With carbon trading very much in the spotlight
Posted by Laurence Marchini.
With carbon trading very much in the spotlight, it’s all too easy to fall into the “lies, damn lies and statistics” trap - particularly if you forget to look at the bigger picture.
For example, this week while reading a story on public perceptions of which industries are the worst CO2 offenders, I came across the statistic that although the IT industry has a “clean” public perception, computers are responsible for more than 0.75% of total global greenhouse gases.
Now, I’m not suggesting that this is a lie (or even a damn lie), but like any other statistic, its sense relies on knowing the fuller picture, and in this case the meaning of the word “responsible”. And, on closer examination, the figure does appear reasonable, as the survey concerned covered both the manufacture and use of computers in arriving at its conclusion.
The need to consider the total carbon picture (covering both manufacture and operation) is becoming a hot topic. But again it is producing a plethora of statistics, any of which can be hijacked and used with dubious context to “prove” any number of spurious points.
For example, it has even been suggested recently that the carbon emissions associated with manufacturing green power generation products such as wind turbines might far outweigh the savings that might be accrued during their lifetimes. The flaw in that argument, of course, is that it relies on a depressingly short lifetime for the turbine. But it does illustrate the dangers of focusing too hard on one end of the equation.
All this has been brought into focus this week by configurable processor specialist Tensilica, which this week has taken a novel approach to power-optimised IC design with the release of its new power estimation software, dubbed Xenergy.
All too many IC designers, it seems, fall into the trap of slavishly following the milliwatts-per-megahertz metric to the exclusion of all other power consumption data - ignoring the simple fact that the ability to process suitably crafted instructions in fewer clock cycles, for example, can have as much (if not more) influence on total power consumption. And when memory accesses are factored in to the equation, that raw processing power measure becomes but a small part of the overall picture.
This does rather beg the question whether we might all be guilty of focusing on the “wrong numbers”. Are there other areas where components are specified based on one (supposed) key feature, at the expense of the bigger picture? Any suggestions?
This comment was originally published in the Electronicstalk Newsletter
Make the first Comment on this •
Trackback • Permalink •