Most computer users wont particularly care about performance too much when theyre buying a computer.
As long as its fast and cheap enough, thats good enough.
One of the ways to squeeze the most performance out of your box is to get high-performance RAM.
The significant flashy sales number is the clock speed of the RAM, such as DDR4-3200 or DDR5-6400.
Technically that second number isnt the clock speed.
Its the transfer rate.
This is double the clock speed because DDR RAM is Double Data Rate.
Still, the higher number sounds better on the marketing material.
That transfer rate is a measure of the bandwidth of the RAM, so higher numbers are better.
Bandwidth, however, isnt the only factor in RAM performance.
Latency is just as important, if not more so.
Contents
What Is Latency?
Latency is a measure of the delay between a process being initiated and it actually happening.
A simple example is the ping of your internet connection.
If youve ever run a speed test, youll have seen your download speed and ping.
As many gamers will know, it doesnt matter how fast your internet is.
You wont have a good experience if youve got high latency.
High-performance RAM will always advertise its speed.
It will often advertise at least one specific measure of latency.
The most common and important measure of latency is CAS Latency, sometimes shortened to CL.
These are tCL/tCAS (the CAS Latency), tRCD, tRP, & tRAS.
Data in RAM is stored in columns, and only one can be interacted with at any time.
Only one row can be open at once.
RAM can come with multiple banks.
In this case, only one row can be available per bank.
Its important to understand that timings are not absolute values.
They are actually multiples of the RAM I/O clock as they are units of clock cycles.
Again, RAM is Double the Data Rate, which is half the advertised speed.
you oughta do some math to determine the actual latency of specific timing.
Alternatively, suppose you want an easier time.
Occasionally, these will be accompanied by either a 1T or 2T on end.
For these examples, the primary timings are 16-18-18-38 and 32-39-39-102, respectively.
The time for a single clock cycle is 0.625 nanoseconds and 0.3125 nanoseconds, respectively.
CAS Latency
The first number in the primary timings is the CAS latency.
This is typically the primary timing to improve if youre trying to overclock the RAM.
CAS is short for Column Address Strobe.
It isnt technically a strobe anymore.
RAS to CAS Delay
The second entry in the primary timings is the RAS to CAS delay.
This will generally be denoted as tRCD and is a minimum value, not an exact value.
A row must first be opened to access a column to read its data.
RAS stands for Row Access Strobe.
The time to be able to read data in that scenario is tRCD + tCL.
Row Precharge Time
The third primary timing is the Row Precharge Time, generally shortened to tRP.
This value is essential when theres another pop in of page miss.
In this case, the right row is not open, but another row is.
To kick off the right row, the other row must first be closed.
The process of completing a row is called pre-charging.
This involves writing the values to the row read from when it opened.
Row Activate Time
The fourth primary timing is the Row Activate Time, generally shortened to tRAS.
It is the time needed to internally refresh the row.
This is the only primary timing that overlaps with another, specifically tRCD.
Our DDR4 example has a tRAS of 38 cycles giving a total time of 23.75 nanoseconds.
Our DDR5 example has an rRAS value of 102 cycles giving a total time of 31.875 nanoseconds.
Interestingly, DDR5 currently also uses the tRCD + (2* tCL) sum.
Interestingly, there is some evidence that its possible to boot with a tRAS lower than tRCD + tCL.
Theoretically, this shouldnt really work.
Or if the tweaks were only partially stable.
Many acronyms exist for this value, such as CR, CMD, CPC, and tCPD.
Most RAM you find will run at 2T, though some may run at 1T.
Secondary and Tertiary Timings
There are plenty of other secondary and tertiary timings that can be changed.
However, doing so is highly complex.
Even experienced memory overclockers can take a day or more to dial in stable configs.
Some are easier to adjust than others and have more significant impacts.
For example, tREFI and tRFC.
These control how often the memory cells are refreshed and how long the refresh process takes.
During the refresh process, the bank must otherwise sit idle.
Tuning these values shows a particular value when your RAM configuration has an insufficient number of banks.
Memory Controller Ratio
Recent generations of CPUs can allow you to configure the memory controller ratio.
This is typically known as Gear 1, 2, and 4.
Gear 1 gets the memory controller to run at a 1:1 ratio with the memory.
However, this results in excessive power draw above 3600MTs, affecting system stability.
This ultimately only offers any benefit from around 4400MTs and up.
Gear 1 is better, but Gear 2 can provide stability at higher speeds.
Again, this is only necessary at high speeds.
However, it is unclear exactly where this switchover would be as the hardware isnt there yet.
Conclusions
RAM timings offer incredible configurability for your system RAM.
However, they are also a deep rabbit hole if you get into full RAM overclocking.
This can offer extra performance in a nearly plug-and-play implementation.
In some cases, the XMP profile will get enabled by default.
Still, often it will be necessary to select it manually in the BIOS.
This automatically applies the higher vendor recommended speeds and tightens the timings to the vendor-tested parameters.
If you decide to configure your RAM timings, knowing what they are and what they do is essential.
