Meeting Business Needs for Deep Learning in the Modern Data CenterJuly 2, 2018
With a data center in Los Angeles as well as other major metropolitan centers, Telehouse must stay at the forefront of the methods in which AI and deep learning neural networks are shaping the data center needs of the present and future. As the use of AI in the data center becomes more prevalent, the number of enterprise and hyperscale data centers that utilize AI and deep neural networks (DNNs) for massive amounts of data are growing.
Leveraging neural networks is increasingly seen as a fundamental part of digital transformation. It’s growing prevalence can be seen in a recent Information Week article explaining how it is being applied in marketing, retail, finance, and operations management across almost every sector.
Because neural networks use vast amounts of data, they require servers capable of extreme amounts of data computations in record time. Consequently, GPUs designed to enable this level of computational speed and volume are quickly being developed and adopted by data centers around the world.
Data centers that support these new high-performance GPU-based servers can deliver greater efficiency and performance and use less power for advanced workloads while decreasing the data center footprint and power consumption needs. For example, Nvidia’s new single server capable of two petaflops of computing power does what currently takes hundreds of servers networked into clusters. The leading GPU developer’s DGX-2 system is aimed primarily at deep learning applications.
While hyperscale data centers have been the traditional users of neural network-focused GPUs, collocation providers are increasingly partnering with major cloud providers that make this capability part of their offering. They can then offer this capability in their data centers for clients in need of providing their developers with cloud infrastructure services that enable them to build AI features into their own applications. This use is prevalent for companies that are in need of High Performance Computing (HPC) for big data.
The use of cloud hardware in the data center that is designed for neural-network training and inferencing continues to accelerate with Microsoft using FPGAs to accelerate these workloads.
Google Cloud Platform’s custom chips for ML known as TPUs or TensorFlow Processing Units are an application-specific integrated circuit (ASIC) developed by Google specifically for neural networks.
In fact, Deloitte Global predicts that 25 percent of all chips used to accelerate machine learning in the data center will be FPGA and ASICs. Data centers in Los Angeles that have close ties to these providers can now offer their enterprise clients the support needed to utilize these features for highly specific workloads. A recent Forbes article shows how the three major cloud providers are offering machine learning as a service (MLaaS) offerings and support providers like Telehouse are helping enterprises make the most of those services
Deep learning neural networks in the data center are previewing incidents and working on operations parameters to oversee the process of pinpointing and addressing root causes. As data center events grow in scale to millions of events per day, neural networks can be leveraged to respond to events in real time much faster than humans via traditional alerts.
The advent of intuitive machines that can make decisions about how events should be addressed in the data center can predict data center infrastructure problems as well as maximize performance while reducing cooling and energy costs. The use of neural networks can derive accurate predictions of future issues to enable proactive circumvention that reduces downtime risks and operational costs while increasing service levels and computational performance.
Researchers are deep into experimentation with using neural networks to enable faster and more accurate network load balancing for the data center in real time as speed and density of traffic across the data center continues to increase exponentially. A recent study featured in the peer reviewed PLoS One Journal details how convolutional neural networks can be used to forecast short-term data center traffic loads.
Los Angeles Data center providers like Telehouse are working with businesses of all type and sizes to plan for public and private cloud strategies that will take them well into the future. AI, ML, deep learning, and neural networks are still in the beginning stages of evolution. Still, their impact on the data center is already being felt and is shaping the networks and workloads that will drive businesses forward in the digital age.