As AI will get greater and higher, it requires way more than mundane commodity compute, community, and storage infrastructure. GPUs are just one a part of this – AI workloads demand a finessed optimization of uncooked compute, information throughput, and energy consumption. Whether or not consumed as a cloud service, footprinted on-premises, or a hybrid of each, AI infrastructure is a crucial piece of the strategic puzzle that enterprise expertise leaders face. The selections tech leaders make about AI infrastructure will decide whether or not their enterprises change into AI leaders or laggards. Forrester defines AI infrastructure options as: Laptop techniques and cloud companies designed to maximise the efficiency of AI workloads: information preparation, mannequin coaching, and inferencing.
In our recently-released Forrester Wave: AI Infrastructure Options, Q1 2024, we seemed on the choices of 12 distributors on this area: Alibaba Cloud, Amazon Internet Companies, Cerebras Methods, Dell Applied sciences, Google, Graphcore, Hewlett Packard Enterprise, IBM, Lenovo, Microsoft, NVIDIA, and Oracle. Listed here are three of the important thing insights that emerged over the course of researching and penning this newest analysis:
Assume by way of workloads. Though GenAI has launched new methods to leverage AI, the three core AI workloads have remained the identical: information preparation, coaching, and inferencing. For a lot of organizations, it is going to be essential to leverage totally different suppliers for every of the workloads they’re tackling. For instance, with regards to information administration, a company could select a vendor with a robust on-premises answer, however for inferencing they could choose a big cloud service supplier.
Match your infrastructure to your computational wants. GenAI has elevated reliance on GPUs, which partly explains the astronomical development of some chip makers since we revealed our first wave on AI Infrastructure in December 2021. Though GPUs have been essential to GenAI’s success, they might not be essential to the success of your AI technique. Throughout the three core AI workloads, computational necessities fluctuate considerably. On the subject of coaching workloads, deep studying fashions like pc imaginative and prescient and huge language fashions require entry – both via the cloud or on-premises – to chips optimized for AI (usually GPUs), however predictive fashions could not profit from such chip architectures.
Plan for a way you’ll combine a brand new answer with current instruments. Earlier than committing to an answer, enterprises should perceive how they’ll incorporate a vendor’s AI infrastructure administration layer with their very own current expertise infrastructure administration instruments. AI infrastructure comes with administration software program to assist operations professionals monitor the system, management entry, allocate utilization, and provision/deprovision infrastructure to optimize prices. If an enterprise has already standardized on a vendor’s IT infrastructure, then utilizing that vendor’s AI infrastructure might be enticing from a administration perspective.
As with all expertise purchases, no single vendor will likely be proper for all enterprises and the best performers in our analysis could not essentially be the only option in your group. Forrester purchasers can learn the total report on our web site. You possibly can simply obtain the spreadsheet and customise the outcomes in your wants and priorities.
Forrester purchasers can focus on these findings – or AI infrastructure usually – by scheduling a steering session with me.
Aaron Suiter contributed to the content material of this weblog put up.
