Heterogeneous Computing and Solutions Provided by oneAPI
Heterogenous computer systems – systems that contain different kinds of computational units (CPUs, GPUs, etc.) controlled by a general-purpose processor (GPP) and augmented by accelerators (XPUs) – are here to stay. However, the computational complexity inherent in these systems creates some unique challenges in the healthcare industry. In today’s “Health and Life Sciences at the Edge” podcast, Intel’s Beenish Zia, Chief Architect for Medical Imaging in Intel’s Internet of Things Group, and Evgeny Drapkin, Chief Engineer for GE Healthcare Digital Platforms, talk with Tyler Kern about those challenges and solutions.
According to Drapkin, both the need for heterogeneous computing and the biggest challenges to its successful deployment can be illustrated by medical imaging. “In medical imaging, being able to deliver results from scans as quickly as possible is a necessity,” says Drapkin. “In many cases, like in stroke management, the speed of delivery can directly impact patient outcomes.” With the level of computational complexity growing every year, finding ways to increase image processing speeds is both imperative and challenging.
Enter heterogeneous computing using the Intel® oneAPI Toolkit. OneAPI is designed to meet the three biggest challenges developers and programmers face when creating heterogeneous systems:
- Determining which hardware architecture to use
- Selecting the proper software model
- Porting legacy software in ways that take advantage of modern technologies
“OneAPI stands for One Application Programming Interface,” says Zia. “It simplifies software development and programming by providing a unified programming model. This model gives programmers the freedom to select the best hardware for their workloads, optimizes hardware performance, and removes hardware vendor lock-in.” Best of all, the same learning model applies to all industries and helps build community and industry collaboration.
Both Zia and Drapkin agree that heterogeneous computing is needed to identify and map algorithms to accelerator devices, then program those devices to deliver faster results. “Industries like medical imaging are driving the need for a common programming language that can transform how software coding is approached,” say Zia. “OneAPI provides a unified programming model that can be applied to all industries. Heterogeneous computing is inevitable. It’s up to us to optimize it.”
Connect with Beenish Zia and Evgeny Drapkin on LinkedIn.
More about how oneAPI could assist in heterogenous computing and rapid software deployment can be found at oneapi.io.
Subscribe to this channel on Apple Podcasts, Spotify, and Google Podcasts to hear more from the Intel Internet of Things Group.
