Pages: pp. 2-3
Abstract—This column discusses hot interconnects and the three keynote speeches from the Micro 2011 conference. It also introduces a new Editorial Board member.
Keywords—Hot interconnects, Micro, CPU, GPU, accelerators, Kinnect, dark silicon, high-performance computing, IEEE Micro editorial board members
Happy 2012, and surprise! This issue is focused on Hot Interconnects, not Top Picks. For the first time since 2005, Top Picks is not the theme of the January/February issue of IEEE Micro. But no worries—Top Picks is not going away. We thought that, to allow time for adequate consideration of papers published and presented at the Micro conference, it would be better to delay Top Picks to the May/June issue of 2012. So, stay tuned.
Once again, we welcome a new member to the IEEE Micro Editorial Board. I am pleased that Babak Falsafi accepted my invitation to join the Board. Please see the sidebar for his biography.
For this issue, we have three interesting articles on the theme of Hot Interconnects. Such interconnects are an increasing part of obtaining good performance, especially in high-end machines, such as those vying for top spots in the Top 500 list ( http://top500.org). The computing deals with increasing data volumes in everything from meteorology and climate to earthquake prediction to brain modeling to business analytics and web traffic, so having appropriate interconnects becomes more important. Guest Editors Torsten Hoefler, Patrick Geoffray, Fabrizio Petrini, and Jesper Larsson Träff have done an excellent job of recruiting and selecting articles on this topic. I hope you find them as interesting and informative as I did.
As I write this introduction, I have just returned from the Micro conference, which had many interesting presentations and sessions. Not surprisingly, virtually all were relevant for IEEE Micro readers. However, I will limit myself to the three excellent keynotes. Pairing the second and third keynotes yields the theme of our recent September/October 2011 issue on GPUs versus CPUs. Clearly this is a hot topic in the IEEE Micro community. Steve Keckler of the University of Texas at Austin and Nvidia gave a thoughtful assessment of the path ahead for GPU chips, particularly as they relate to high-performance computing. In particular, Steve observed that a primary distinction between GPUs and CPUs is that GPUs can be viewed as having a primary focus on high throughput, while CPUs have primary focus on low latency. Because GPUs are designed for high throughput as needed for HPC, Steve argued that they will have a natural advantage. On the other side, Avinash Sodani of Intel argued that conventional wisdom is faulty in viewing CPUs as impractical going forward due to their supposed wastage of area and energy on things like caches, coherence, and speculation. Contrary to conventional wisdom, Sodani showed much evidence that these things either offer considerable help to HPC or are not terribly costly. Whomever of Keckler and Sodani turns out to be more prescient, they will almost certainly need the high-performance interconnects described in this issue.
Doug Burger of Microsoft gave the third keynote at Micro, and challenged the audience to take a more applications-driven approach to microarchitecture. Building on his coauthored ISCA 2011 paper and others, Burger observed that a coming era of "dark silicon" is likely. (The July/August 2011 issue of IEEE Micro also featured an interesting article on this topic, "Toward Dark Silicon in Servers," by Nikos Hardavellas et al.) Burger argued that, because of dark silicon's power limitations, efficient accelerators and microarchitectures are needed. In turn, to make such accelerators successful, Burger argued that we must make microarchitecture more application-oriented, and furthermore, that we enjoy a rapidly increasing set of interesting applications from games and their sensors—especially Microsoft's Kinect—to tracking collars to language recognition to computational holography, and many more applications. Taking up this challenge, IEEE Micro would be happy to receive articles addressing this view of application-driven microarchitecture. Send us your thoughts!
And, in the meantime, good reading.
Erik R. Altman
Editor in Chief
IEEE Micro welcomes a new member to its editorial board.
Babak Falsafi is a professor in the School of Computer and Communication Sciences at École Polytechnique Fédérale de Lausanne and an adjunct professor of electrical and computer engineering and computer science at Carnegie Mellon University. He's the founding director of the EcoCloud research center at EPFL, innovating future energy-efficient and environmentally friendly cloud technologies. He also directs the Parallel Systems Architecture Lab (PARSA) at EPFL, where he conducts research on technology-scalable data center design, design for dark silicon, data-centric memory systems, and techniques for computer system performance evaluation. He received an NSF CAREER award in 2000, three IBM Faculty Partnership Awards, and an Alfred P. Sloan Research Fellowship in 2004. He's a fellow of IEEE.