EggXpert

Looking through the specs of your RAM of choice, you come upon some form of code.  Four numbers.  They have to mean something!  Sometimes they can be similar, and sometimes the lot are different. Separated by dashes, these four digits have mystified humanity for ages.  Or, well, at least the last couple decades...

RAM Latency is an often-attended and often-misunderstood spec to practically all memory.  More specifically, the four number sequence is SDRAM specific - not that the same isn't measured for RDRAM (if that's still holding out).

Once again, I'll delve into this topic as deeply I can, as a simple knowledge article on the topic.  Once you understand the general concept, you shouldn't have to pull this up over and over again to see how good the RAM you're looking at is in comparison to some other RAM you're looking at.

Timing is one of the few things on RAM you can alter.  You can, however, only alter it so far before you'd want to look into hardware with better base timing.

And a bit of warning: This topic is a bit more technically complex than the last topic.  Hope I don't walk you in circles...

* * * * * * * *

4-4-4-12

The first thing you have to understand about RAM is that all the information on it is organised into rows and columns, forming a matrix.

Here's an example matrix (in binary for simplicity):
[0,0,0,1,0,1,1,0,
0,1,0,1,1,0,1,0,
1,1,0,1,1,1,1,0,
0,1,0,0,0,1,0,1]

Your RAM stores memory, and, when asked to (or called), provides it's stored information.  The way information is called is by coordinates (like [x,y]).  So, in my simple example, if I call for the memory at (2,4), I would get a '1' in response.  (On your RAM, this '1' would be a set of information, not just a single number.)  You can also write to a coordinate, like if you want to store a variable (or, per my example, change this '1' to a '0').

Wow.  A lot of tech garble up there.  If you don't completely understand it, don't worry about reading it a second time (and don't worry - sometimes it even makes my head hurt...), as it's only in preparation for my explanation of timing.

RAM Latency is spelled out in four numbers.  An example would be 4-4-4-12 (simply pulling numbers off the top of my head).  Each one of these numbers serves a different meaning, and all are important when it comes down to what you want out of your RAM.  Each number is how many clock cycles it takes to do each procedure, which is denoted by the PCxxxx or DDR(2) xxx spec for your RAM.  I can explain this later if you'd like, but it'd probably be just a stub (or a reply to a comment).  Before comparing Latency, first compare the PC or DDR(2) numbers (to make sure the RAM is on a fair playing ground).  The larger the number, the faster the base RAM (and speed after that is determined by RAM Latency).

I will explain these four numbers here, in four sections:

1. The first number (in my example, 4) is how long (how many clock cycles) it takes to access one column of data.  So, the larger your DDR(2) number and the smaller your first number, the faster your RAM can access a column of data.  This number is also known as your CAS Latency (the 'CA' standing for 'Column Access', get it?).
   
2. The second number (a 4 in my example) is the second half of the previous location method - finding the row of data.  So your computer has found the column of data it needs - whoopdy-friggin'-do!  What next? Without a row, the cycle so far is useless!  So the second number is how long it takes to finally obtain the data requested by the computer, as well as the time it takes to read or write data into that location.
3. The third number (also a 4 in my example) is how long it takes to 'close' that coordinate so the RAM can access another one.  After a read/write, it is necessary to 'close' the memory location, or something surely catastrophic will happen (probably a memory dump, but you can't really tell unless you have RAM without this number - which does not exist - probably to prevent something catastrophic from happening!) *breathes* So, in order for the RAM to move on, it must close previously accessed memory locations, and the time it takes to do that is the third digit of our cryptic code...
4. Numero four (a 12 in my example) is always the biggest number.  Now, you might be tempted to look and go, 'Hey!  It's all the other numbers all added up!!1'.  Well, don't.  I am aware of many people that think this, but the timing could also be 5-5-5-12 as seen here.  So, what's the explanation?  In tech jargon, the fourth number is "The minimum number of clock cycles necessary to access a certain row of data in RAM between the data request (number 1) and the precharge command (number 3)."  Basically, it's how long it takes to get the data out of the 'closest' coordinate in the matrix (like column 1, row 1) after the RAM has already been 'fired up'. (like, after the cycle has been done previously).  It is a minimum number, and is as fast as your RAM will go (disregarding overclocking).

