Slashdot Mirror
Understanding Moore's Law
S. Blocher writes "Ars Technica has a great article up, 'Understanding Moore's Law', that I think most geeks should read. The misrepresentation of Moore's Law in the media has always been a real pet peeve of mine, and this article does a great job of looking at the flipside of the 'bigger and faster' thesis to show how the Law isn't really just about doubling computer power."
It seems like a good start would be to stop calling it a "law," a term that has some kind of meaning, in a scientific sense. Exactly where that line gets drawn may be a little fuzzy, but I think it's fairly obvious that Moore's observations don't make the grade.
It Is the Nature of Information to Transgress Artificial Boundaries
I'll comment w/o reading the article.
Roger Moore's Law is: you get more chicks when you are James Bond than when you're Roger Moore. That's it right?
Microsoft's law: what you get when you put Moore's law and Murphy's law together.
...just my 2 gil.
For every news artical about Moore's Law, there will be a two fold reduction in the amount of Moore's Law that is explained. Which at some point it becomes impossible to comprehend the difference between Moore's Law and Artical Fodder about why you need a new computer.
Neck_of_the_Woods
#/usr/local/surf/glassy/overhead
it is a theory, or a hypothesis, or an observation. A law in the scientific jargon must be something which is:
6 a : a statement of an order or relation of phenomena that so far as is known is invariable under the given conditions b : a general relation proved or assumed to hold between mathematical or logical expressions
Moore's Law only holds true to a point. There comes a time when only so much can be fit on that piece of silicon. The term conjecture might also be applicable.
In SOVIET RUSSIA... erm...NSA AMERICA, the Internet logs onto YOU!
Moore's Law is so perennially protean because its putative formulator never quite gave it a precise formulation.
I tried to read further, but started twitching uncontrollably. How about Mace's Law: "The skill of tech writers halves, and their pride doubles, every 12 months."
I always gave about the same credence to both Moore's Law and Murphy's Law.
Ok, I finished the article. I learned some history, saw some graphs, and care not one bit more about Moore and his infernal Law.
...
Is to give you an excuse to avoid work. See http://arxiv.org/abs/astro-ph/9912202 for a paper (in PDF) describing this
If you're referring to that recent Red Herring article, my article was indeed "inspired" by it in the sense that I thought it was sensationlistic crap and I just couldn't take it anymore. For more info, see the news blurb that announces the article:
h tm l
http://arstechnica.com/archive/news/1045747027.
Senior CPU Editor | Ars Technica | http://arstechnica.com/
Basically, Amdahl's law says that you can only optimize to a point before it becomes pointless. Also, it is often used to demonstrate that it is useless to optimize a single aspect of a system when the other aspects are still unoptimized.
For example, let's say you have a computer that executes something within 10 seconds. 5 of those seconds are spent reading from disk, 5 are spent doing calculations on the CPU.
Now if you upgrade the disk to a disk twice as good, you're gonna get an execution time of 7.5 seconds(5 for CPU, 2.5 for disk). So you gained 2.5 seconds.
Let's say you still think it's too slow, and upgrade the disk again to a disk twice as good. You're now getting an execution time of 6.25 seconds(5 for CPU, 1.25 for disk). You thus gained 1.25 seconds.
You should get the trend here, if you continue upgrading only the disk, it will come to a point where, even by increasing the speed of the disk tenfold, you will only gain small fractions of second on your execution time, and small fractions of second over more than 5 seconds is definitely not a good improvement.
The same reasonning can be applied(this is the original intent of Amdahl's law) to multi-processor machines. Assuming only 50% of a program is runnable in parrallel processing, it comes to a point where adding processors brings very little improvement, even were you to increase the number of processors tenfold. (The explanation for this is left as an exercise to the reader... Hint: it's the same explanation as the disk/cpu above.)
My new law/observation/theory/detriment/prophesy is:
Every 18 [eighteen] months a new faster and more powerful processor is released to the masses that makes my 1Ghz seem obsolete.
Not a sermon, just a thought.
[n8.r0n] http://petesweb.spymac.net/
Quite simply, companies expect Moore's law to remain true. Software companies plan product lines in anticipation of processor capability doubling every n years. Processors are going to get improve at this rate, therefore we know how quickly bus performance and peripheral performance need to improve.
Semiconductor manufacturers know this. They plan product lines in a Moore's-law-consistent manner (not necessarily explicity, but surely as a matter of economics). If they're a little behind the curve, more money gets put into keeping up with it... or somebody else steps up and keeps it true. If they're at or ahead, they hold the course.
If someone were to introduce a processor that was 10x the density/speed of current processors, don't you think more resources would then go into peripheral design/heat management/software development to utilize the improvments, rather than continuing to focus on improving the processor?
The reason the law has held so long and seems to be so consistent is that it sets everyone's expectations, and people plan towards those expectations. Not less, not more.
You've all been Jedi mind tricked.
This has been achieved by getting a more and more detailed understanding of the processes and eliminating the fundamental sources of the problems. The costs of doing this are immense, but it works. It's striking to look at micrographs of chips today - everything looks so good. No ragged edges anywhere. Think for a moment about what that means. In some of those pictures, you can see atoms, and they're in the right places. Atoms.
It's not like the bad old days of the "purple plague", ceramics with traces of radioactive minerals, or the HP fab with the 4% yield.