[c627627]

Ed Stroligo had a great one three or four years ago but I can't remember what it was.




EDIT: The thread is in trouble even though I know you all don't answer
"...but this one has more m'gahurtz" with talk about 'work per clock cycle'.

How do you illustrate it to them?


EDIT EDIT: Simplest analogy requiring the least amount of thought through the use of something that a minimum wage person can recognize instantly. Preferably a two sentence hit:

It's like A vs B.
B xxx faster but you get more xxx with A.


* * *

If you wish to skip the first page or so, here's a quick summary:


Graphic67: 8-cylinder engines vs. 4-cylinder engines story.

Hoot: Two school backpacks with different number of books story.

Gnufsh: Why do you need an analogy?

borg: posted to say he has no analogy.

mustrum: posted to say Graphic67's analogy would be correct if US car producers wouldn't suck so bad in engine technology.

deeman: posted to say we need to put an automotive thread in the forums.

JigPu: a 283 word post about floating point operations.

Finally deeman: posted a link to a story about a child and an adult walking together, and the child needing to take more steps to keep up with the adult.

Cyrix_2k: posted a story about 40mph 4-lane road vs. a 65mph 2-lane road.

barton2500: story about cycles per second.

c627627: took barton2500's story to Joe at the gas station to which Joe replied "Bicycles, what?"



Gnufsh: posted full blown Einstein's Theory of Relativity as an example. (OMG)

[Graphic67]

Using megaherts to compare different processors is like using RPM's to compare different engine technologies when you are really interested in horsepower/torque.

small 4-cylinder engine ==> high rpm, low horsepower/torque

8-cylinder engine ==> moderate rpm, good horsepower/torque

diesel 8-cylinder engine ==> low rpm, excellent horsepower/torque

[Hoot]

Two students are in the same homeroom and have the same four classes before lunch. Student A has a backpack that he takes to every class. In that backpack, he puts the four books he will need for the four classes. Student B does not have a backpack and only takes the book he will need for the first class. As the two head for their first class, student B makes better time through the hallways and gets to class first. After the class lets out, student B makes better time through the hallway back to his locker to drop off his book and get the book he needs for the second class, while student A heads directly to his second class, albeit at a slightly slower pace. They both arrive at the same time. After the second class lets out, the process repeats, but their class is further from their lockers. Student A arrives first. Student B arrives a little later and a little out of breath. After the fourth class, well, you can see where this is heading...

Hoot

[Gnufsh]

Why do you need an analogy? Why not just say that, on average, a P4 completes the execution of fewer instructions than does an AthlonXP or a G4e. If someone actually wants to know why this his, talking about RPMs and school lockers won't actually give them an understanding of what is going on. Goingin-depth into an architecture and talking about what each pipeline stage does, the penalties for a branch mispredict, how often the branch predictor is wrong on average, caches, memory latency, and that sort of thing might eventually lead to actual understanding. That's just the way I like to learn. An analogy makes a good firs sep to understanding, but we should be able to go beyond that.

[borg]

have no anology, but can u (gnufsh) post some links to tutorials/articles explaining how the pipeline/branch predicition/cpu generally in-depth works? would be *very* grateful...

to stay a bit ontopic... isn't MIPS or FLOPS better cpu power indicator than Mhz? why isn't it displayed?

[mustrum]

Originally posted by Graphic67
Using megaherts to compare different processors is like using RPM's to compare different engine technologies when you are really interested in horsepower/torque.

small 4-cylinder engine ==> high rpm, low horsepower/torque

8-cylinder engine ==> moderate rpm, good horsepower/torque

diesel 8-cylinder engine ==> low rpm, excellent horsepower/torque

Then check out REAL cars.
Us cars: 8 cylinders low RPM low horsepower high torque
European/Japanese: 4 cylinders high RPM high horsepower low torque
Diesel: as stated above

Sure your explanation WOULD be correct if US car producers wouldn't suck so bad in engine technology.
Well most likely the could do better but what for if you can burn gallons of fuel that cost close to nothing.

I used to drive one of those US cars with a 4 litre V8 and 170 horesepower. Hard to imaging how they could produce this engine THAT WEAK. The torque couldn't save it... just plain sucked.
A 2 litre 4 cylinder owns that engine while using half the fuel.
Only advantage of the US crap engine. SOUND. Just great sound that makes you happy everytime you start it up.

Yeah i AM digressing.

[deeman]

Ok just out of curiosity this thread started talking about computer analogy and now we are talking about 4cyl to 8 cyl cars do the mods need to put an automotive thread in the forums so you can place all this dribble in its own section and maybe these threads will stay computer related.

