Frequency

silvater

Junior Member
May 1, 2002
17
0
0
In computers frequency is a term that is very common. but not many of us understand what it really means. We just memorized the numbers(133Mhz, etc) and repeated like robots.
I want to understand what do the frequencies show us, It sounds very general Im going to try to break it down.
What I understand (I could be wrong) is that most computers parts received a frequencie from a clock generator and they multiplied/Divide this frequency. This frequency is measured in Megahertz per second.
I understand until here. So if we say the FSB has a frequency of 533Mhz what do we really mean, how much data are tranferring?
Thanks :confused:
 

Demon-Xanth

Lifer
Feb 15, 2000
20,551
2
81
Lately computers have been using the tricks of "effective frequency".

The Athlon takes the FSB and multiplies it by two, the P4 multiplies it by 4. So they're really still back in the days of 100 and 133MHz clocks. They're just transferring more data per clock.

Each line can carry one bit at a time, unless multiplexed (such as DDR, QDR, and RSL), so you need 8 lines to transfer a Byte. Systems typically have a CPU w/ 64 bit data bus, 64 bit AGP, and 32 bit PCI. The RAM is usually at 64 bit for DDR/SDRAM, and 2x16 bit for RDRAM.

In the case of DDR, a bit can be transferred at both the rising edge and the falling edge of a clock. In the case of QDR there is effectively 4 logic levels allowing two bits to be transferred per transition.

For the max _THEORETICAL_ bandwidth, you need to multiply the bus width, but the bus speed, by the bus's effective multiplier.
 

Evadman

Administrator Emeritus<br>Elite Member
Feb 18, 2001
30,990
5
81
I saw a really good picyure on the breakdown somewhere, but I forget where. It had a square waveform, and arrows use to denote when data was being transfered.

SDR transfers data only on the "rising" edge of the clock
DDR transfers on botht he rising and falling.
Intel's transfers twice on the rising and twice on the falling.

The "rise" is going from 0v to 1.7v ( or CPU voltage I beleve ) and falling is back from 1.7 to 0v.

The pulse is generated by the "clock generator" and is then multiplied up to meet the different speeds of different parts of the board. 33.3 for PCI bus, 100-133 for CPU / memory ect.

The bus is made to run @ a specific speed or slower as running it faster can over stress components. The 1st to be overstressed is the PCI cards and bridge. that is where most overclocking attempts go awary.

Unrelated: I remember doing a calculation to figure out what the max speed the PCI bus could run at as 1.0 ghz. Then you run into the light speed barrier. ( assuming 12" x 12" board ) depending on placement of clock gen. any faster and 2 pulces will be on the board at the same time.

<edit>
The bus speed you are talking about (533) is actually only a 133 bus, but since intel transvers data 4 times per clock cycle, you end up with a 533 "effective" rate. Athlons and DDr do not actually run on 266 mhz bus speeds, that is effective rate. They run @ 133, but since they transfer data 2 times per clock cycle, then you end up with the effective 266 rate.

Oh, and DDR memory already has the theoretical bandwith labeled. 2100 DDR can transfer 2.1 GB of data, 1600 1.6 and so on.
 

imgod2u

Senior member
Sep 16, 2000
993
0
0
The word "frequency" is kinda self-explanatory, it's how frequent something is. Frequency in terms of computers is how many signals per second, or rather, how many electrical pulses are sent per second. This can have vastly different properties considering a pathway can send 10 signals per second (and receive also, called a clock) but have a much wider pathway, meaning each signal carries more data or it can have a very narrow pathway and send 10000 signals per second.
 

TalShiar

Member
Nov 29, 2001
32
0
0
Clock speed doesnt = computing power.
One chip could multiply 2 numbers in 30 steps running at 500 MHz while another processor layout can do the same operation 15 steps while running at 350 MHz. After doign that a few thousand times the second proc will be done first. That is why we use benchmarks to compare computers.
Then you add in bandwidth and latency for all the other devices in the computer and it gets really complicated. So we use things like FLOPS (floating point operations per second) to comare different computers.
 

Belegost

Golden Member
Feb 20, 2001
1,807
19
81

This frequency is measured in Megahertz per second

Just to clarify what the units of frequency are, and mean. Frequency is the number of cycles an oscillating object goes through in one unit of time. For instance, RPM (revolutions per minute) is a frequency, it tells you how many time the object completes an entire revolution per minute. The units of Hertz are cycles per second, a cycle can be different things depending on the system in question, but for waves a cycle is the full wave, meaning you can completely reproduce the wave by copying the cycle many times, from peak to peak generally. So if we had a wave like this:

/\_ _/\_ _/\
\/ \/

I hope that comes out looking right... Then a cycle could be:

\_ _/ or
\/

_/\_
\/

The prefix mega- tells you that the original units have been multiplied by 10^6, so that gives megahertz a value of 1 million cycles per second. As this relates to computers, the data signal is a square wave, the the ons and offs coming from the lows and highs on the wave. So you can see as the frequency of the wave increases, so does the amount of information that can be moved per unit time, more 1's and 0's every second.

Of course this doesn't take into account all the marketing nonsense...