Sign in with
Sign up | Sign in
Your question

Memory Question for technically savvy

Last response: in Memory
Share
October 8, 2004 6:24:21 PM

From Anandtech article: "RAM - Memory Technology Overview"
http://www.anandtech.com/memor...oc.aspx?i=2223&p=5 (paragraph 5)

After a row in a memory chip is activated, it cannot be closed and another row in the same bank be opened until a minimum amount of time has elapsed. This is why having more memory banks can help to improve memory performance, provided it does not slow down other areas of the memory.

However, it is also known that data travels faster inside a memory chip than on the physical data bus and pin connections. Perhaps its safe to say that having fewer high density chips is better in this regard.


Wondering if you guys can shed some more light on this subject:

Which memory module configuration offers better performance in theory:

DDR400 PC3200:
1GB module 128Mx64, built using: 64Mx8, 16 chips, 2 memory banks
1GB module 128Mx64, built using: 128Mx8, 8 chips, 1 memory banks
October 8, 2004 9:01:57 PM

Hey, somebody PM that Crucial guy and let him have a crack at this one. I can't remember his name.

Mobile XP 2600+ (11X200)
Abit NF7-S v 2.0
Maxtor 60GB ATA 133 7200RPM
512MB Corsair Twinx 3200LL
BBA 9800 Pro
Enermax Noisetaker 420 watts
Win2K sp4
October 8, 2004 9:21:39 PM

I believe the guys at crucial and other manufacturers will go with the configuration that is more cost effective whether it be (64Mx8) chips or (128Mx8) chips. Between the two, usually (64Mx8) is a more cost effective configuration.

Also, many retailers hide the chips they are using from consumers by placing heat sinks on the modules. However, the heat sinks do offer better performance. In few cases the heat sinks can be used to cover countefeit chips. Crucial is a very reputable source and I highly doubt you would get bad modules from them since they typically use Micron original chips to build their modules.

But it still leaves the question of which configuration offers better performace ...in theory.

I am hearing that fewer banks/module offers better performance, but I am not sure. The reason being that it takes less time to find a block of data with fewer banks, less phyical connections, and fewer higher density chips.

This could be incorrect.
Related resources
October 8, 2004 10:55:28 PM

Quote:
I believe the guys at crucial and other manufacturers will go with the configuration that is more cost effective

Actually the guy from CrucialLabs (I think that is his user name) that hangs out here, is really good. He generally isn't try to sell anyone anything, just good for answering RAM questions like this one.

Anyways yes, you are correct the theory is that fewer banks/fewer modules offer better performance. However in real world tests this difference tends to be very minimal.

My Desktop: <A HREF="http://Mr5oh.tripod.com/pc.html" target="_new">http://Mr5oh.tripod.com/pc.html&lt;/A>
October 8, 2004 11:49:38 PM

Yes for PCs the difference is minimal.

But for example... if you have a server with a high volume of memory... such configurations could become an issue.

The more memory you use, the more significant small issues get.

I would be curious to know if this is an issue.

If it were possible to access two banks of a memory module simultaneously, then it might make sense to say that two banks is better than one bank.

2 banks = twice the money! haha. (ok lame joke)
October 11, 2004 6:39:00 PM

Ok, this would really take a long time to cover fully. Just think about it this way. In single channel mode the system can only access one of the modules using a 64bit data path. Now in dual channel (remotely like stripping in a raid config.) the motherboard is able to use both modules and the data path is extended to 128bits making dual channel a little faster. Fewer components on a module are not necessarily better.

What happens is on a double ranked module is two components are sharing the same data path so when information is being sent; the other component is in a stand by mode waiting for its turn to send data. There is little or no wait time for the information to start transferring, more of a seamless operation. On a single ranked module the information is sent through the pipe and after that information is sent it goes to a standby mode to wait for more data, more of a wait to receive process and not very efficient.

The 1GB modules only come in a 64x16 configuration. You will not find any 1GB modules using 128x8 that I am aware of.

I know I did not cover too much information but it all sums up to squeezing the most performance out of your system and right now the 1GB modules are going to stay 64x16 for a while longer. Once we get to 128x8 there will be a lot of happy people.

Funny BrentUnitedMem, hehe,

hope this helps to some degree.


Crucial Performance Lab

www.crucial.com
The Memory Experts(SM) at Crucial Technology is now on board to give you straight answers to your memory-related questions.
cpl@micron.com<P ID="edit"><FONT SIZE=-1><EM>Edited by CrucialLabs on 10/11/04 10:08 AM.</EM></FONT></P>
October 11, 2004 7:24:13 PM

Thanks for your reply Mr. Crucial Labs. Very informative. =P

I just wanted to add that 64Mx8, 16 chips is a popular configuration for 1GB modules.

And also that all though I'm not seeing many 128Mx8 configurations for 1GB modules, I have seen many 2GB modules built from 128Mx8, 16 chips.

cheers !
October 11, 2004 8:00:47 PM

Thanks, hmm I wonder who is making the 2GB modules using 128Mx8, its not us..

Crucial Performance Lab

www.crucial.com
The Memory Experts(SM) at Crucial Technology is now on board to give you straight answers to your memory-related questions.
cpl@micron.com
October 11, 2004 8:24:14 PM

There are a few module manufacturers, including ourselves, who are building 2GB using 128Mx8, 16 chips.

Though we do not represent the retail market.
!