[ChillPhatCat]

It degenerates because someone is misinformed and posts something either completely wrong or opinion based... I'm gonna delete my post even though it was so masterfully written. I suggest someone else (not gonna name names) deletes their reply too...

[Quailane]

Ok, ok, I can take a hint.

This is my analogy- it's like comparing runners by how much they weight.

[mustrum]

Oh comeon. Don't be so stiff.
It'S not like this thread is about anything serious and he (thread starter) started the analogy himself.

[JigPu]

Originally posted by borg
to stay a bit ontopic... isn't MIPS or FLOPS better cpu power indicator than Mhz? why isn't it displayed?

Somewhat, but even they're aren't free from bias.

Using MIPS requires you to have a standard benchmark, which paces through a piece of 'average' code. The program knows how many lines of assembly are there, and can determine how many million lines are executed each second.

However, that's exactly like using a single benchmark to determine the real-world performance of any chip. Seeing how fast it is with average stuff tells us NOTHING about how fast (or slow) it is with non average stuff. Worse, the defenition of average can (and is) set by the creator of the benchmark, unintentially or not biasing it to one design and "average" use.


FLOPS is in the same boat. FLOPS measures the number of floating point operations per second. This is only measuring one small aspect of the chip. Again, this shows a biased view of who's on top in the world.


Even listing out the execution times of each command (while better) does not totaly eliminate problems. It would show you what chip executes classes of commands the fastest, and you could choose the best chip from those (for example, finding a chip really fast at trig if your program is trig intensive). However, the commands themselves are only part of the equation. Because everything is pipelined, execution order, branch prediction, pipeline stalls, etc all contribute to the overall performance. If a chip is really good at trig, but chokes on your program because of constant pipeline stalls, it dosen't matter how efficient it is


Real world performance can't be summed up in one number. Nor can it be summed up in several numbers. There's just way too many variables for stuff that simple
JigPu

[c627627]

I know you all don't answer
"...but this one has more m'gahurtz"
with talk about 'work per clock cycle'.

How do you illustrate it to them?

Thank you Graphic67 and Hoot very much for trying.
If anyone posts another good example that can be used on the less litterate, I'd be grateful.

[deeman]

heres a AMD link on analogies
http://www.amd.com/us-en/Processors/...5E4593,00.html

[c627627]

Thanks deeman, the first two AMD link analogies are so-so but the third one is kind of OK:
"If a child and an adult are walking together, the child needs to take more steps to keep up with the adult. Since the adult has a longer stride than the child and travels further with each step, the child has to work harder – by moving faster – to try and keep up."

[c627627]

Originally posted by Gnufsh
Why do you need an analogy? Why not just say that, on average, a P4 completes the execution of fewer instructions than does an AthlonXP or a G4e.

Ah yes .
Anyway, it's because you're going to be met with blank stares by the people who say "...but this one has more m'gahurtz"

Would you seriously respond to a question posed to you in that way with the kind of talk we use at the forums? Have you met any of the majority of the ig'nant and illiterate population of any country in the world?

[Cyrix_2k]

It's like a 4-lane road with speed limit of 40mph vs a 2-lane road with a speed limit of 65mph. The 4-lane road moves more traffic(work) at lower speed(mhz). The 2-lane road moves less to the same traffic(work) at a higher speed(mhz).

Also, some #'s to through in:
Work able to be done by the 4-lane road:160 units per hour
Work able to be done by the two-lane road:130 units per hour

Does that work ok?

[c627627]

They all work and thanks.

But the goal is to find the simplest analogy possible requiring the least amount of thought through the use of something that a minimum wage person can recognize instantly. Preferably a two sentence hit:

It's like A vs B.
B xxx faster but you get more xxx with A.

[Cyrix_2k]

It's like volume vs speed.
You get more speed with a P4, but you get a larger volume of work with AMD (when it's run at a comparable speed).

[barton2500]

Here is an interesting example: Pentium 4 1.8A beaten by a Duron 1.3
http://www6.tomshardware.com/cpu/200...charts-28.html

Now that is impressive!

Here's my analogy:

CPU A can do 4 calcuations / cycle at 100 cycles per second = 400 calculations / second

CPU B can do 3 calcuations / cycle at 133.3 cycles per second = 400 calcuations / second

MHz is just how many cycles per second, and does not indicate the absolute performance of the CPU. It is only one of several factors that affect performance.

[c627627]

Thanks barton2500, I'll try that with Joe down at the gas station and see what he thinks.