| 1:40 - 1:45 | Opening remarks | Brandon Reagen Harvard |
| 1:45 - 2:15 | Processor thermal management: from energy harvesting to advanced cooling [Abstract] |
Carole Wu Arizona State University |
|
High temperature in computing devices is becoming the show stopper for high-performance computing platforms from smartphones that serve our personalized computing needs to high-capacity data center servers. High temperature stems from the high heat flux produced as a result of high computation demands. Instead of dissipating the processor waste heat, one potentially plausible approach is to harvest the heat that would otherwise be wasted and use the recovered energy to improve the energy efficiency of the processor. This talk will present the processor energy harvesting opportunity and the approaches undertaken. Then it will transition into designs with advanced active cooling technologies to improve the overall system energy efficiency.
| ||
| 2:15 - 2:45 | Pre-silicon power-performance modeling and management: challenges and pitfalls [Abstract] |
Pradip Bose IBM |
|
Cycle-accurate performance modeling and associated efforts to project power consumption constitute key aspects of pre-RTL concept definition of microprocessors. Defining on-chip power management architectures is an added task in the modern power-constrained design environment. In this talk, the speaker will present an experience report, based on his involvement with a range of POWER and mainframe processor development projects at IBM over the last several decades. In particular, the pitfalls of pre-silicon modeling (in terms of potential sources of errors) that can lead to major post-silicon surprises or mid-stage project cancellations will be summarized. Real-life examples, taken from major processor development project experiences will be cited.
| ||
| 3:00 - 3:30 | Coffee Break | |
| 3:30 - 4:00 | The unrealized potential of data value speculation [Abstract] |
Antonio González Universitat Politècnica de Catalunya |
|
The concept of data value speculation started to be researched in the mid 90s. The initial idea was focused on boosting performance in single threaded microprocessors by breaking data dependences. Later, the scope was expanded to multithreaded/multicore processors, giving rise to a plethora of works on what was called speculative multithreading. These topics attracted the interest of many researchers and were the focus of many published papers for more than a decade. However, in spite of many interesting, innovative and promising ideas, these works have had a limited impact in industry. In this talk, I will review these technologies and discuss its strengths and weaknesses, and the main reasons for their limited impact in commercial microprocessors.
| ||
| 4:00 - 4:30 | Through fire and flood – what it took to get Catapult into production [Abstract] |
Andrew Putnam Microsoft Research |
|
Microsoft Catapult has brought the power and promise of reconfigurable computing to hyperscale datacenters, accelerating major production cloud applications such as Bing web search and Microsoft Azure. But the getting Catapult to production required overcoming innumerable roadblocks, uncertainties, and challenges. In this talk, we will reveal some of the challenges, how they were overcome, and the stories behind what it took to get Catapult from a caffeine-fueled PowerPoint presentation into the largest deployment of reconfigurable computing in the world.
| ||
| 4:30 - 5:30 | Panel followed by Q&A | |
Not all research projects end up with positive results. Sometimes ideas that sound enticing at first run into unexpected complexity, high overheads, or turn out simply infeasible. Such projects often end up in a proverbial researcher's drawer, and the community as a whole is not aware of dead-end or hard-to-advance research directions. NOPE is a venue that encourages publishing such results in all their "badness".
The best negative results help us learn from our mistakes. They can illuminate hidden obstacles or demonstrate why we need a change of course. An ideal submission to NOPE has a novel idea which sounds plausible from first principles or design intuition, but yields little to no improvement (in performance, power, area, …) in practice. The paper drills down into the reasons for the lack of improvement and proposes a plausible explanation – different technology trends, unexpected implementation complexity.
Our goal is to find papers which the community can learn from and might otherwise have trouble finding a suitable venue, so we take a broad view of what constitutes a "negative" result. A good NOPE submission might entail:
Questions? Send us an email.
We believe in substance over style, and we encourage authors to prioritize delivering their message over conforming to a particular template. That being said, we anticipate papers will probably end up in the 4–6 page range, and we encourage authors to use a two-column format. Papers need not be anonymized.
Additionally, we ask that you also include a short, 1-paragraph abstract in your submission email. This should be suitable for inclusion on the NOPE website and program handouts.
Finally, while one of the goals of NOPE is to find a home for papers that can sometimes be difficult to publish elsewhere, we do not wish to preclude publication elsewhere. NOPE 2016 will not be indexed with IEEE or ACM, so authors should feel free to expand and submit their work to larger venues. We discourage resubmission of previously published papers, though "second-look" papers or retrospectives fall squarely within the scope of the workshop and are welcomed.
Please submit your papers via email to: submissions@nope.pub by 11:59pm (anywhere on Earth) on the deadline.