So if you have some RAM at PC5400 with timing of 4-4-4-12 and another bit o' RAM at PC5400 with timing of 5-5-5-15, (depending on your brand loyalty) you would go with the one with lower timing.  Simple as that.  But... now you know why!

* * * * * * * *

This should clear up some questions, and I hope it's as easy to understand as my other explanations (even though it is a bit of a more technical concept).  If there's anything you need cleared up, please ask!

Future coverage will probably include IDE vs. SATA in depth looks, as well as airflow techniques and other ways to make your computer a better computer, whether you're planning out a build, upgrading, or simply looking to maximise your current setup.  Let me know your ideas and how I can improve for future issues!

(If you see mistakes, or have additions, please let me know so I can change this OP and credit you. Thanks!)

Previous TL Explains: available here!

thank you, larrikin. this is a great writeup.

You should ask to have a sub forum of stickied help articles.  Thanks once again!

I'm glad you find this information so useful.  I won't ask for any special priviledges to be sent my way, but if they want to, I won't stop them.  You could ask if you like, as I'm sure some people would find this informative, but I ask no trouble to be gone through for my sake.

Oh! That's a real cool composing. We truly need more writeup like this to make the forum more informative. Thanks alot.

I'll try to have another in a week, probably. Any ideas on what to cover next?  I know TK's listed a few, I just need to decide on order. Pretty much anything you can think of. :)  Let me know!

This is kind of technical, but thats what the purpose of this is anyway :) But how about flow rate vs pressure in watercooling systems and the benifits and weaknesses of each?  I really need help on this on personally, as im about to buy some watercooling parts...

~TK

great reference! and that's why 3-4-3-9 RAMs are way out of my lead atm =.=

 

b 

shootingfreak101:

weird...

?

Edit: Ah. You must be referring to the odd timing of wee's example.  I've seen few like this - and they just keep getting 'odder'. *shrug*

I've one additional question on this...

I've read that with some mobile processors, PC2 4200 may actually outperform PC2 5300.

My question isn't just about the validity of this wild wikipedia claim, but if there is ANY truth to it (which I believe there is some), would it be wiser to purchase, at the same price, 2G of PC2 4200 RAM with cas 4, or 2G of PC2 5300 RAM with cas 5?

This will be for a laptop with a Turion 64 X2.  I only mention that because I understand the above claim is based on "somebody's" measurements with a Core 2 Duo.

 
Anyone?
 

I'd like to take this one, if I may, panel. ;)

I'm not sure how valid that claim is, but it could have to do with many different factors.  ECC and non-ECC RAM actually operate at different speeds.  ECC RAM uses 1 bit for every nine for error checking and correction.  Non-ECC RAM uses only 8 bits at a time, and thusly runs a little faster (although less reliably - most RAM is Non-ECC).  It could also have to do with bottlenecking at the processor or Northbridge (depending on what the Motherboard recommends).  You would need exact specs on the laptop (which is hard to do often, because they're usually from larger manufacturers unwilling to release such information) to determine the validity of the claim.

As for your next question, let's do a little math.  I'll simplify it a bit, so as to not get so completely technical I might lose everybody.  We just need to remember that CAS latency is the first of our four-digit number.  Usually when a manufacturer only lists a CAS latency, it means it's a solid sequence of numbers (such as 4-4-4-12 or 5-5-5-15), but it can also mean that they just got lazy and threw us the first number. :)

I probably should have included this in the article, but I was sure it would come up sooner or later.  Here goes:

There are 1000 milliseconds in one second.  You know that the speed of the RAM (in this case, 4200 and 5300) is how many clock cycles in one unit of time (say, a second).  So we'll take the numbers and divide them into 1000.  1000 ÷ 4200 = .238 and 1000 ÷ 5300 = .189.  That's basically how many millisecond it takes for one clock cycle to pass.  They're fairly close, but when you factor in latency, the lines draw even closer.  Here's the rest of my example: Now, we have .238 and .189 rounded.  When we multiply them by 4 and 5, respectively (remember?  number of clock cycles per action?), we get .952 and .943.  That's one hundredth of a millisecond difference, but, the 5300 RAM is faster.

