Performance is marred by memory
Brian Bailey, EDN
August 30, 2012
For many years the industry created faster and faster processors. This was possible because more transistors were available in each technology node that could be used to produce even more complex and optimized pipelines. At the same time, voltages dropped enabling the processors to keep a lid on power consumption, a problem that if left unattended would have resulted in thermal breakdowns in the devices. But this came to a screeching halt a few years back. Physics started to work against us and while additional transistors were available, that were having diminishing impact on performance. At the same time, leakage power started to rise, meaning that power consumption was increasing, and not easy to turn unused parts off when deeply embedded in a pipeline. Everyone quickly accepted that the answer to both of these problems was to stop producing faster processors and to start producing more processors. Dual cores, quad cores and many core devices quickly appeared and since that time the software industry has been trying to work out how to deal with the sudden introduction of concurrency.
Click here to read more ...