MT/s vs MHz (Datarate vs Frequency) in RAM Modules

10
CG Director Author Jerry Jamesby Jerry James   /  Updated 

The misuse of ‘Megahertz’ by media, motherboard manufacturers, and even memory manufacturers has always bothered me.

Even we’ve been guilty of sticking to the technically incorrect unit for simplicity’s sake – just to avoid this long-winded explanation.

Here’s MSI using MHz on one of their QVL pages:

MSI's Qualified Vendors List (QVL) page using MHz.

Source: MSI

And here’s Kingston doing the same for their own memory modules:

Kingston's memoy page using MHz.

Source: Kingston

Are they wrong? Well, yes. Technically.

Let’s quickly cover why the speed of a DDR4 memory module is 3200 MT/s and NOT 3200 MHz.

We’ll also go over what MT/s (Megatransfers per second) is and why it’s the perfect unit to indicate the effective speed of any DRAM module.

Differentiating RAM Frequency and Effective Speed: Why is the Usage of Megahertz Incorrect?

Each DRAM module has a certain clock frequency that it can maintain without compromising stability. This frequency indicates the number of clock cycles per second and is measured in Hertz (Hz, MHz, GHz, etc.).

The universal use of Megahertz to denote the data transfer rate stems from early generations of memory – SDR or Single Data Rate memory. At the time, your DRAM frequency was EQUAL to your memory speed because data was transmitted only once every clock cycle.

SDR vs DDR

However, things changed with the advent of DDR (Double Data Rate) memory. Modern DDR memory modules transmit data TWICE per clock cycle. Once at the rising edge of the square wave and once at the falling edge.

So, what you often see advertised as 3600 MHz DDR4 Memory is a module operating at an 1800 MHz clock rate with an effective transfer rate of 3600 MT/s (Megatransfers/second).

What is a Megatransfer? How Much is it?

Megatransfer is a bandwidth-agnostic term usually used for the rate of signal on I/O buses – making it the perfect unit to measure memory ‘speed.’

Let’s think of your memory access like a water pipeline. Now, just saying that water is flowing at 3600 meters/second doesn’t really tell us exactly HOW MUCH water is flowing through it.

If it’s a thin pipe, you’re clearly sending much less water down it compared to a thicker one – even though both might be going at 3600 m/s. You need the width (radius) of that pipe to find out exactly how much water is flowing through it.

Similarly, Megatransfers is the equivalent to the data transfer speed. It tells you how many transfers are taking place per second but is independent of the memory bus width (pipe thickness).

So, this number can’t really tell you how much data is being sent/received per second until you also factor in memory bus width. This means something transmitting at 10 MT/s over a 1-byte wide bus will offer speeds of 10 Megabytes/s.

Here’s an example:

Consider a mainstream DDR4-3200 memory module.

Now, DDR4 has a bus width of 64 bits (1 byte is 8 bits, so 64 bits/8 = 8 bytes) for data.

So, DDR4-3200 memory transmits data at a maximum rate of 3200 MT/s*8 Bytes = 25,600 MB/s.

Here are JEDEC specifications for DDR4 memory showcasing the same.

There’s an advantage to measuring memory speeds this way. Why? Well, because you can use this unit across memory generations as it doesn’t need to take bus width into account.

Why Does Everyone Use Megahertz Then?

Since a DDR4 module with a 1600 MHz frequency does ACT like a module with 3200 (1600*2) MHz frequency, it became quite widely used as it did still indicate its effective speed.

That said, continuing this convention will only cause more confusion if/when things like QDR memory become mainstream, which would be able to transfer 4 times each clock cycle.

Isn’t Frequency Just ‘Something per Second’?

Yes, but one of the basic tenets of frequency is that it is a repeating activity per unit of time. The key word here is ‘repeating.’

A data transfer, on the other hand, isn’t a repeating activity. A clock rate is constant and will not change as long as the required voltage is supplied to the module.

Just because you’re measuring the rate of something doesn’t make it frequency.

It’s not a perfect analogy, but suppose you measure the speed of a car and it comes out to 6 m/s (meters/second). Does this make 6 Hz or even 6 kHz the ‘frequency’ of said car? No, that’s silly.

The transfer rate of memory is how fast you can expect data transfer to happen, while a clock rate is the number of clock cycles per second.

How Will CGDirector Handle Memory Speed Notations in the Future?

