In my previous blog entry, I spoke about a 15-year analysis showing how the performance improvements due to software were 43 times greater than the hardware improvements from Moore’s law. With hardware gains at 1,000X and software gains at 43,000X over 15 years, neither is too shabby. What was the key to the 43,000 times improvement in software? In short, a general purpose, computing environment that was flexibly programmed to whatever the style of computing required by the application and the algorithm.
Now let’s fast forward to 2011, where a study from the National Research Council, “The Future of Computing Performance: Game Over or Next Level?” concludes that, “Future growth in computing performance will have to come from parallelism”. The report further finds that, “There is no known alternative to parallel systems for sustaining growth in computing performance; however, no compelling programming paradigms for general parallel systems have yet emerged.”
Well, I am overjoyed that the world has woken up to the sequential single-processor performance and power wall (general purpose and hardwired) and realized the future of computing is parallel computing. I’ve been living, breathing and developing parallel computing systems for over 25 years. From this experience comes some insight, and we’d like to think, a meaningful head start.
Taking what we have learned from the last 50+ years of computing, a general purpose, flexible computing environment has proven to be the catalyst for both software productivity and home to Moore’s law performance bounty. So it would seem we have to create an analogous environment for the evolving parallel era; its all about a flexible software environment! With this in mind, the real gem in this report shows up on page 110, where the software implications of parallelism are discussed. The report outlines the 5 main challenges to increasing performance and energy efficiency through parallelism:
- Finding independent computational work
- Communication between units of computational work
- Preserving locality within and across units of computational work
- Synchronizing units of computational work based on overall computational dataflow
- Load balancing units of computational work across resources.
The good news- there are no surprises here! Insights into these 5 key challenges have led to Panève’s Rhino™ architecture. Rhino hardware architecture’s powerful primitives and abstractions enables a software environment where all of these challenges are addressed today and also adds a sixth key benefit- complete flexibility to enable applications and algorithms to continuously evolve and improve.
In my next blog entry, I’ll discuss these in more detail.