The increasing maturity of artificial intelligence is the most important business trend I’ve seen in many years, with profound implications across all industries. Every business and IT leader should prioritize answering one question: How can we harness the power of AI to improve our organization’s competitiveness and profitability?
If the meetings I’m having with those leaders are any indication, that’s exactly what they’re doing. Virtually all of them at least include AI in whatever the organization is trying to achieve. In fact, many of my conversations are squarely focused on AI.
I recently wrote about one of the most popular use cases for AI, customer service. I noted that many businesses, recognizing that service is the ultimate differentiator, have identified opportunities to leverage AI technology to take better care of their customers, especially in the call center.
RELATED: What is high performance computing power?
AI Requires HPC-Level Computing Power
But there’s another issue — separate from but critically related to AI — that is becoming increasingly urgent as more businesses begin their AI journeys. That is the question of how organizations will acquire the computing resources necessary to power the kinds of outcomes they’re seeking from artificial intelligence.
AI, which often requires running computations on gigabytes of data, needs enormous computing power compared with ordinary workloads. Businesses have discovered that they can’t effectively operate AI-driven applications with their existing data centers or cloud environments. In fact, AI requires so much computing muscle that merely adding another couple of servers or marginally increasing the organization’s cloud budget won’t be enough.
READ MORE: Find out how AI is helping organizations improve customer service.
What’s required for AI workloads is the kind of power that comes from high-performance computing, in the form of graphical processing units. In fact, GPUs are replacing traditional CPUs (central processing units) in our AI-driven world. Businesses can access GPUs by purchasing them and installing them on-premises or via the cloud, but we’re seeing most organizations choose the former approach.
That’s mainly for two reasons. The first is to eliminate the latency that is inevitable in cloud environments as data moves between cloud data centers that may be located around the world. That kind of latency is barely noticeable with day-to-day computing needs, so in most industries it hasn’t been much of a barrier to cloud adoption. But with the gigabyte-level computing demands of AI, the latency of cloud environments can become pronounced.
Click the banner below to learn how to sharpen your edge with high performance computing.
An even bigger challenge is cost. If you were among those who thought a cloud migration would mean big savings, only to wince when you saw those first monthly bills roll in, then you can imagine what those bills would be like when you need the kind of cloud-based computing power necessary for AI.
That said, smaller-scope AI projects can often be run in the cloud. A bank we worked with recently, for example, deployed a new, AI-powered, cloud-hosted call center and is perfectly happy with the results.
DISCOVER: Learn how to deploy AI and analytic solutions to advance your business.
Why AI Requires GPUs
Bigger projects, though, will likely require on-premises GPU hardware. This can be acquired as part of traditional data center technology, with clusters of GPUs included in the box alongside conventional CPUs or as external GPU clusters that come complete with attached network and storage (just like a hyperconverged infrastructure setup), essentially taking the place of a traditional data center.
Purchasing an external GPU cluster is perhaps the best way an organization can acquire the most elite level of processing power, and it is best suited for those that have large needs right now or ambitious projects planned. These organizations have a few options in terms of how they buy such clusters, which come in groups of eight: A business that needs more can tie two or more clusters together or buy a “superpod,” which includes 256 GPUs.
This is no small investment, but these clusters come with the high-performance networking and low latency that organizations need for significant AI-based projects, as well as the software necessary to compute for AI.
Few organizations have a clear view of the best approach right from the beginning. Typically, the path forward emerges from a series of conversations with a partner that has seen many of these projects from start to finish.
CDW is here for you when you’re ready to have that conversation.
