New thread from another post to discuss RAM chips - What is the best configuration?

First the comment that prompted the question?

>>>>>... I looked up your motherboard's notes r.e. memory population:   if you use single-sided modules you can install four modules with no reduction in latency/speed.   If you are using double-sided modules there will be a reduction if you install four modules.   In fact, it doesn't just reduce the CAS latency, but slows the memory down to DDR333 speed.   So if you want to maximize both memory and speed, you need to use 2 x 1GB single-sided modules and 2 x 512mb single-sided modules.

Here's the specific comment from MSI:

The Maximum memory speed decreases when the following two memory combination is selected:
- Each channel is installed with two double-sided memory module
- Both DIMM1 and DIMM3 are installed with double-sided memory module<<<<<


Mobo is an MSI K8N Diamond Plus currently running two sets of 1GB dual channel two sided chips.  For those not on the other thread, it has 4 slots but will only run 3 GBs (so yes I wasted some money there.  Also, trying to get the optimal setup for my PC)  I'm not sure how much difference it makes for the chips to be dual channel.  Also I'm assuming that single sided chips can be dual channel if they are identical.  That being said here is my bottom line question.  Would the opitmal chip configuration for my mobo be 1 set of 1GB dual channel single sided chips and 1 set of 512 dual channel single sided chips?

If so, where can you find single sided chips.  Maybe I'm not searching correctly but I can't find them (at least not on Newegg).
Who is Participating?
>Are you seeing all of the memory

Some motherboards show 3GB without the /3GB switch, such as this one and the one I have, an Asus A8R-MVP.

>I'm not sure how much difference it makes for the chips to be dual channel.  Also I'm assuming that single sided chips can be dual channel if they are identical.

Dual channel doubles the bandwidth, and can increase the total system speed by up to 10%, so by all means go for it.  I think the pair of 1GB modules and pair of 512K modules in single sided versions is about as good as it gets.
Aslo known as MS-7220
Check this link for more info.
Are you seeing all of the memoryThe virtual address space of processes and applications is still limited to 2 GB, unless the /3GB switch is used in the Boot.ini file.

I believe the operative spec here is single rank versus dual rank modules.  All the single rank 1GB modules I could locate were ECC modules for server use.

This is a good memory primer.  Read last paragraph in "SDRAM modules: Organization" section.
How do you know if your security is working?

Protecting your business doesn’t have to mean sifting through endless alerts and notifications. With WatchGuard Total Security Suite, you can feel confident that your business is secure, meaning you can get back to the things that have been sitting on your to-do list.

abpowellAuthor Commented:
Ok great.  A link to find these?  I guess I'm not searching with the correct keywords because I don't see them
abpowellAuthor Commented:
Here is Windows example of how to switch
[boot loader]
[operating systems]
multi(0)disk(0)rdisk(0)partition(2)\WINNT="????" /3GB

Note: "????" in the previous example can be the programmatic name of any of the following operating system versions:

Windows XP Professional.......

Here is what I show:

multi(0)disk(0)rdisk(0)partition(1)\Windows="microsoft xp professional"/noexecute=optin/fastdetect

I can't tell if it's on or not
Gary CaseRetiredCommented:
The /3GB switch is NOT needed to take advantage of memory beyond 2GB.  The 2GB limit is for the virtual address space allocated to applications -- but the executive software (i.e. XP) will use any additional memory ==> so effectively having more than 2GB frees up all of the memory that XP would have used (since it will utilize the additional GB).

You definitely want to operate in dual channel -- that doubles the bandwidth for successive memory accesses => common when filling the cache after a cache miss.  But you're operating in dual channel - and will continue to do so as long as you use matched pairs.   The difference is that you will operate at 400MHz (dual channel) with 2GB of your current memory; or at 333MHz with 3GB (still dual channel).
Switching to single rank modules would simply let you use 3GB at 400MHz.   Personally, as I noted in the previous question, I'd just keep what you have if you're satisfied with the performance; or unplug two of the modules and run with 2GB at 400MHz  (in any event, I'd benchmark it with both configurations).

