RAM-What Is It? CAS Latency and Speed

[Hello man]

Hello everyone!

It's another tutorial on RAM! With my last one just having crested 11,000 reads I figured I would write another, this one slightly more detailed. Rather than focusing on the exact amount of RAM you should have, this tutorial will go over the following: RAM speed (CAS Latency & Mhz), overclocking and cooling. A mix of both personal experience and factual knowledge, I will focus on bringing everything to the table I can. If you have feedback, PM me.

There are two primary methods of measuring RAM speed. The first and most common is megahertz, usually expressed like "1600Mhz". The other way of writing this would be "PC3 12800". Megahertz is the speed at which the RAM operates. As a general rule, the faster the RAM operates, the snappier you're system will be. I run 16GB of 1600Mhz and find it 100% sufficient for streaming to twitch while playing any game.

The second method of measuring relative RAM speed is CAS Latency. CAS stands for Column Address Strobe. If you Google CAS Latency, you get a load of technical jargon with a side of understandable information. Because I would hate for you to be confused, I'll sum it up. CAS Latency is the measured delay between the moment the memory controller queries or asks the RAM for information, and the time at which it becomes available. The smaller the CAS number is, the faster the RAM can provide data to the memory controller. Unless you use FBDIMM (fully buffered memory), you're memory controller is located on the CPU die. If you know what none of this means, then this tutorial should not be you're choice reading material......yet!

As soon as I mentioned overclocking, some of you probably started drooling. Yes yes, you can overclock RAM. However, it is a complicated process and can have considerable repercussions. If you're RAM is crap, overclocking could lead to premature failure. An bad overclock can destabilize a system and make it unable to boot. DDR3 has a stock voltage of 1.5, and the JEDEC recommended max is 1.575. Supposedly a module can handle 1.975 volts without permanent damage, but it is not guaranteed to function at this level. That said, don't live in fear. Play around with RAM speeds in the BIOS, and if you have them, ask questions in the forum.

As a general rule, RAM cooling is over advertised. I personally recommend that you ignore Nike on this one and don't believe the hype. However, if you're into it, there are ways to cool RAM. The cheapest by far is a set of heat spreaders. Probably the only method I would recommend for the causal user, spreaders are quite helpful if you're case has decent airflow. If you have FBDIMM RAM, you should DEFINATELY have spreaders. The best ones have fins, like these:

If you have a preexisting water cooling loop, or are adding one, you can water cool RAM with a relatively simple water block. But, if you're loop comes out over budget, this is the first thing I would cut out. Its pretty unnecessary unless you plan on overclocking you're RAM.

The middle ground between heat spreaders and water blocks is a simple fan assembly. If used in conjunction with a set of spreaders, it should provide more than adequate cooling!

Thanks for reading, advice and or questions are always welcome!

 

[Pinhedd]

There's a fair bit of inaccurate information in here, so I'm not really comfortable leaving this up as a tutorial.

 

[Hello man]

There's a fair bit of inaccurate information in here, so I'm not really comfortable leaving this up as a tutorial.

Care to elaborate?

 

[Hello man]

Its not every day my tutorials get yanked. In fact, its never happened before. Did I mess up describing CAS in some way? There is not much technical stuff in here to screw up on.

 

[Pinhedd]

Its not every day my tutorials get yanked. In fact, its never happened before. Did I mess up describing CAS in some way? There is not much technical stuff in here to screw up on.

1. You've confused the data rate, bus clock frequency, and bus bandwidth. These can be converted between eachother easily, but they are not identical. Marketing departments may advertise a DDR3-1600 module as "1600Mhz" but that doesn't make it accurate.

2. CAS latency isn't a measure of relative speed at all. This is one of the most common mistakes made on technical websites. CAS latency is more properly known as "column read to data ready delay" and is measured in either nanoseconds (asynchronous DRAM), or IO bus cycles (synchronous DRAM). It measures the delay between the read command and the first word of the DRAM burst. Each subsequent word of the burst is transferred every half cycle until the burst is finished. The burst length for DDR3 and DDR4 is 8 words (or 4 words with burst chop). Good memory controllers can easily interleave commands during the CAS wait period, so it has a far less significant impact on performance than most people think that it does. Furthermore, the delay looks far less impressive when the entire burst is taken into consideration.

3. There are no DDR3-FBDIMMs, those are a legacy of DDR2 and was only used on one Intel platform that I know of.

4. AMD and Intel didn't integrate a DRAM controller onto their CPU dies until around 2009. Prior to this, the DRAM controller was located on the North Bridge. While these platforms are end-of-life, they are still in use.

5. There is no "stock" voltage for DDR3. Most DRAM manufacturers adhere to the JEDEC standard for DDR3 which stipulates a 1.9 volt minimum tolerance relative to ground and recommends 1.5 volts +/- 5% for maximum interoperability (SSTL_15). Intel and AMD both recommend this as well. While the DRAM integrated circuits themselves may be able to withstand 1.9 volt signal levels, the memory controller itself may not. In fact, this may exceed the absolute maximum specified by AMD or Intel and can cause damage to the memory controller.

