Stephen S. Pawlowski is Intel’s Senior Fellow and Chief Technology Officer for the Datacenter & Connected Systems Group (DCSG) and General Manager for the Architecture Group & DCSG Pathfinding. On June 18, Steve will be keynoting at the International Supercomputing Conference (ISC’13) and the topic of his talk is Moore’ Law. Here’s a Q&A session between Steve and ISC.
1.In his book, Physics of the Future, Michio Kaku says that Moore’s Law is nearing its end. Many in the high-performance computing community also share this belief; they are saying that after 2013 transistor counts and densities will double only every three years. Will Intel prove the skeptics wrong?
I expect that we will. The pace of innovation in the electronics industry is accelerating actually. The demand for smaller, lower power devices with new capabilities is not just being driven by the PC industry anymore, but is strongly being driven by the consumer electronics industry as well. When we look at the exascale space and the big data analytics space, the demand continues to be very strong. Intel continues to drive process innovation to meet those demands. It’s a law of economics really… that supply and demand are closely related… were motivated to serve the exploding global marketplace…and we will.
2. Chip design in the past was mostly focused on the aspect of performance increase. This appears to be changing. Do you agree that energy efficiency, that is performance per watt, is now the predominant design criteria?
I disagree that design for performance per watt is new. Intel has been focusing on energy efficiency in semiconductors for 20 years. What has changed is the explosion in the ways computing reach the consumer. We have been squeezing more performance per watt into our processors not only because of data center power limits, but also to increase performance and battery life of our notebook and mobile products.
3. For HPC systems, it’s essential to have more and more cores on a chip, but memory access isn’t keeping pace with the additional cores. How do you view this memory bandwidth bottleneck?
It’s a very important area of innovation. Workloads require balanced system performance, performance density, memory bandwidth, IO bandwidth, storage systems. It’s important to realize though that effective memory bandwidth is heavily impacted by processor architecture. How a processor accesses memory and the use of its cache impact the memory bandwidth an application truly requires. This is where some of the solutions to the problem will lie. The other area to look into is the speed and density of memory interconnects. There are some amazing technologies coming that address this area. It’s not a simple spec that drives real workload performance.
4. Many core systems with millions (or more) of cores pose a significant challenge to programmers. What is Intel doing to make life easier for the programmer who need to develop applications for these extreme-scale systems?
Intel is constantly working on improving programmer productivity. Bringing this complexity down to manageable levels is not only a key area of value for us…but also a key area of research.
5. Will the old and “successful” vector unit of former supercomputer systems experience a renaissance?
I really doubt it. Most HPC data centers have a wide range of workloads with a wide range of requirements. Few people run “just one thing” any more… Intel is proving that a standard architecture with long term continuity can efficiently serve a wide range of workloads… That’s the key learning from Intel Many Core work.
6. On a more personal note, you’ve worked for Intel most of your professional career. What has encouraged you to remain with the company for 31 years?
I don’t know if my wife could tolerate me being around the house… Seriously, this is a company of innovation and impact and there are very FEW with Intel’s scale. I’m fortunate that I get to work here, and I come to work and I learn something new every day. A new idea here can change the world relatively quickly. In addition, I also get to interact with some of the best and brightest that our industry and universities have to offer. There is a culture of growth and vision, and it deeply respects the human factor as well. I come to work each day because I do love it.
7. What are the biggest changes you have experienced at Intel since you joined in 1982?
The biggest change? How Moore’s Law and Intel’s focus on driving down the cost and increasing the density and performance of our process technology has profoundly changed all of our lives. For Example:
1. The bill of materials, not the price, for just the electronics components of my first PC design, a 25 MHz 80386 based machine was almost $2000, and its performance pales in significance to a cell phone today.
2. The Human Genome has been mapped and now we can use that information to find cures for diseases that in 1982 were considered unimaginable.
3. Social networking has become pervasive. One day my daughter ‘texted’ my wife and I not more than five minutes after we left the restaurant we were at to tell us that she had read, on Facebook, what we had just eaten. This would not be unusual, but my wife and I actually don’t do Facebook J.
4. Tremendous amounts of information are at our fingertips. Encyclopedias Book sets, which were essential when I was a kid, are a thing of the past.
5. Cameras are everywhere, Pictures are ready to share instantaneously and film is almost a thing of the past!
6. In 1982, CDs came out in August…I still bought my music on vinyl records and cassette tapes. High definition televisions are now ubiquitous and cheap.
7. Young people rarely wear wrist watches any more.
8. Cell phones in 1982? What cell phones?
So, at some point we all should look at how, over the last 30 years, Moore’s Law has profoundly changed our lives! And here’s the best part…at least for me…I have gotten to watch, participated, and STILL participate in many of those changes!