can address the unique requirements and challenges of different domains by linking industry data with more general LLMs. HPE said it plans to launch a series of industry and sector-specific AI applications in the future. These applications address a variety of fields such as climate modeling, medical and life sciences, financial services, manufacturing, and transportation.
Each industry has a huge amount of specific data, so it makes sense to work on these areas. These industries also have use cases where the economic benefits of successful outcomes far outweigh the cost of AI supercomputing. Speed is another area of interest for AI supercomputing. For example, detecting fraud and security threats in financial services transactions requires a combination of large amounts of data processing in a compressed timeframe.
You may be wondering why you can’t use a general-purpose cloud product for AI. The answer is yes for certain workloads, but AI workloads are not the same. HPC workloads require an AI-native architecture specifically designed to handle single, large-scale AI training and simulation workloads with full computing power. Unlike parallel workloads, HPE GreenLake for LLM supports AI and high-performance computing jobs on hundreds or thousands of CPUs or GPUs simultaneously.
This infrastructure design enables more effective and efficient AI training, resulting in more accurate models that accelerate enterprise problem solving. This doesn’t mean hyperscalers rest on their laurels. Google recently announced its A3 supercomputer with Nvidia H100 GPUs, and Amazon recently updated its HPC7g instances. However, HPE’s offerings should be particularly attractive to his existing HPE GreenLake customers who are building hybrid cloud strategies and have HPC requirements.
Sustainability is another issue in the AI environment. Most of the large companies that Lopez Research consults with define sustainability metrics, add ESG monitoring solutions, and actively explore ways to reduce energy consumption. The processing power required to support new AI models contradicts these goals in many ways. As a result, businesses have come to rely heavily on infrastructure providers to provide solutions that address sustainability concerns.
All hyperscaler cloud companies design different infrastructure, cooling and optimization solutions to address this issue. HPE is also in this camp. GreenLake for LLM runs in a colocation facility that uses nearly 100% renewable energy, according to the company. HPE GreenLake services for LLM require specialized supercomputing data centers that are optimized for power and cooling. This means HPE GreenLake for LLM will be available in North America by the end of 2023, followed by Europe in early 2024. In North America, HPE ensures that our services are in line with environmentally friendly practices.
However, not all AI use cases require supercomputing power. For businesses that don’t need supercomputing power, but want servers optimized for AI workloads, HPE updated his ProLiant Gen11 servers with 4th Generation Intel Xeon Scalable CPUs to boost performance for AI inference tasks. improved.
It’s exciting and amazing to see HPE bring new as-a-service computing products to market during the Discover conference. HPE’s move aims to revolutionize the AI landscape by making HPC accessible to a wider range of organizations. While this type of service seems best suited to large enterprises with the money to pay for massive amounts of data and on-demand supercomputing, the market is still taking leaps and bounds. Even if large enterprises could afford to build AI infrastructure at scale, there is a well-known shortage of NVIDIA GPUs. You can’t train a model without the right hardware. It will be interesting to see what kinds of companies adopt these new services and what workloads the increased processing power enables.
follow me twitter Or LinkedIn. check out my website.