 

[Hello man]

Its not every day my tutorials get yanked. In fact, its never happened before. Did I mess up describing CAS in some way? There is not much technical stuff in here to screw up on.

1. You've confused the data rate, bus clock frequency, and bus bandwidth. These can be converted between eachother easily, but they are not identical. Marketing departments may advertise a DDR3-1600 module as "1600Mhz" but that doesn't make it accurate.

2. CAS latency isn't a measure of relative speed at all. This is one of the most common mistakes made on technical websites. CAS latency is more properly known as "column read to data ready delay" and is measured in either nanoseconds (asynchronous DRAM), or IO bus cycles (synchronous DRAM). It measures the delay between the read command and the first word of the DRAM burst. Each subsequent word of the burst is transferred every half cycle until the burst is finished. The burst length for DDR3 and DDR4 is 8 words (or 4 words with burst chop). Good memory controllers can easily interleave commands during the CAS wait period, so it has a far less significant impact on performance than most people think that it does. Furthermore, the delay looks far less impressive when the entire burst is taken into consideration.

3. There are no DDR3-FBDIMMs, those are a legacy of DDR2 and was only used on one Intel platform that I know of.

4. AMD and Intel didn't integrate a DRAM controller onto their CPU dies until around 2009. Prior to this, the DRAM controller was located on the North Bridge. While these platforms are end-of-life, they are still in use.

5. There is no "stock" voltage for DDR3. Most DRAM manufacturers adhere to the JEDEC standard for DDR3 which stipulates a 1.9 volt minimum tolerance relative to ground and recommends 1.5 volts +/- 5% for maximum interoperability (SSTL_15). Intel and AMD both recommend this as well. While the DRAM integrated circuits themselves may be able to withstand 1.9 volt signal levels, the memory controller itself may not. In fact, this may exceed the absolute maximum specified by AMD or Intel and can cause damage to the memory controller.

Dude! I that's way more than I knew. I know about how CAS latency is really the delay between memory controller request and data in a specific column becoming available. It seemed to me that CAS latency had an effect when I was into bench marking everything and comparing with friends. Anyways, I will revise and try to keep it understandable for most people, my goal in the first place. Thanks for the help, I'm of an age where I certainly don't know everything! (14) Plus I have been pretty inactive for a while, I got into building bikes.

 

[Pinhedd]

Its not every day my tutorials get yanked. In fact, its never happened before. Did I mess up describing CAS in some way? There is not much technical stuff in here to screw up on.

1. You've confused the data rate, bus clock frequency, and bus bandwidth. These can be converted between eachother easily, but they are not identical. Marketing departments may advertise a DDR3-1600 module as "1600Mhz" but that doesn't make it accurate.

2. CAS latency isn't a measure of relative speed at all. This is one of the most common mistakes made on technical websites. CAS latency is more properly known as "column read to data ready delay" and is measured in either nanoseconds (asynchronous DRAM), or IO bus cycles (synchronous DRAM). It measures the delay between the read command and the first word of the DRAM burst. Each subsequent word of the burst is transferred every half cycle until the burst is finished. The burst length for DDR3 and DDR4 is 8 words (or 4 words with burst chop). Good memory controllers can easily interleave commands during the CAS wait period, so it has a far less significant impact on performance than most people think that it does. Furthermore, the delay looks far less impressive when the entire burst is taken into consideration.

3. There are no DDR3-FBDIMMs, those are a legacy of DDR2 and was only used on one Intel platform that I know of.

4. AMD and Intel didn't integrate a DRAM controller onto their CPU dies until around 2009. Prior to this, the DRAM controller was located on the North Bridge. While these platforms are end-of-life, they are still in use.

5. There is no "stock" voltage for DDR3. Most DRAM manufacturers adhere to the JEDEC standard for DDR3 which stipulates a 1.9 volt minimum tolerance relative to ground and recommends 1.5 volts +/- 5% for maximum interoperability (SSTL_15). Intel and AMD both recommend this as well. While the DRAM integrated circuits themselves may be able to withstand 1.9 volt signal levels, the memory controller itself may not. In fact, this may exceed the absolute maximum specified by AMD or Intel and can cause damage to the memory controller.

Dude! I that's way more than I knew. I know about how CAS latency is really the delay between memory controller request and data in a specific column becoming available. It seemed to me that CAS latency had an effect when I was into bench marking everything and comparing with friends. Anyways, I will revise and try to keep it understandable for most people, my goal in the first place. Thanks for the help, I'm of an age where I certainly don't know everything! (14) Plus I have been pretty inactive for a while, I got into building bikes.

No worries. Do some research, fix it up, and resubmit it.

 

[Choong Keong]

May be you should extend your tutorial and give some advice as to when come to buying RAM which is more important and look out for ? Mhz or CL? I noted the more expensive RAM for example Kingston DDR 4-3000Mhz comes with a CL15, and also I noticed those Premium/gaming RAM never comes wit anything less than CL7! Why?