PDA

View Full Version : New formula for MHz myth


ryanweb
Aug 29, 2001, 01:53 PM
Does anybody know how to measure REAL speed and producivity of the computer?

Now people in general think, that MHZ=speed, but may be we can measure speed by instructions per clock cycle, and call it ICC or whatever, or find any other way to compare different computers?

If the name for this formula will be catchy, people will use it.

Nobody uses engine RPM to measure car speed, why we do use MHZ for computers?

mnkeybsness
Aug 29, 2001, 11:37 PM
there is no real formula, it's very technical to tell how fast a computer actually is. pipelines have big factors in it, g4s have short pipelins about 3 times shorter than p4s-which makes the mhz pretty much null. blah blah blah, tons of other stuff factor in too

spikey
Aug 30, 2001, 01:24 PM
I think alot of the confusion is that some processors are better at some operations than others. I think it has to do with vector processing units and floating point blahs.

The best thing to go by is benchmarks off non-zealot sites, or probably just experience.

iamnotmad
Aug 30, 2001, 01:38 PM
Isn't the only thing that really matters to us, time?

Time to do a task is the best measurement. Mhz OR # of pipeline stages does not matter as much as how long it takes to perform a given task.

How long does it take to compress my 3 minute video?

How long does it take to render my 900 frames of animation?

How long does it take to recalculate my spreadsheet?

How long does it take to scroll through this document?

How long does it take to query my database?

How long does it take to display my HTML page?

How long does it take to apply this filter to my image?

How long does it take to rotate this image?

I think that's the best way to measure speed. I am not belittling technical detail of processors, I find that very interesting, but what end users generally should want to know is the above things, not what the MHZ is or even how many pipeline stages the chip has.

spikey
Aug 30, 2001, 02:11 PM
true
although pipelines and processing units are relative to the speed and time a processor is at a certain task, which i think is what monkeybusiness means.

MasterX (OSiX)
Sep 4, 2001, 03:12 PM
Think about how great It was when Mhz ment something, soon no one will even hear of them, they'll be as useless as knowing pipeline stages was 3 years ago. Mhz were great at telling computers of the same class apart, not with DDR memory and such a variety of cache sizes and CPUs in Apple's line-up alone, CPU Clock Freq. is practically irrelivent except for configs of the same model (Quicksilver vs Quicksilver Mhz is useful, Quicksilver vs PowerMacG4 Yosemite is more useless)

MasterX (OSiX)
Sep 4, 2001, 03:21 PM
What we need is a test to measure "full system, real world performance" One test for the following:

CPU - Math Calculations w and w/o CPU acceleration (Altivek/3D-Now/ect)
FPU - Gigaflops
GPU - FPS in a time demo for OpenGL
RAM - time to copy 1GB inside RAM
VRAM - time to copy 50MB to VRAM from Ram and HD using CPU w and w/o acceleration
HD - Time to copy 1GB file to itself
Overall test - time to copy movie info into memory, trasform it (a/v codecs) then store it and replay it, all at 100% of the systems power (ie replay of the 30fps movie could occur at 200fps)

I think this would measure the system's most used tasks, and each task's results would get a number based on it's time to complete the task. Since all these tests can be done on any system on any OS, it makes it a good universal test, that would in fact benifit from special instructions on the CPU, like AltiVek, 3D-Now, or SSE.

mnkeybsness
Sep 4, 2001, 10:09 PM
your thoughts are flawed....


"HD - Time to copy 1GB file to itself "
this can depend on your hard drive itself
"RAM - time to copy 1GB inside RAM
this can depend on quality and amount of ram
"VRAM - time to copy 50MB to VRAM from Ram and HD using CPU w and w/o acceleration "
depends on the entire cpu/motherboard and ram and whatnot

thecube84
Sep 13, 2001, 07:20 PM
wait a minute-- lets just strip it down to the processor. If you just want to measure the CPU, nothing else, consider this.....

g= g4 mhz = 867 p = pentium mhz = 1400
a= g4 bit rate =128 b = pentium bit rate=64
(i think)

g4 speed = ga pentium speed = pb

867 * 128 = 110,976 bits processed

1,400 * 64 = 89,600 bits processed


so while the pentium has more cycles per second-- the cycles do less and the g4 has fewer per second, but a bigger bandwidth and therefore a faster process.

