CAS2.5 vs. CAS2

[billd222]

Howdy all,

Is there really a big difference between PC2700 cas2 & CAS2.5? I know 2 is supposed to be faster but is it that much faster?
I have Gigabytes GA7VRXP. Right now I only have 256mb of Corsair PC2700 CAS2.5.
I need to increase my memory to about 725-1Gig.
Would it be worth it to chuck my CAS2.5 and start over with CAS2.0?
Thanks?

__________________
Rock me Amadeus,

Bill D.

 

[billd222]

hello??
Anyone??
I checked the faq and this question wasnt in it.

__________________
Rock me Amadeus,

Bill D.

 

[gaviota]

The performance difference is not enough to throw away your 256 MB of CAS 2.5 RAM.

__________________________________________________
It's not important to know all the answers, as long as you know how to contact someone who does.

 

[nja469]

I suppose it's not worth the effort. But then there are ppl like me thst need everything as fast as possible. Personally I'd ebay the current [-peep-] and get some good CL2 RAM, but again that is me. No flaming ppl. I don't need to hear that I'm weird, I know that already )

"Opinions are like assholes, everyone has one"

 

[billd222]

Thanx for input.
How unstable would it be if I mixed different brands of PC2700 ?(If I got samsung DDR as opposed to another corsair)

One more question:
I've always mixed different sizes of ram with no problem --such as a 256 dimm+ a 512 dimm with no problem.

My friend ordered more ram from dell (where he got his comp) and the tech support guy said that he should order the same size dimm as his last dimm and always install in pairs. as in only install dimms in sets of twos or fours.

Is there any truth in this? I told my friend that it was absolute bull-shyte. I hope I gave him the right advice.
so

__________________
Rock me Amadeus,

Bill D.

 

[nja469]

No, even though it came from Dell it's not total BS. It's recommended for best compatiability and for best performance to get identical chips. Having mismatched chips doesn't exactly spell out doom, but it can in some instances, therefore it's best to avoid the possibility altogether.

Try to go same brand, same model, speed.. etc. As identical as possible. It's even better to get one module instead of 2. i.e., one 512MB module over 2 / 256 modules.

Some chipsets get weird with more than one module, even when identical. This used to not be such an issue as it is today, but it seems with the newer chipsets and memory it is.

 

[xxsk8er101xx]

it's just a delay. The less the delay the faster.

Delay plays a role in performance. Delay, aka Latency, times the bandwidth is your total throughput.

So if your delay is 2.5 (CAS2.5), and your bandwidth is pc2400 it would equal about a CAS2.0 PC2100. I could be wrong but thats what i noticed.

<A HREF="http://www.anandtech.com/mysystemrig.html?id=9933" target="_new"> My Rig </A>

 

[gringocorp]

Ah ha. I had a similar question. I am looking at a SiS648 based mobo and thought about getting DDR400 (XMS3200) but I can only find CL2.5 memory. So according to your calcs the DDR333 (PC2700) at CL2 would be quicker.
2700/2 > 3200/2.5.
Is this definitely correct?

 

[HammerBot]

So if your delay is 2.5 (CAS2.5), and your bandwidth is pc2400 it would equal about a CAS2.0 PC2100. I could be wrong but thats what i noticed.

No! The bandwidth of PCx is independent of the latency. Latency IS only latency (or delay if you prefer) and becomes significant when performing many memory accesses to different locations. This reduces the performance gain in making long burst, and latency becomes significant. However, if you access sequential memory locations, you can use long burst. In this case CAS2.5 vs CAS2 only means that you get your data a few nanoseconds later. But the effective bandwidth is the same. This is the reason why the performance gain of CAS2 is very small (probably only a 1-2%) and also highly dependent on the benchmark method. I.e. benchmarking with large amount of data at consequtive memory locations will show almost no difference. Whereas benchmarking with single byte accesses to random memory locations shows a larger difference.

 

[billd222]

wow, thanx for all the help everyone.

__________________
Rock me Amadeus,

Bill D.

 

[xxsk8er101xx]

according to the benchmarks i've seen, yes.

<A HREF="http://www.anandtech.com/mysystemrig.html?id=9933" target="_new"> My Rig </A>

 

[xxsk8er101xx]

what? your post is very cryptic.

and about your "probably only a 1-2%)", is actually wrong. it's from 1% to 14%. .. and from how you say it, it sounds like you are guessing so why should anyone believe what you said.

My finding is from benchmarks which you can easily look up and find for yourself if you have a brain.

so there you have it... look it up isntead of the probability that it might be 1% or 2%. right?


<A HREF="http://www.anandtech.com/mysystemrig.html?id=9933" target="_new"> My Rig </A>

 

[HammerBot]

No need to be rude. If I have made a cryptic post there are ways for you to ask me to clarify it.

I have seen different benchmarks getting highly different results. As I explained in my first post benchmarks scores are very dependent on exactly how the memory is tested. In real life applications where you have mixed accesses to the memory (large blocks small blocks etc.) I estimate a 1-2% performance gain.

It is really not very interesting that the CL2 memory has 14% performance gain over CL2.5 memory in some situations where you perform non-bursting accesses to random memory locations. Such accesses are very seldom and hence overall the 14% does not weigh a lot.

Thats it. Did I clarify it enough so your brain is capable of conceiving it or do I have to spell it out?

 

[Grub]

OOOOh....Ooooh....let me try to spell it out....
F-R-A-G-G-E-D...fragged...that spells fragged. Way to drop the hammer...bot

Just because you're not paranoid, doesn't mean they're not watching you.