看了RISC之父David Patterson文章后的一些摘要:The Trouble With Multicore
http://spectrum.ieee.org/computing/software/the-trouble-with-multicore/0
- 背景:
2003年左右,chipmakers开始意识到 处理器的性能无法继续提升了;无法提升的原因:虽然可以继续make transistors smaller and faster,但是散热问题无法解决;于是产生了multicore microprocessors;
- multicore上software application的执行速度,取决于“程序的并行性”;
- 提高程序并行性的挑战?--- load balancing, sequential dependencies, and synchronization
- 提高多核利用率,已经try的解决方法:
- Languages for parallel programming,Since 1960;--- failure
- New computer architecture,such as EI Dorado; --- failure
- Software that will automatically parallelized existing programs; --- maybe success
- 该问题今后10年可能的解决方法:
- GPU成为主流,替代现有的multicore;
- 提高程序的并行性;
Research Accelerator for Multiple Processors (RAMP) project ---http://ramp.eecs.berkeley.edu/