Artificial intelligence is grabbing many of the GPU headlines. But a host of other applications, from business intelligence to data-intensive analytics, also underscore the growing need for GPUs.
GPUs are essential to the success of your organization. In your CEO’s eyes, so is winning the talent war, finding new growth opportunities, and transforming into a digital organization. Within these priorities where do GPUs for the data center sit? How should CIOs convince their boss to significantly increase investments in GPUs and related software?
To begin with, AI and machine learning are reshaping the future of businesses, in much the same way as big data did a few years ago. CEOs are grasping the potential of machine learning and AI for everything from anticipating customer needs, to lean manufacturing, to building blockchain platforms. Machine learning and AI apps require the massively parallel processing capability of GPUs. Where a typical CPU server may have 10, 20, or 25 cores, a GPU server can have over 5,000. In fact, the latest TOP500 supercomputer rankingsOpens a new window reported that for the first time in history, more flops were added by GPUs than CPUs.
But the advantages of GPUs don’t stop with AI. Enterprises face all sorts of increasingly data-intensive analytics problems—and GPUs are ideal to address these as well. The dense computational power of a GPU allows it to query billions of rows of data in milliseconds. It’s akin to having supercomputing power at your fingertips for general purpose applications, but without the exorbitant costs typically associated with supercomputers.
For example, a telecommunications enterprise we know of, understands the importance of GPUs for data-intensive analytics. Monitoring the reliability of their vast network is a gargantuan task involving billions of records that stream in continuously. Prior to implementing a GPU-based analytics solution, queries would take hours, meaning operational analysts had no real-time analytics visibility into network reliability. Now the same queries happen in real time, enabling the carrier to deploy corrective action to performance issues as soon as those problems are detected.
The power of GPUs, in fact, has multiple ramifications—benefits that create huge advantages in cost and competitiveness. First, GPUs provide orders of magnitude processing speedups. These increases provide not only quantitative but also qualitative leaps, as analysts are able to ask questions never thought of before.
Data center operations is the third area of value. The performance of thousands of CPU cores can be replaced by only 10 or 20 GPU nodes, and perhaps only a few servers, versus hundreds. The savings in data center space, power, and human resources can more than offset the higher per-unit cost of GPU cards. Fourth, GPUs are highly versatile in application, meaning they can be applied to general purpose analytics, visualization, deep learning, and other subcategories of AI. This versatility allows GPUs to satisfy the organization’s evolving next-generation needs, wherever those needs take it. Even if you adopt GPUs for deep learning, you can still use them for general purpose analytics between deep learning cycles.
If your business is concerned with staying abreast of the exponential growth in computing, the time will soon come—if it hasn’t already—to consider including GPUs in your analytic ecosystem. Cloud services are perhaps the most cost-effective way to tap into GPU capabilities; a thorough needs analysis will help you make that decision. Any way you go, it’s important to inform your top executives of all the capabilities that GPUs can provide, not just the ones that garner the headlines. Who knows? The improvements in business agility, competitiveness, and productivity just might pay for the next company picnic others for years to come