Jump to content

  •     

Banks, Ranks, & Double-sided Ram


  • You cannot reply to this topic
6 replies to this topic

#1 jeffk

jeffk

    Sr. Active Member

  • Active Member
  • PipPipPipPip
  • 11 posts
  • Gender:Male
  • Location:UK

Posted 27 November 2011 - 08:32 AM

I'm studying for my A+ & am trying to get my head round the relationship between memory banks, memory ranks & single-/dual-sided memory.

Reading around, it seems that 'banks' is used in similar but subtly different ways:

1) as a group of sockets or modules that make up one logical unit, with the logical unit referring to a match between the processor's data bus width and the RAM's bit width e.g. one 64-bit DIMM or two 32-bit DIMMs makes a full bank with a CPU with a 64 bit data bus.

ok, that makes sense

2) but then other people refer to banks of memory addresses, or as banks of "switches" in a RAM chip (each switch being 1 transistor/capacitor set)

3) & then Scott Mueller says that dual-ranked memory has internal banks of memory chips called ranks. (He was explaining why single-/double-sided memory has nothing to do with whether there's chip one one or both sides of a RAM stick, it's to do with memory ranks).

I think my brain's gonna explode!  biggrin.gif


Any help from the more theory inclined techs here would be much appreciated. Thanks!


Also, which is faster, single- or dual-rank (often called single-/double-sided memory)?

(Hopefully it won't be too long before I can contribute to the forums by answering others' technical questions!)

#2 Insomnia

Insomnia

    Sr. Expert

  • Active Member
  • PipPipPipPipPipPipPipPip
  • 183 posts

Posted 07 December 2011 - 02:21 PM

Jeez, good questions.  I kind of cringed at trying to answer these... It's been a while since I've studied the low-level stuff for RAM, but I'll see if I can help clarify a bit.

QUOTE (jeffk @ Nov 27 2011, 04:32 PM) <{POST_SNAPBACK}>
Reading around, it seems that 'banks' is used in similar but subtly different ways:

Yeah, us techs do like to re-use our lingo.

QUOTE (jeffk @ Nov 27 2011, 04:32 PM) <{POST_SNAPBACK}>
1) as a group of sockets or modules that make up one logical unit, with the logical unit referring to a match between the processor's data bus width and the RAM's bit width e.g. one 64-bit DIMM or two 32-bit DIMMs makes a full bank with a CPU with a 64 bit data bus.

Makes sense, this term of banks refers to how modules are organized on the mobo.  Most motherboards have two or three DIMMs per bank, and generally 2 banks per board.

QUOTE (jeffk @ Nov 27 2011, 04:32 PM) <{POST_SNAPBACK}>
2) but then other people refer to banks of memory addresses, or as banks of "switches" in a RAM chip (each switch being 1 transistor/capacitor set)

This actually deals with how a memory module works on the inside.  Generally speaking, in the old days most RAM modules (what you plug into a mobo) contained 8 or 9 memory chips on the board.  These are the black chips you see on the outside of the module.  Inside, these RAM chips were aligned in such a way that an equal portion of each byte of data is stored in each chip (each portion stored in a switch).  These bytes, known as words, were then stored logically in address "pages."  Each page holds pointers to the memory locations where the data for the word is held.

The banks of memory addresses they mention are these logical pages, and deal with the logical ordering of data stored in the chip.  The term "Banks of switches" is probably their way of trying to tie the idea of logical ordering to the internal structure of the chip.  Note that the RAM chips I'm describing are no longer ordered this way.  This is why we see many RAM modules with 3-6 chips instead of the old 8 or 9.

QUOTE (jeffk @ Nov 27 2011, 04:32 PM) <{POST_SNAPBACK}>
3) & then Scott Mueller says that dual-ranked memory has internal banks of memory chips called ranks.

I think the ranks he's referring to are the words.  IE, if a word was stored across multiple chips, think of the physical chip as vertically splitting data (think coloumns on a spreadsheet), while a rank would be horizontally splitting the data (think row)

QUOTE (jeffk @ Nov 27 2011, 04:32 PM) <{POST_SNAPBACK}>
(He was explaining why single-/double-sided memory has nothing to do with whether there's chip one one or both sides of a RAM stick, it's to do with memory ranks).

This could be true.  Generally I see dual-sided memory having the RAM chips on both sides of the module, but this could be due to companies re-using cheap parts rather than being required.

QUOTE (jeffk @ Nov 27 2011, 04:32 PM) <{POST_SNAPBACK}>
I think my brain's gonna explode!  biggrin.gif

Mine already has...

QUOTE (jeffk @ Nov 27 2011, 04:32 PM) <{POST_SNAPBACK}>
Any help from the more theory inclined techs here would be much appreciated. Thanks!


Also, which is faster, single- or dual-rank (often called single-/double-sided memory)?

My pleasure.  Also, neither is faster.  Single/dual refers to (as stated above) how memory is stored in a chip, this won't influence speed.  Speed is instead influenced by clockrate.  Now actual chip statistics, such as how long it takes for electrical signals to stabilize inside of the chip, will influence speed as clockrates are set to give the characteristics of the hardware time to fully react or change state.


Hopefully this is all clear; let me know if it isn't.  I could probably get more indepth into the mathematics and theory behind this stuff, but I'd honestly have to review it before I could remember it all.  It's also well out of scope of anything expected to be known by a tech.

Edited by Insomnia, 22 December 2011 - 11:28 AM.


#3 Transk53

Transk53

    Sr. Advanced Member

  • Active Member
  • PipPipPipPipPipPip
  • 63 posts
  • Gender:Male
  • Location:Hove UK
  • Interests:Computers and Technology. Gaming and Overclocking. Reading and Writing (fanfic) If only I could add keeping fit to that :D But first and foremost, A+ and N+ to get nailed.

