Ratio v FSB

I read several explanations on memory ratio but really still fail to fully understand it. The simple explanation I seen a few times is that anything higher then 1:1 ratio is like speeding between syncronized traffic lights which force you to stop anyway. Not really understanding it still makes me wonder which of the following settings will deliver better overall performance.

1. FSB 3.2 GHz, memory ratio 4:5, memory speed 800 MHz or
2. FSB 3.33 GHz, memory ratio 1:1, memory speed 667 MHz

well it is kinda like that... but the higher the mhz the faster those lights go from red to green and back.
The other part is like... say your fsb is light a and d, your ram has to stop at those lights but it has other lights b and c between those lights... and the faster those go, the quicker traffic will get to light d. Dunno if that makes any sense... if lights b and c are slow less info will get to d before it changes.

Two men are moving a pile of bricks.
Man A (Mr. Memory) has to pick up one brick at a time and give it go man B (Mr. Processor)

1.
As Mr. Memory hands each brick to Mr. Processor, he finds that Mr. Processor is waiting with his hands on his hips because Mr. Memory is just not going as fast as Mr. Processor. (this would be an example of a 3:4 ratio)

2.
Now they have to move the bricks back to where they started. As Mr. Processor is giving the bricks to Mr. Memory, Mr. Processor is having to stand there with the brick and wait for Mr. Memory to take the brick from his hands and put it down next to him. (this is another example of 3:4, but going to opposite direction)

3.
After going to the gym for a couple of weeks, Mr. Memory is a little faster, and is able to keep up with Mr. Processor. Now when they move their bricks, neither of them wastes time waiting for the other to pick up a brick, or put one down. (an example of 1:1 timing)


...Well, at least I tried

I think thats a fairly good basic way to state it... however I'd compare it more to them having to fill up a wheelbarel.... since they move more than 1 brick at a time.
However many bricks Mr. memory can get put into the barel (latency) AND taken to Mr. CPU (speed) by the time Mr. CPU has unloaded his bricks would be more like it.

Wow this is an eye opener. I never imagined my computer besides being a driver is also builder and barrel pusher as well. Actually I am quite surprised coz I have an acrylic case and can see everything happening inside it and I dont see any construction activity going on inside.

So based on the explanations I guess you are saying that the speed of all this racing around and moving of bricks individually or in barrels is really dependent on Mr Processor and not at all on Mr Memory so long as the memory ratio is at least 1:1.

Meaning I have to take out a gym membership for Mr Processor to try to get him more fit and run faster. Thanks guys. Have one on me!

You Almost Have it.

The point I was trying to make is that the slowest one is what will be limiting the bandwidth between the processor and memory. At 1:1, Neither is slower, so both are working to their full capacity. at 3:4, either the ram or processor will be doing some waiting around for the other to catch up.

Ok got it that is why the OC as in my sig is now as is.

In my rig's real life that means that I can't run my memory at rated speed coz I cant OC my CPU to more then 3.5GHz. Well I can but the temps go to high so I backed of to 3.45GHz. Will try to see if I can then tighten the memory timings to pick up the loss of speed.

you could try lowering the multiplier to get a higher FSB.

Already tried it a few weeks ago and couldnt make it work. Considering the changes since then its worth another try. Thanks will do.

If Mr memory is faster than Mr CPU... Mr memory has to wait around... So whats the point of buying high end ram eg Dominator 8888 at 1111Mhz.

Ram would run a 555 x 2 = 1111 and Mr CPU would be doin a quad pumped ~500 (If your lucky )

This would mean that MR CPU will never be able to keep up with Mr RAM...

Well I am not fully confident that this is true but the way I understand it this additional unused speed allows some tightening of timings which you cant do if you are running your memory at its top speed or above.

Timings are supposed to be a somewhat lesser factor in overall performance but if you are trying to push your system to the limit it may be what you decide in spite of the high price.

MrWizard2U said:

Well I am not fully confident that this is true but the way I understand it this additional unused speed allows some tightening of timings which you cant do if you are running your memory at its top speed or above.

Timings are supposed to be a somewhat lesser factor in overall performance but if you are trying to push your system to the limit it may be what you decide in spite of the high price.

Click to expand...

Wizard, what you said is the way I understand it. In my current rig I'm running Ballistix ddr500. I'm running it slower that it's rated speed and it tightens up the timings. Just my observations.

Faster memory speeds do not always mean higher performance. Chipset straps and latencies may actually result in a high FSB/memory config being slower than lower FSB/memory settings on some boards. You really need to run some benchmarks in Windows to determine which gives better performance. Everest is pretty accurate, the trial version will give you memory read/latency numbers and is a good place to start.