 |
 |
G3 Calculations Per Second
|
 |
|
 |
|
Fresh-Faced Recruit
Join Date: May 2003
Status:
Offline
|
|
How many billions of calculations can a 900MHz G3 processor perform per second?
|
|
|
| |
|
|
|
 |
|
 |
|
Mac Elite
Join Date: Mar 2001
Status:
Offline
|
|
Originally posted by Cykon:
How many billions of calculations can a 900MHz G3 processor perform per second?
Heh.
0.9 billion, to be exact (900 million).
(A 900MHz processor can perform 900,000,000 operations per second. 1 Hz = 1 cycle/second. 1 MHz = 1,000,000 cycles/second. However, there are many more factors to how many operations a processor can effectively perform. For raw, non-vectorized mathematics, the performance is tied very closely to the clock speed, no matter what kind of processor it is. For more complex and specialty tasks, there are many more factors. This is where the "MHz myth" comes in. For some background, see: http://www.apple.com/g4/myth/.)
|
|
|
| |
|
|
|
 |
|
 |
|
Professional Poster
Join Date: Jul 2001
Location: Dis
Status:
Offline
|
|
It's actually more complicated than that, piracy. Processors are divided up in to units capable of doing different types of work (integer, floating point [numbers with decimals], altivec, etc], and many processors can use more than one of these units at the same time. So the G4, for instance, can do three instructions per cycle, IIRC. This is a theoretical maximum, though, and far from what actually happens. Two important factors come in: the processor has to have three things it can do at the same time in different functional units (not easy, especially if the chip doesn't "look" for things to do, and relies on the compiler programmer to set things up optimally, like the G4), and any time there's a branch in the code (where what the next instruction is depends on some condition) the processor "guessing" the wrong branch to execute will lose valuable cycles. The second comes from the fact that processors execute instructions in stages (7 for the G4, like 20 for the P4) and thus it actually takes 7 cycles to complete any one instruction, but like an assembly line it can have each stage working on an instruction.
As for the G3, I don't know. If I recall correctly (big if, actually), the G3 has 4 stages, two scalar units, on floating point unit, and I don't know what else, or how many instructions it can complete on any one cycle. G3s are actually pretty sad chips in terms of sticking them in full blown computers and expecting raw performance. They're probably as much as most people will ever need, but still...
BlackGriffen
|
|
|
| |
|
|
|
 |
|
 |
|
Mac Elite
Join Date: Mar 2001
Status:
Offline
|
|
Originally posted by BlackGriffen:
It's actually more complicated than that, piracy.
Yes, indeed. Short of writing a book here on microprocessor architecture and instruction processing techniques, I think the generalized answer about clock cycle is fairly to-the-point in terms of one type of "operations per second". Though you are correct: it is a lot more complicated than that.
|
|
|
| |
|
|
|
 |
|
 |
|
Posting Junkie
Join Date: Sep 2001
Status:
Offline
|
|
Oh, come on, I want the whole book on it.
|
|
|
| |
|
|
|
 |
|
 |
|
Senior User
Join Date: Nov 2000
Location: New Orleans, LA
Status:
Offline
|
|
For what it's worth, my beige g3 @ 300mhz does 91 million floating point calculations per second.
I used XBench to get this figure.
-vasu
|
|
|
| |
|
|
|
 |
|
 |
|
Mac Elite
Join Date: Feb 2000
Location: Nashua NH, USA
Status:
Offline
|
|
|
|
|
|
| |
|
|
|
 |
|
 |
|
Mac Elite
Join Date: Oct 2001
Location: Colorado Springs
Status:
Offline
|
|
Originally posted by piracy:
Yes, indeed. Short of writing a book here on microprocessor architecture and instruction processing techniques, I think the generalized answer about clock cycle is fairly to-the-point in terms of one type of "operations per second". Though you are correct: it is a lot more complicated than that.
yes, some instructions take longer than others. A simple ADD OR MOV can usually be executed in one cycle, while floating point operations tend to take longer, not to mention LOADs and STOREs to main memory (since you have to wait on the DRAM). Of course sometimes you get lucky and what you want is in cache. But that's up to the MMU, most of the time.
Of course, I'm sure it gets even more complicated. In my computer architecture class we focused on some of the earlier (and therefore- more pure) RISC designs. Because they were easier to understand (we were only undergrads, after all).
|
RhythmScore
iMac 27" Quad i5 | PMG4 2x867 (RhythmScore test server) | iPhone4
|
| |
|
|
|
 |
|
 |
|
Mac Elite
Join Date: Aug 2001
Status:
Offline
|
|
Both the G3 and the G4 average 2.31 scalar integer operations per cycle (dhrystone mips or however you spell it. It's probably all cached stuff). So a 900MHz G3 would do 2.31*900 integer ops per second. The G4 can do much more than that when it uses Altivec. Also, the correct address for that page is
www.arstechnica.com
not
www.ars-tecnnica.com
|
|
|
| |
|
|
|
 |
 |
|
 |
|
|
|
|
|

|
|
 |
Forum Rules
|
 |
 |
|
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts
|
HTML code is Off
|
|
|
|
|
|
 |
 |
 |
 |
|
 |
|