It’s almost unimaginable to have a dialog within the high-performance computing (HPC) house (or any house for that matter) with out synthetic intelligence (AI) creeping into the dialogue. The know-how’s versatility has skyrocketed its total deployment, and as such, AI can seem to be a miracle remedy that’s come to resolve all computing issues.
After all, that is incorrect. Whereas AI is a helpful know-how, reducing by means of the hype is important if the HPC group desires to make use of AI to its full potential.
With this in thoughts, Intersect360 Analysis performed the 2024 HPC-AI Software program Survey. This complete survey of working environments, middleware, and end-user purposes stretches throughout 11 domains, together with the adoption of machine studying, AI, and huge language fashions (LLMs). The survey respondents got here from numerous sectors, with 68% coming from Business shoppers, 26% from Tutorial establishments, and 6% from Authorities.
This investigation into the real-world use of AI – together with the opposite subjects coated within the survey – sheds gentle on simply how helpful this know-how is and the place it is likely to be headed.
Intersect360 Analysis 2024 HPC-IA consumer survey questions and solutions. (Supply: Intersect360 Analysis)
AI Engagement and Utilization
One of many least shocking items of knowledge to come back out of the 2024 HPC-AI Software program Survey was the massive share of respondents who’re utilizing AI in some type. Answering the query, “Is your group partaking with AI proper now, in any of its types?” 92% of respondents replied, “Sure.” As one would think about, AI is in every single place proper now.
That stated, AI is such a flexible device that merely asking if a company is utilizing it isn’t sufficient. Digging deeper, the survey additionally requested respondents how they had been at present utilizing AI and what they deliberate to make use of it for sooner or later.
Respondents had been requested to answer to the next assertion: “We’ve built-in AI or Machine Studying into our HPC purposes, comparable to for computational steering or simulation optimization.” Out of these surveyed, almost 75% both agreed or strongly agreed with that assertion.
Conversely, respondents had been additionally requested to answer to the next assertion: “AI/ML are separate from our HPC initiatives and purposes.” A little bit greater than 30% responded both “agree” or “strongly agree” to this assertion. This comparability exhibits simply how deeply built-in AI/ML instruments are in HPC purposes. Organizations are discovering a variety of use for this know-how of their HPC operations.
Seeking to the long run, it might appear that even these at present not adopting the know-how plan will accomplish that quickly. Respondents had been requested to answer to the next assertion: “We’ve plans to combine AI/ML into our HPC extra carefully, however we’re not there but.” Almost 65% of respondents both agreed or strongly agreed with that assertion, displaying that even those that should not at present integrating AI into their HPC purposes no less than have a need to take action.
Generative AI and LLMs
Whereas AI has all kinds of makes use of, generative AI (GenAI) and LLMs are a number of the most mentioned purposes of this know-how. As such, Intersect360 Analysis devoted a complete part of the 2024 HPC/AI Software program Survey to those purposes.
The survey discovered that almost 75% of respondents agreed with the assertion, “We’re actively utilizing generative AI.” Almost 55% of respondents additionally agreed with the assertion, “We’re contemplating constructing our personal customized LLM.” Moreover, almost 65% of respondents agreed they “anticipate generative AI to decrease our prices.”
Whereas that is solely a small snapshot of the general research, it paints an image of what organizations are on the lookout for from their GenAI/LLM deployments. It seems that most respondents anticipate this know-how to avoid wasting them cash. Furthermore, constructing customized LLMs is harder than utilizing off-the-shelf merchandise. The solutions from respondents present that clients need management over their LLM deployments – or, on the very least, they suppose they do.
The assertion furthers this sentiment that 42% of respondents might be retaining their LLM deployment at their very own information heart colocation for improvement and design, 34% for testing, 35% for mannequin coaching, and 31% throughout manufacturing.
For a full presentation of the survey outcomes (with slides) , see the total Intersect 360 Analysis Webinar.
This text first appeared on sister website HPCwire.