Catfish_Man
Sep 13, 2001, 09:54 PM
"your thoughts are flawed....


"HD - Time to copy 1GB file to itself "
this can depend on your hard drive itself
"RAM - time to copy 1GB inside RAM
this can depend on quality and amount of ram
"VRAM - time to copy 50MB to VRAM from Ram and HD using CPU w and w/o acceleration "
depends on the entire cpu/motherboard and ram and whatnot"

This is actually more useful than a processor only test. Most users (the ones who aren't building custom machines) want to know the speed of their computer, not the speed of their processor.

"g4 bit rate =128 b = pentium bit rate=64"
aren't pentiums 32 bit processors? I thought that was why the Itanium is supposed to be so spiffy, it's 64 bit.
As far as I know, it goes something like this
g4 bit rate=32b 128b floating point (or whatever Altivec accelerates) pentium bit rate=32b.

thecube84
Sep 14, 2001, 02:29 PM
yea i know i was just guessing off the top of my head. i guess I thought it was 64. I dunno. anyway it's still faster.

Catfish_Man
Sep 17, 2001, 10:50 PM
I suppose you could do a weighted average of cpu speed, ram latency/bandwidth, hd speed, cache, bus speed, etc... and come up with an "average performance indicator" or something like that. The problem would be, someone could skew the average by drastically increasing the speed of just one part.

MasterX (OSiX)
Sep 19, 2001, 08:16 PM
My test was designed to cover COMPLETE system performance. That's why some of the numbers (especially for Ram and VRam) are above the basic sizes of todays machines, but inside the bounds of a maxed out system. A system with over 1GB of ram and 64MB of vram will benifit greatly on this test. This makes the test good at showing how a system performs in different areas (Ram, Graphics, CPU) and overall. If you looked at a box for a G4 w/a 32MB Radeon and a ATA/66 it would be worse than the same procesor G4 with a GeForce3 and a U160 HD. This is why my test is useful. Not only do you see how much a game will benifit with more graphics power and a better HD, you can tell overall performance gains too.

A sidenote I just remembered, Intel's benchmarks are always very promising because each piece of their P4 systems are top of the line (fastest Ram, CPU, sometimes HD, Bus Rates) but when it comes down to the overall performance it tripps up. Just like the Pippin system, looked good on paper but...

Gelfin
Sep 20, 2001, 11:18 AM
This is definitely not a "flawed" idea. System-level testing is actually the norm for people who are serious about performance. Relying on raw processor speed is like a race car mechanic relying on engine displacement as the only measure of a car's performance. Typically only business customers are interested in spending the time and money to perform detailed testing and interpret the results. For consumers, as long as they've got the biggest number available and they can see a higher frame rate in Quake, they're happy.

This sort of thing was my job for many years It's sort of an art, and the single-number "result" you get out of marketing departments doesn't actually tell you much at all about what you're going to get when you plug it in and stick it on your own network.

spikey
Sep 20, 2001, 11:52 AM
You will find that nearly all companies like intel do that, apple do it too.
And yeah, it is not a fair thing to do.
Not only that, intel pays reviewers to give false test results and benchmarks. They give them anything from money to a free pc.

KingArthur
Sep 27, 2001, 03:14 AM
HELLO!!!!!! I don't think many actually understand it. Computers have an INFINATE number of variables affecting the speed of a computer. Can you say I-N-F-I-N-A-T-E? Come on! The freeking allignment of two molecules in one transistor on the video card can affect the outcome. So why not just get over this MHZ myth crap and just say "Yes, mac G4s are excelent at vector processing which is a big plus when making graphics because vectors are the infinate number of trajectories which goes into a 3D model, but the enormous clock speed of a Pentium won't let a certin program skip because MOST programs DON'T require vector processing intensive tasks."
Don't get me wrong or anything. I admire mac's quality before quantity approach, but as in every law of nature, there are always exceptions!