Of course, now, when you multiply that into minutes and hours and days, etc, it becomes more and more time saved.  Then you can see that the difference is a little more understandable, a little larger, and, well, easier to see.  Just remember, this is only the first action (column access), and the following actions take more time (though the time ratio will usually be similar).  Also, remember that your motherboard is rated for a maximum speed of RAM (determined by the speed of your Northbridge), so going over the limit won't do anything, and may actually even hurt your speed a bit.

Now for the quick answer to your question - go with the 5300 RAM with a CAS latency of 5, as long as your motherboard is rated for 5300 RAM or higher. :)

Does that help?

Any questions?

That helps a ton.  Thanx, TL!

And for the record, I don't think it was too technical.  Any more "layman" and I might not have got it.

Considering the minute difference in speed, I might've been convinced to go with the lighter RAM, but since they are the same exact price...

Well, thanx again! 

Well, this 'minute' difference can lead to quite a bit of time, when you start to look at it in days and weeks or more. :)

I'll try to have something new out today, but I can't guarantee it. :)

I have a question too :]

On Newegg you see some memory sticks that have a "recommended use for high performance or gaming memory". Well, what is the differences between a memory stick with 4-4-4-12 timings that is recommended for high performance or gaming memory vs any regular stick of the same size and same timings?

The difference between high performance and regular memory is that high performance has been tested under more constant dataflow.  They are usually a more sturdy product, and often come with some form of shield or heatsink on them.  One's just as fast as the other - that's the only difference. :)

Hope this helps!

Thats good to know, I was just assuming it was something the manufactuer just added to sell more product.

Hehe.  Kind of like rebranding?  Not quite. :)

"Here's an example matrix (in binary for simplicity):
[0,0,0,1,0,1,1,0,
0,1,0,1,1,0,1,0,
1,1,0,1,1,1,1,0,
0,1,0,0,0,1,0,1]

Your RAM stores memory, and, when asked to (or called), provides it's stored information. The way information is called is by coordinates (like [x,y]). So, in my simple example, if I call for the memory at (2,4), I would get a '1' in response. (On your RAM, this '1' would be a set of information, not just a single number.) You can also write to a coordinate, like if you want to store a variable (or, per my example, change this '1' to a '0')."

Hmm according to my calculations in a binary matrix the location (2,4) doesn't exist on your post.

     0,1,2,3,4,5,6,7

0   0,0,0,1,0,1,1,0,
1   0,1,0,1,1,0,1,0,
2   1,1,0,1,1,1,1,0,
3   0,1,0,0,0,1,0,1

Heres the correct way:

     0,1,2,3,4,5,6,7,8,9,A,B

0   0,0,0,1,0,1,1,0,0,0,0,1
1   0,1,0,1,1,0,1,0,0,0,0,1
2   1,1,0,1,1,1,1,0,0,0,0,0
3   0,1,0,0,0,1,0,1,0,0,0,0
4   0,1,0,0,1,1,0,0,1,1,0,1
5   0,1,1,1,0,0,1,0,1,0,1,1
6   0,0,0,1,1,0,1,0,0,0,1,0
7   0,1,1,1,0,0,0,0,1,1,0,1
8   1,0,0,1,1,0,1,0,0,0,0,0
9   1,0,0,0,1,1,0,0,1,1,1,0
A   1,0,1,1,0,1,0,1,0,1,0,1
B   0,0,0,1,1,1,0,0,1,1,1,0

IM A SMARTASS thankyou very much.

Krazykaveman:

Hmm according to my calculations in a binary matrix the location (2,4) doesn't exist on your post.

This is a learning article to provide for the learning needs of everyone, from the computer genius who thinks he knows everything, to the layman.  Thusly, my 'home' coordinate was set to (1,1), for simplicity's sake, with regards to the layman.

My matrix was not a binary matrix for any reason other than simplicity.  I could have used arbitrary numbers, but 1 and 0 are much faster to type, and they get enough of the point across.  Each 1 and 0 in my 'sample' matrix is symbolic of a set of data, not a single binary digit.

Thank you.