Welcome Guest ( Log In | Register )

Outline · [ Standard ] · Linear+

 DDR2 RAM overclocking - the confusion, Tighter timings or higher frequency

views
     
TSkmarc
post Feb 7 2008, 03:35 PM, updated 18y ago

The future is here - Cryptocurrencies!
Group Icon
Elite
14,576 posts

Joined: May 2006
From: Sarawak



This guide is to explain in simple terms regarding the issue of ram overclocking, whether a tighter timings at lower mhz is faster than a looser timings at higher mhz.

Note : If there is any incorrect facts or errors, please inform me through pm. Thx.

Let's say you have a DDR2 rams that is capable of the following:
- 1050mhz @ 5-5-5-15
- 950mhz @ 4-4-4-12


Which one would you choose? Which one is faster?

Important
Most of the references are taken from this excellent guide : The Tech Repository - by freecableguy
You can read the original guide for detailed information
I'm just making it simple so that new overclockers can understand it better
The guide deals mainly on Cas Latency (CL)
It is assumed that you know more or less about ram overclocking. If not, you can refer to my guide on DDR2 overclocking (See here : Basic guide to DDR2 ram overclocking)
Other references are indicated in the guide

------------- Part 1 : Understanding the basics -------------------------------

CAS - Column Access(or Address) Strobe*(or Select)

Latency - English word which means the delay before something is done

CAS Latency (CL) - CL is the most important memory timing in terms of memory performance.
- CL refers to the number of clock cycles it takes before a column can be addressed on the DRAM chip (in other words, it tells us how many clock cycles the memory will delay to return a requested data)

Attached Image

Source : http://www.hardwaresecrets.com/article/26/3

Look at the above picture for better understanding.
- for a CL3 timing, after the "read" command is issued, it takes 3x clock cycles before the data is sent out
- for a CL5 timing, after the "read" command is issued, it takes 5x clock cycles before the data is sent out

Obviously, the lower the CL, the faster the memory can run, as it can send out data faster!

Also note that CL is a ratio
- In terms of real time measurement, it can be expressed in nanoseconds (ns)
- E.g. a DDR2 ram has a CL4, and each clock cycle is 2ns.... so, in terms of ns, the CL of the DDR2 = 2 x 4 = 8ns

Clock cycle - depends on bus speed
Calculated by T = 1 / f
Where T = Clock cycle in ns, f = bus speed
Bus speed ----- clock cycle duration ----- CL4 rams
200 mhz ------------ 5.0 ns ---------------- 5.0 x 4 = 20.0 ns
266 mhz ------------ 3.75 ns -------------- 3.75 x 4 = 15.0 ns
333 mhz ------------ 3.0 ns ---------------- 3.0 x 4 = 12.0 ns
400 mhz ------------ 2.5 ns ---------------- 2.5 x 4 = 10.0 ns

So, obviously if your bus speed is higher, each clock cycle is shorter......

Conclusion
If you understand the above explanation, then you will understand that:
- CL4 is faster than CL5
- CL is only a ratio, and the duration is dependent on the duration of a clock cycle
- A clock cycle length is dependent on the bus speed
- CL4 running at 200mhz bus speed is slower than CL4 running at 333mhz bus speed

Study has shown that regardless of memory speed and CAS latency specification :
Value Series memory is usually binned for 10.5 - 12.0 ns of CAS latency
Mainstream memory is often about 8.0 - 10.5 ns
Performance Series memory can be found to be as low as 7.0 - 8.0 ns

------------- Part 2 : Timings (CL) vs Raw speed (mhz) -----------------------------
To help you determine which speed/timings is better, take a look at this graph

Attached Image

The above graph assumes that your aim is to have a CL of 7.5 - 8.5ns
Let's also assume that your rams are only capable of running the lowest @ 7.5ns (the lower the better!!!)

Note : Obviously, the 7.5-8.5ns range is only an example. If you have value rams, the speed binned for these rams are higher (10.5-12.0ns) as mentioned above.

Y-axis - the CL
X-axis - the RAM speed in mhz
Curve/slope lines - indicates that the CL gradually reduces at higher mhz
- In simple terms, the higher you OC your rams, the lower the CL duration (which is good!!!)

