Accelerating Data Insights with Intel’s® Deep Learning Boost
These days, keeping up with the times as an organization means leveraging massive amounts of data that’s constantly being collected on a myriad of devices. Broadly speaking, the more data-driven you are, the more opportunities you can create for your business. If you’re familiar with data analytics, you’ve likely encountered terms like “unstructured data” or “data bloat,” and have experienced the pitfalls of the process firsthand. In short, creating value from all that data can be a cumbersome process. Entire companies exist now whose sole purpose is to crunch the numbers for you and deliver actionable insights.
Regardless of who’s at the helm, the fastest way to gain those sorts of insights is through using deep learning models. The fastest way to do that is to accelerate the speed at which a trained model can make itself useful and provide accurate predictions through a process known as inferencing. And the best way to speed up inferencing is to utilize Intel’s® Deep Learning Boost.
Briefly, Deep Learning Boost extends the AVX-512 instruction set to include VNNI, or Vector Neural Network Instruction, which delivers significant inference acceleration. In layman’s terms, it basically means you get the information you need faster through a more efficient method of calculation for complex deep learning workloads. It does this by decreasing compute and memory bandwidth requirements, as well as switching to using a single instruction that used to require two or three.1
Okay, that sounds great, but what’s it for? Image classification, speech recognition, language translation, or object detection are all notable examples. Every industry you can think of could potentially benefit from this newly optimized workload. Take the medical field, for instance. We will never replace the need for doctors, but what we can do is provide them with data-driven insights in real-time that they can use to treat patients more effectively. A single doctor cannot analyze millions of x-rays to determine if someone is in the early stages of a hard-to-detect disease—but a well-trained deep learning model can. The best treatment is preventative treatment, and the better we can get at detecting the signs, the better we will get at eradicating even the most virulent diseases altogether.
AI-powered insights like this also means we can provide better care, faster, to more people. In the healthcare field, it isn’t about replacing the human element, but rather leveraging the technology to make healthcare professionals more effective. Thankfully, this technology is embedded in all the latest high-end Intel CPUs.2 As long as you’re running an optimized framework (e.g., PyTorch, Caffe, MXNet, PaddlePaddle, TensorFlow) and are using a Cascade Lake processor, you’ll be able to get the crucial performance gains of Intel’s Deep Learning Boost.
Even if you work outside of the healthcare field, if you have lots of data that needs crunching, it’s likely you can reap the benefits of this amazing new technology. If you’re looking to get serious about leveraging your company’s data to provide better insights, or are simply curious about what’s possible, consult with a BOXX performance specialist. BOXX provides a wide range of workstations, from desk side data science workstations all the way up to high-density servers designed to massively accelerate the most demanding deep learning workloads.
1 For 16-bit operations, two common operations were combined into a single instruction. For 8-bit operations, three common operations were combined into one.
2 2nd gen Xeon® SP, Xeon® W, and 10th gen Core™ X-Series.