Cray suggests supercomputers for the job
by Max Smolaks 25 September 2019
Venerable supercomputer maker Cray has published a report that, unsurprisingly, suggests IT teams need more data center hardware to tackle the growth in artificial intelligence workloads.
The company surveyed more than 300 IT professionals, and more than 70 percent said their companies already have AI applications in development or production.
At the same time, nearly 65 percent of respondents stated that they need to expand their on-premises systems to keep pace with increasing AI performance requirements.
More than 34 percent believe AI can be considered a “business critical” capability in their organization, or that it will become critical at some point during 2019. Another 41 percent believe AI will become critical to their business within the next three years.
The greater need for on-premises hardware could benefit high performance systems vendors like Cray, which has been making supercomputers since 1972.
But the same survey found that just 39 percent of organizations are currently using supercomputers, HPC systems or dedicated AI hardware to run their AI workloads – the rest rely on general purpose servers and workstations.
Intel, the world’s dominant CPU supplier, has long maintained that its Xeons will handle AI workloads just fine. The industry has always seen such claims with suspicion and has welcomed every new entry into the AI hardware field, whether from established players like Nvidia, Xilinx, Huawei, or startups like Ampere, Graphcore and Cerebras.
In August, Intel launched its own family of specialized AI silicon, built from the acquisition of Nervana Systems, perhaps the greatest testament to the fact that we can’t tackle AI with the standard hardware platforms.
“The research highlights that supercomputing plays an important role in enabling mainstream AI adoption in the enterprise,” said Frederick Kohout, senior vice president and CMO at Cray. “IT professionals see the need to expand their use of on-premises infrastructure, like supercomputers and HPC clusters, to meet the expected growth of business-critical AI applications.”
Nearly 50 percent of respondents said that overall, AI had a positive effect on their daily working experience in the past year.
For organizations that have implemented mainstream AI applications, 40 percent of respondents reported using on-premises systems for some of their machine learning work, 53 percent noted that their organization placed some AI workloads in the cloud, and laptops and workstations were still used for AI projects by 41.7 percent.
“Training for machine learning requires high performance computing, either on-premise or in the cloud, and there is a significant overlap between AI investment and systems for HPC and hyperscale,” said Addison Snell, CEO of Intersect360 Research. “One way or another, those interested in AI need to access a high-performance infrastructure.”
Of special interest to AI Business was the high level of educational activity among the AI professionals: 72 percent of respondents participated in one or more activities to educate themselves on AI in the past year. More than 48 percent attended AI conferences or received training, more than 41 percent participated in vendor webinars, about 40 percent took self-study courses and more than 35 percent downloaded and read reports on AI.