The end of scaling looms large, and computer architects need to find innovative ways to deliver increased functionality and integration in the face of the end of Moore’s Law. What other important trends will guide the development of computer architecture in the coming years? There are a great number of challenges for architects, and it’s an exciting time for research in this area! The community is also considering accelerators in a range of platforms from ultra-low-power IoT devices up through server/datacenter class devices. Much of the early work in this domain was largely focused on inference for convolutional neural networks already we are seeing the architecture community pivot toward other algorithms and to also consider training as an important aspect for acceleration. Accelerators for AI need to be general enough to not become obsolete before they are even manufactured but not so general that they give up their energy and performance advantages. One challenge facing architects is the continued evolution of AI algorithms. AI applications are showing tremendous promise across a range of applications yet require greater compute and energy efficiency to fully deliver on that promise. The computer architecture community has seen a huge explosion in terms of research volume in hardware acceleration for AI applications. How will the growing use of deep neural networks for AI applications impact computer architecture? We should not be oblivious to the carbon footprint of our computers and the emerging killer applications that are running on them, such as machine learning. Finally, there is the role of computing in our overall climate. At the high end (datacenters), energy translates to cost-cost to operate the computers and cost of extracting the heat generated. Energy efficiency is a concern across devices, from the smallest IoT devices that harvest ambient energy from their environment and must operate under the strictest of budgets, to battery-operated devices such as cell phones up to datacenters and warehouse-scale computers.Īt the low end (IoT), maximizing energy efficiency will allow these devices to operate for longer periods of time and do more meaningful computations locally. Since then, there have been remarkable advances in improving energy efficiency in modern computer architectures. The end of Dennard scaling in the mid-2000s catapulted energy efficiency to the front lines of computer architecture. Given the current state of computer architecture and technology trends, why does energy efficiency continue to be a priority?
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |