At what memory speed does CAS latency no longer matter?

[MichaelD]

I've wondered this for over a year now and haven't been able to find a definitive answer.

We all know that CAS latency plays a big part in memory performance. That's how come we always try to set our memory latency (CAS) to 2 as opposed to 3 or 2.5

But, as we reach a certain memory clock, the memory won't stably run at CAS 2 anymore, so we lower it to 2.5 or three.

My question is; at what memory speed does the speed negate or outweigh the fact that the memory is now at a higher latency? I hope that makes sense. Thanks.

 

[Lord Evermore]

Let me try to translate, tell me if I'm right:

Would 133MHz memory at CAS3 be faster than 100MHz at CAS2? Or maybe would it take 150MHz to allow CAS3 to still be faster?

I think that's what you're trying to ask. It would depend on the exact memory type, but also the particular use you're putting it to. For some applications, very high bandwidth is more important than latency. For others, lower latency is more important. There isn't a particular speed that would be the magic point, rather the amount of difference between one speed and another.

I don't have any hard numbers, but I think many people were willing to switch to CAS3 to run memory at 133MHz when PC133 CAS2 was still expensive. We see that higher bandwidth can be very very useful, as we see that RDRAM, even with higher latency, is a very good performer.

 

[Shooters]

Well, I've always been told that going from CAS2 to CAS3 results in only a 3 or 4 percent drop in memory performance. So, I guess as long as your reference memory clock is no more than 4 percent lower than your overclocked memory clock, then the increase in memory speed outweighs the higher CAS latency.

Example:

133MHz = reference memory clock.

overclocked memory - .04 * overclocked memory = 133MHz

Therefore, overclocked memory = 138.5 MHz

So, I guess 138.5 MHz would be the point where the benefits of a higher memory clock would outweigh the drawbacks of going from CAS2 to CAS3.

Of course this is all assuming the average 3 to 4 percent difference I mentioned earlier, but like Lord Evermore said, the exact difference in performance would depend on the application being run.

 

[KenAF]

Latency always matters.

That said, some applications / games benefit more from the increased bandwidth (through higher FSB) than they do from lower latency.

 

[MCS]

I pesonally always like to go to the highest memory speed @ CAS 2 that my memory will go. I don't even try CAS 3...but that's just me.

 

[MichaelD]

Nice replies, thanks! I guess I'll have to do some experimentation/benchmarking to find out, huh? My new setup allows me to change the memory clock independent of the CPU FSB...that's a first for me. (Shuttle AV35GTR).

Right now, everything is bone stock except that the memory is set to CAS2 and 4-way interleave is enabled. Have a nice day!

 

[miken]

From Crucial..






What is the performance difference between CL2 and CL3?


Description:

What is the performance difference between CL2 and CL3?



Solution:

CL2 parts process data a little quicker than CL3 parts in that you have to wait one less clock cycle for the initial data. However, after the first piece of data is processed, the rest of the data is processed at equal speeds. Latency only affects the initial burst of data. Once data starts flowing, there is no effect. Bear in mind, a clock cycle for a PC100 module is 10 nanoseconds so you probably won't notice a significant performance difference. Most systems will accept either latency part. However, there are some systems that require either CL2 or CL3 parts. These requirements are built into our Memory Selector.

I doubt anyone can see the difference between CAS2 and CAS3. Sometimes I think we would be better of without such things as Sandra and benchmarking, too much hype IMHO.

 

[Kilrsat]

From my benchmarking tests, I was getting the same scores at 145mhz cas 2 as I was at 158mhz cas 2.5 (this is with ddr)

This was fairly consistent on both my 8k7a and Dragon+, so for me I was looking at needing a ~8% memory speed increase to overcome the CAS latency change.

 

[MichaelD]

<< From my benchmarking tests, I was getting the same scores at 145mhz cas 2 as I was at 158mhz cas 2.5 (this is with ddr)

This was fairly consistent on both my 8k7a and Dragon+, so for me I was looking at needing a ~8% memory speed increase to overcome the CAS latency change. >>



Now THAT is the kind of info I'm looking for! Thanks, Kilrsat! I'd like to squeeze some more performance out of this rig; I know there's more to be had. But I don't want to sacrifice any stability in the process. (There I go...living in that ideal world again....)

 

[CraigRT]

<< I pesonally always like to go to the highest memory speed @ CAS 2 that my memory will go. I don't even try CAS 3...but that's just me >>

 

[pillage2001]

First of all, does anyone here even know what RAS and CAS do?? I know but I'm too lazy to explain it here.

CAS latency will always matter no matter how fast is your RAM.