Memory Knowledge

[Shez]

Hey all - I'm on a learning rampage as I get set to build a new rig, so I'd like to ask the memory section some overall questions.

:?: In general is it better to get 2 sticks of dual channel memory or 1 stick of the same ammount? I could imagine that reading from 2 sticks would be faster but what do I know

:?: I know having a lower CAS latency is best but is there a significant difference between CAS 3, CAS 4 and CAS 5 memory? This is assuming that it's installed on a good rig, running memory intensive games.

:?: Finally, could someone dumb down the meaning of memory timings for me. I know that they come in a variety of flavors 4-4-4-12 or 5-5-5-18 etc... but I'll be damned if I know what the H those numbers mean. And just like with the CAS latency, is there going to be a significant difference in preformance between a 4-4-4 and a 5-5-5?



Thank you all in advance for any insight you have and any help you may give.

 

[fueled_by_ramen]

Well, I came onto the Memory Forums to ask some questions myself, since I'm building my first AMD system and from what I understand, the Athlon's integrated memory controller is sensitive to certain types of RAM (as opposed to Intel's chips).

From what I do know, I believe I can answer your first question and at the same time bump this thread a bit so it doesn't go unnoticed. I'll hold off from my questions for now...don't want to hijack your thread.

It is better to have 2 sticks of dual channel RAM as opposed to one big stick. ie. 2 x 512MB in dual channel is better than just having one stick of 1GB.

I can kind of extrapolate from what little bits and pieces I know regarding your other questions, but it would be best to wait for a "Memory Expert" to come on and answer them for sure. If you are building an AMD system, I believe it's best to get memory with as tight of timings as possible because of the integrated memory controller. The lower the latency the better, so CAS 5 wouldn't be the best choice for an AMD system. It might be adequate for an Intel chip however.

Hopefully someone with more experience can come on and explain it...since I need clarification as well.

 

[nobly]

:?: In general is it better to get 2 sticks of dual channel memory or 1 stick of the same ammount? I could imagine that reading from 2 sticks would be faster but what do I know

Generally dual channel is better, but real-life improvements are under 10%.

:?: I know having a lower CAS latency is best but is there a significant difference between CAS 3, CAS 4 and CAS 5 memory? This is assuming that it's installed on a good rig, running memory intensive games.
:?: Finally, could someone dumb down the meaning of memory timings for me. I know that they come in a variety of flavors 4-4-4-12 or 5-5-5-18 etc... but I'll be damned if I know what the H those numbers mean. And just like with the CAS latency, is there going to be a significant difference in preformance between a 4-4-4 and a 5-5-5?

Yes if you're overclocking. Not o'c? Not really or negligible.
Reference on memory timing definitions:
http://www.tomshardware.com/2004/01/19/ups_and_downs/page3.html

Read this if you're more interested in overclocking w/ the CAS timings.
http://forumz.tomshardware.com/hardware/modules.php?name=Forums&file=viewtopic&t=223944&highlight=

Also you will probably find the Memory Sticky useful (please read this first before asking any more questions - saves us typing)
http://forumz.tomshardware.com/hardware/FAQ-read-posting-ftopict55024.html

 

[qwertycopter]

1. Yes, dual channel would be superior to single. However, I doubt it will make much real-world difference in terms of performance.

3. So there are four timings:

# of clock cycles to access a particular column

# of clock cycles to access a particular row

# of clock cycles it takes to start read/write once column and row have been selected

# of clock cycles to jump from the current row to a new row

Generally you want the lowest timings possible, but performance also depends on the combination of the four timings. You want the combo to be in sync (memory access is pipelined). If timings are out of sync you start hitting wait-states, which result in additional latency.