Posted 20 December 2011 - 11:50 AM

QUOTE ("jeffk")
Also, which is faster, single- or dual-rank (often called single-/double-sided memory)?


If by this you mean Dual Channel and Single Channel, then yes Dual Channel is quicker. Being Double Data Rate even on a single stick. The trick with Dual Channel though, is to match and pair up. basically a macthed pair will process in sync and be out of step of each over.

#4 Insomnia

Insomnia

    Sr. Expert

  • Active Member
  • PipPipPipPipPipPipPipPip
  • 183 posts

Posted 22 December 2011 - 11:22 AM

Alright, I think we're getting confused here.  I believe you're referring to three different technologies and considering them all to be the same.

QUOTE (Transk53 @ Dec 20 2011, 08:50 PM) <{POST_SNAPBACK}>
If by this you mean Dual Channel and Single Channel, then yes Dual Channel is quicker. Being Double Data Rate even on a single stick.


Dual Channel/Single channel refers to the amount of memory sticks per each bank on a motherboard.  Most motherboards now are dual channel, although some are triple (X-58 boards by Intel).  This technique essentially allows a computer to access a larger stream of data within a single transfer.  If RAM stores 64-bit words, and there are two memory modules in a bank, then when accessing that bank the computer can theoretically access 128-bits of data within a single transfer.  Some say this can help speed up memory, but the gains are relatively insignificant (under most tasks)

Double Data Rate refers to a technique where data is sent on both the rising and falling edge of a clock cycle.  Double Data Rate is a technique that is used in a lot of different hardware specifications.  It's well known with RAM due to how they've been named.  Performance-wise, DDR can theoretically double the amount of data that can be transferred per cycle, in real life the increase is still high but not quite double.

QUOTE (Transk53 @ Dec 20 2011, 08:50 PM) <{POST_SNAPBACK}>
The trick with Dual Channel though, is to match and pair up. basically a macthed pair will process in sync and be out of step of each over.


Dual channel technology does work by matching the sticks of RAM, but they do not work out of step.  They work in parallel.  I think the technology you're talking about by working "out of step" is known as multi-threading.   This is a processor technology that allows a single CPU to process multiple jobs by handling parts of each job "out of step," or by piecemeal.  Instead of handling one part of a job, waiting for the required data, and then completing the job a multi-threaded CPU will start a job, start the second job while waiting for the first job's data, and then complete the first job.  

This does not apply to RAM, all RAM currently in use is synchronous (bound by the system clock cycle), with a bank working in parallel, not out of step.  There may be some form of optimization used that would cause data to be split onto multiple memory banks, but this doesn't have anything to do with Dual Channel/Triple Channel RAM.  As a side-note, yes I know there's asynchronous RAM, but I don't believe anything currently uses it.

Edited by Insomnia, 22 December 2011 - 11:45 AM.


#5 Transk53

Transk53

    Sr. Advanced Member

  • Active Member
  • PipPipPipPipPipPip
  • 63 posts
  • Gender:Male
  • Location:Hove UK
  • Interests:Computers and Technology. Gaming and Overclocking. Reading and Writing (fanfic) If only I could add keeping fit to that :D But first and foremost, A+ and N+ to get nailed.

Posted 23 December 2011 - 02:41 AM

Used the the wrong phrase here "but they do not work out of step" I did not mean to cite "out of order execution" In terms of Channels, Intel are now providing Quad on the new Sandy Bridge E processors, but it is too early to tell if a Quad Channel setup will be of benefit. Ethier way, Quad Channel will be with the LGA2011.

QUOTE ("Insomnia")
As a side-note, yes I know there's asynchronous RAM, but I don't believe anything currently uses it.


Not sure what you mean by this. AMD have deployed Asynchronous RAM for years. IE, not tied to the system clock. And Yeah, Intels could run a 1.1 Divider. Anyway obviously I have read the OP wrong.

#6 Insomnia

Insomnia

    Sr. Expert

  • Active Member
  • PipPipPipPipPipPipPipPip
  • 183 posts

Posted 23 December 2011 - 01:06 PM

QUOTE (Transk53 @ Dec 23 2011, 10:41 AM) <{POST_SNAPBACK}>
Not sure what you mean by this. AMD have deployed Asynchronous RAM for years. IE, not tied to the system clock. And Yeah, Intels could run a 1.1 Divider. Anyway obviously I have read the OP wrong.


I had meant that everything I'd written regarding ram was based on synchronous RAM, not asynchronous.

Where does AMD deploy asynchronous, out of curiosity?  Also, what's Intel 1.1 divider?

#7 Transk53

Transk53

    Sr. Advanced Member

  • Active Member
  • PipPipPipPipPipPip
  • 63 posts
  • Gender:Male
  • Location:Hove UK
  • Interests:Computers and Technology. Gaming and Overclocking. Reading and Writing (fanfic) If only I could add keeping fit to that :D But first and foremost, A+ and N+ to get nailed.

Posted 23 December 2011 - 02:04 PM

QUOTE (Insomnia @ Dec 23 2011, 01:06 PM) <{POST_SNAPBACK}>
I had meant that everything I'd written regarding ram was based on synchronous RAM, not asynchronous.

Where does AMD deploy asynchronous, out of curiosity?  Also, what's Intel 1.1 divider?


Ahh right. AMD use RAM in Asynchronous Mode and Intel both. At least pre QPI I believe. The 1.1 Ram Divider would allow the Synchronous Mode between CPU and RAM. Mainly at stock speeds though.





0 user(s) are reading this topic

members, guests, anonymous users

Due to excessive spam attacks, as of November 2012, these forums are no longer operating. We would be back with a much improved forum soon.