Your memory will be faster with 2GB -- but whether you'll notice that;  or whether you're using applications that would benefit from the additional space a 3rd GB would provide the system, is difficult to predict.   It is also, as willcomp noted, difficult to find single sided 1GB modules.   Last time I found them they were nearly double the price -- but I couldn't find any in a brief search tonight.

Gary, isn't that single rank vs single sided?  Single rank 1GB ECC modules are double sided (10 chips).

Generally, but not always, current 256MB and 512MB single sided modules are single rank and double sided modules are dual rank.  I suspect manufacturer was just trying to make choice simple by specifying single sided modules.
Gary CaseRetiredCommented:
In this case I think the manufacturer is right in specifying single sided.  The issue is the number of chips;  since every chip represents a load on the address bus with unbuffered memory.  It's true that a single sided module WILL be single rank (since it will use x8 chips).   Of course a double-sided module with x4 chips could also be single rank, but in that case it would use 16 chips -- and have twice the load on the address bus.  There are two issues at play here:  (1)  the addressibility of the memory -- that's where the single/dual rank issues can come into play; and some systems have limits on these; and (2)  the load on the address bus -- which is where the # of chips is the driving (no pun intended) factor.

The first issue exists with either buffered or unbuffered memory;  the second issue is not a factor with buffered (registered) modules -- since in that case there's only one load per module instead of per chip.

As we've discussed before, however, I do agree that the memory makers could make this whole issue a lot easier if they would include more details on the physical organization of their modules in the descriptions.
abpowellAuthor Commented:
God you guys make it hard to pick an accepted answer.  I appreciate ALL the info.  I'm going will Callandor on this one though (Just because he agreed with me on the setup, you other guys gave me more to think about though.  Thanks again.  This is the BEST site for info on the entire web!
abpowellAuthor Commented:
I'd give an A+ to all if I could.  Don't forget to look for my new questions.  Probably tomorrow, maybe tonight though.
Thanks for the kind words.  Glad to help.
Gary CaseRetiredCommented:
abpowell -- I was writing up another answer and thought you might appreciate a bit more detail on WHY the system slows down when you install four modules.   Here's what I wrote for the other question r.e. memory accessing issues:

The problem is conceptually simple:   When memory is referenced, the address is placed on the address bus by the memory controller, and then the control signals that indicate a good address, and that strobe the actual transfer of the data (after the appropriate CAS/RAS delays, etc.) are sent.   These signals go to ALL of the memory chips in the system -- so, for example, if you have 4 memory modules with 8 chips on each you're loading these signals with 32 "receivers".   Note that if the modules are double-sided you would have 64 "receivers" !!   This notably degrades the signal -- if you looked on an oscilliscope (remember those??) you wouldn't see a square wave anymore;  you'd see a much-less-clean signal with poorly defined leading and trailing edges.   That's why many systems automatically clock the memory slower if you have more than 32 "loads" (i.e. use double-sided modules and install more than 2).

Buffered memory (i.e. "registered" RAM) uses a small buffer chip on each memory module to buffer the control lines -- so the load to the system is ONE load per module, instead of one load per memory chip.   MUCH more stable -- the signal stays nice and "square" and the clocking is much more consistent.

How many errors are caused by unbuffered memory??   Probably very few.   And if you use ECC memory (I always do) those that you do get are probably always corrected.   But many server motherboards support more than 4 RAM modules -- and with larger numbers of memory chips the load factors get far too high unless you use buffered modules.   If it wasn't for the higher complexity (and cost) ALL memory modules would be buffered -- electrically it's MUCH more reliable.  Note that this is why systems sometimes run fine with TWO memory modules, but have "glitches" when you use four modules.   It's also why single-sided modules are preferable to double-sided modules (1/2 the electrical loading).
Question has a verified solution.

Are you are experiencing a similar issue? Get a personalized answer when you ask a related question.

Have a better answer? Share it in a comment.

All Courses

From novice to tech pro — start learning today.