It's doubtful. A typical soldering iron is in the 20-40W range. This one is only 100mW. However if you had 200 of them all aimed at the same spot...:-)
You've just re-stated Wirth's Law:
Software gets slower faster than hardware gets faster.
I suppose it is no surprise that linux distributions get slower at almost the same rate as other OSes. If you'll recall, Win3.11 was faster than Win9x, was faster than Win2k.
At least with the linux distributions, it is because they're bolting on extra software that I can remove. Under that Other OS, I'm just screwed.
And no, there is no direct corelation between between "clock speed" and power consumption (if that were true, then why does the MC68000 at 16mHz use 28 watts and the Dragonball EZ (same architecture) use 950 mW?)
Every VLSI design textbook I've ever read lists power is being proportional to f^2. Now maybe I just need to read more textbooks, but I doubt they'd be teaching this relation if it weren't true.
Keep in mind that the same ISA does not imply the same architecture.
And let's not forget something as equally important as clock speed: data piplelines. The Pentium 4 has a 24 stage pipeline whereas the G4e has 7: in broad terms this means that while the p4 can work on 3 times the instructions concurrently, the G4 executes it's stack in a third of the time.
Those are very broad terms you use. Let's look at an example program of 1000 instructions, with no branches, running on processors with a single execution unit.
A pipelined design means that once you fill the pipe, you get one result per clock cycle. This is the definition of pipelining.
A machine with a 24-stage pipe would take 1000 cycles of execution, plus 24 cycles to fill the pipe, yielding 1024 cycles to complete the program.
A machine with a 7-stage pipe would take 1000 cycles fo execution, plus 7 cycles to fill the pipe, yielding 1007 cycles of execution.
Now if the 1 cycle in the deeper pipelined machine takes half as long as 1 cycle in the shallow pipelined machine, the deep pipe wins.
However: This whole comparison is bunk.
Modern processors like the P4 and the G4e are not simple pipelines, but very complex post-RISC OOO machines. In such machines, pipeline length is not a terribly important factor. It only really matters when your branch predictor guesses wrong or when you execute a context switch. If I keep typing, I'll ramble on forever.
I was trying to make two points.
Your comparison wasn't accurate
The comparison you wanted to make would be accurate only for very simple processors. Unfortunately, the P4 and G4e are not simple processors.
Oh yes, and we also all know that benchmarks are the be all and end all of performance evaluation. Especially those benchmarks that conform to no known standard, and aren't terribly well documented.
Please, spare me.
Benchmarks can give you a general idea of how performande might compare if you used exactly the same programs, input files, OS configuration, network load, other running processes, etc...
What are the chances that the military uses Photoshop for their image processing? I'd think not very high (unless there's a series of photoshop plugins I am unaware of that will process an image looking for convoys of trucks, bunkers, and other such things that the military cares about satellite images for). If he is not using photoshop, then the benchmarks you're getting so excited about are meaningless.
MacBibble has shown us that a Macintosh can perform quite well on image processing, if you run optimized code.
Benchmarks are just that, benchmarks. If this guy finds that for his application a Macintosh is faster, then let him use a Macintosh.
Slashdot Mirror
It's doubtful. :-)
A typical soldering iron is in the 20-40W range. This one is only 100mW. However if you had 200 of them all aimed at the same spot...
What are people who use Windows at work, Linux at home, and a Mac for school? Helecopters?
crh
You've just re-stated Wirth's Law:
Software gets slower faster than hardware gets faster.
I suppose it is no surprise that linux distributions get slower at almost the same rate as other OSes. If you'll recall, Win3.11 was faster than Win9x, was faster than Win2k.
At least with the linux distributions, it is because they're bolting on extra software that I can remove. Under that Other OS, I'm just screwed.
Every VLSI design textbook I've ever read lists power is being proportional to f^2. Now maybe I just need to read more textbooks, but I doubt they'd be teaching this relation if it weren't true.
Keep in mind that the same ISA does not imply the same architecture.
Those are very broad terms you use. Let's look at an example program of 1000 instructions, with no branches, running on processors with a single execution unit.
A pipelined design means that once you fill the pipe, you get one result per clock cycle. This is the definition of pipelining.
A machine with a 24-stage pipe would take 1000 cycles of execution, plus 24 cycles to fill the pipe, yielding 1024 cycles to complete the program.
A machine with a 7-stage pipe would take 1000 cycles fo execution, plus 7 cycles to fill the pipe, yielding 1007 cycles of execution.
Now if the 1 cycle in the deeper pipelined machine takes half as long as 1 cycle in the shallow pipelined machine, the deep pipe wins.
However: This whole comparison is bunk.
Modern processors like the P4 and the G4e are not simple pipelines, but very complex post-RISC OOO machines. In such machines, pipeline length is not a terribly important factor. It only really matters when your branch predictor guesses wrong or when you execute a context switch. If I keep typing, I'll ramble on forever.
I was trying to make two points.
Oh yes, and we also all know that benchmarks are the be all and end all of performance evaluation. Especially those benchmarks that conform to no known standard, and aren't terribly well documented.
Please, spare me.
Benchmarks can give you a general idea of how performande might compare if you used exactly the same programs, input files, OS configuration, network load, other running processes, etc...
What are the chances that the military uses Photoshop for their image processing? I'd think not very high (unless there's a series of photoshop plugins I am unaware of that will process an image looking for convoys of trucks, bunkers, and other such things that the military cares about satellite images for). If he is not using photoshop, then the benchmarks you're getting so excited about are meaningless.
MacBibble has shown us that a Macintosh can perform quite well on image processing, if you run optimized code.
Benchmarks are just that, benchmarks. If this guy finds that for his application a Macintosh is faster, then let him use a Macintosh.