Example 1
Look at the first curve/slope, showing a ram running at CAS 4:
At 800mhz, the CL is 10.0ns
At 941mhz, the CL is 8.5ns
At 1066mhz, the CL is 7.5ns

So, in order for your rams to run at CAS 4 in between 7.5-8.5ns, you would have to OC the rams in between 941mhz to 1066mhz
- Anything lower than 941mhz, your rams would be running at longer CAS 4 (Longer than 8.5ns)
- Anything higher than 1066mhz, your rams can's run (as it will be shorter than 7.5ns)

Example 2
Look at the 2nd curve/slope, showing a ram running at CAS 5
At 1066mhz, the CL is just below 9.5ns
At 1176mhz, the CL is 8.5ns
At 1333mhz, the CL is 7.5ns

So, in order for your rams to run at CAS 5 in between 7.5-8.5ns, you would have to OC the rams in between 1176mhz to 1333mhz
- Anything lower than 1176mhz, your rams would be running at longer CAS 5 (Longer than 8.5ns)
- Anything higher than 1333mhz, your rams can's run (as it will be shorter than 7.5ns)

So, do you understand the graph?
- if you do, you will notice that the following ram settings are having similar ram performance
--- CAS 4 at 941mhz vs CAS 5 at 1176mhz vs CAS 6 at 1412mhz (all have 8.5ns CL)
--- CAS 4 at 1066mhz vs CAS 5 at 1333mhz vs CAS 6 at 1600mhz (all have 7.5ns CL)

From the graph, you can also roughly conclude that:
--- CAS 4 at 1066mhz is faster than CAS 5 at 1200mhz
--- CAS 4 at 941mhz is faster than CAS 5 at 1070mhz
--- CAS 4 at 941mhz is faster than CAS 5 at 1140mhz

Aos from the above example, you can also roughly say that the DDR2 ram performance for:
--- 1066mhz @ 4-4-4-12 is faster than 1200mhz @ 5-5-5-15
--- 941mhz @ 4-4-4-12 is faster than 1070mz @ 5-5-5-15
--- 941mhz @ 4-4-4-12 is faster than 1140mhz @ 5-5-5-15

Important : Of course, it is not that straight forward in real-life performance, as you have to take into consideration other factors such as bus speed (where faster bus speed results in faster communication between components), other ram timings, etc....

-------------- Part 3 : Determining the best performance for your RAMS ---------------------

Download the "Overclocking Guide v1.00.0001 (8.12.07) from their guide : http://www.thetechrepository.com/showthread.php?t=195
-- The download link is at the bottom of the page

Run the excel file "TTR overclocking guide"
- It has 3 sheets (look at the tabs at the lower right corner - Goal seek tools, dual-core guide and quad-core guide

You will be concentrating on sheet 1 which is the goal seek tools

Look at the "Minimize CAS Memory access latency" table.

This is where you determine the CL of your ram settings.

Example 1
I have DDR2 rams that I can OC to these levels:
960mhz @ 4-4-4-8
1090mhz @ 5-5-5-10

Which one is faster? Let's use the table and see....
- Look at the columns with the red bolded words "Primary consideration". They are drop down menus, where you can change the parameters
- Look at the column with the green bolded words "Primary goal". That would be your result.

For the 960mhz @ CL4, do this:
- Change DDR specification to : 960mhz
- Change CAS selection to : 4

Once you do that, you will notice that the value of CAS Memory Access Latency changes to : 8.3ns

For the 1090mhz @ CL5, do this:
- Change DDR specification to 1090mhz
- Change CAS selection to : 5

CAS Memory Access Latency changes to : 9.2ns


So, the results are:
960mhz @ 4-4-4-8 = CAS memory access latency - roughly 8.3ns
1090mhz @ 5-5-5-10 = CAS memory access latency - roughly 9.2ns

Conclusion : If you run your rams at 960mhz @ 4-4-4-8, the ram performance itself is faster that a setting of 1090mhz @ 5-5-5-10.

Important :
Again, remember that there are other factors that come into play when considering the speed of your rig e.g. bus speed, other memory timings, FSB strap, CPU speed, etc.....
If you really want to know which settings are faster, the general rule is to use a ram benchmarking program such as Sisoft Sandra, SuperPi, etc..... smile.gif

THAT'S IT!!! I hope you can understand this guide as I have made it as simple as I can......

Addition
A very good review/benchmark on ram speed vs timings (must see!) : http://www.tomshardware.com/reviews/ram-sp...tests,1807.html

This post has been edited by kmarc: Jun 14 2008, 11:01 PM

 

Change to:
| Lo-Fi Version
0.0148sec    0.43    6 queries    GZIP Disabled
Time is now: 26th November 2025 - 07:04 AM