One of the best ways (in my opinion) to indicate memory speeds is using the following notation:

64GB (2×32) DDR4-3200 CL16

The first number is the total memory capacity, while the brackets denote the memory config (2 sticks of 32GB each).

The memory generation comes next, and the number after it is a memory kit’s MT/s speed. The final number is the kit’s CAS Latency.

Moving forward, we’ll be sticking to this notation when describing or recommending DRAM modules/kits.

Over to You

Have any other questions about Memory Speed Notations? Feel free to ask us in the comments or our forum!

CGDirector is Reader-supported. When you buy through our links, we may earn an affiliate commission.

Jerry James

Hi, I’m Jerry – a Freelance Technical Content Writer and Strategist.
I’ve been building PCs for the past 15 years, and I’m not stopping anytime soon.
Feel free to comment and ask for my inputs on your PC builds; I’ll do my best to help out!

10 Comments

Also check out our Forum for feedback from our Expert Community.

Leave a Reply

  • newbie

    First of all i’d like to thank you for the explanations! I think i finally understand the difference from MTs to MHz now. If you don’t mind, i got a question since i’m building a new setup for myself. I found this great Viper Steel DDR4 16GB (2 x 8GB) 4400MHz Kit (PVS416G440C9K) but can’t understand how could i reach its maximum speed even with a mobo QVL compatible with it since even the latest intel chipset models (13th gen) show they can run DDR4 up to 3200 MT/s only… is there something i’m not seeing here?

  • I

    Could’ve just simply stated, speed != (or depending on the language) frequency. Frequency is defined as: Number of CYCLES per unit of time. If you shine a laser at a mirror 300 million meters away, it will take roughly 2 seconds to come back to you. Looking at just one photon, the speed would be 3×10^8 (m/s). It would have a frequency of 1Hz at that distance. Using the illustration for SDR and DDR you provided, you can see that the frequency stayed the same, however the sampling rate, or “Data Rate” as it’s effectively called, doubled because now it’s being measured at both the rise and fall of the signal. The frequency of the clock rate never changed, but the “Data Rate” has doubled. THAT is why MHz is technically correct for both even though they are different… and why it’s caused so much confusion. Instead of using water flow, it would be more like measure the speed of the photon on the way to the mirror and coming back. Rather than report on the base clock rate (which is kind of an deprecated metric with DDR and even QDR) you can simply refer to the “Data Rate” as MHz too because it’s equal to the number of data signals you receive, or 2xClock Rate which is in MHz. While yes, it is technically different from the clock rate, if what you still worried about is nomenclature they can still technically use MHz too.

    • PC Pro

      Your response to the article is incorrect and much harder to dig through than the article, which article I found simply refreshing and easy to read.
      MHZ is the cyclic frequency and is not technically correct to use as you state.
      Using the MT/s value for MHZ is inaccurate and misleading.
      The correct MHZ for 3200MT/s memory is 1600MHZ. The data throughput at maximum is 3200MT/s. The motherboard memory controller decides on the effective transfer rate based on CPU, memory capabilities, bus capability and other factors, which results in a final data transfer speed that varies over time.
      The reason we call it MHZ is because of laziness, same reason we call Cable modem “Modems”, they are not modulating anything. The proper name for a cable “modem” is a Cable terminal adapter. Who’s going to re-educate the public? Nobody. We go with what we got.

      • Jerry James

        Yep, pretty spot on. It’s just a laziness to re-educate and manufacturers wanting to push higher numbers for marketing.

  • Prince

    You are wrong on one part buddy. If a manufacture mention 4000 Mhz, and if the spd rate is 2133 Mhz, it doesn’t mean it’s data rate is 4000 MT/s and it’s clock rate is 2000 Mhz. It simply means the RAM has an XMP profile/EOCP profile which will overclock the ram’s clock rate to 4000 Mhz

    • Alex Glawion

      Hey Prince,
      I feel like we’re talking about the same thing.

      DDR4-4000 RAM Modules will have 4000 MT/s and clock at 2000 Mhz XMP. The base clock rate will most likely be somewhere around 1066/1333 MHz (2133/2666 MHz DDR), but since the modules run at XMP clocks (2000Mhz x Double Data Rate = 4000 MT/s), that doesn’t really matter.

      Did you see something in the article that you’re commenting on specifically?

      Thanks!
      Alex

    • PC Pro

      If the memory manufacturer mentions 4000MHz, it is actually 4000MT/s (mega transfers per second.).
      If a memory’s clock rate is 2000MHz, on each of the clock ticks, one transfer of data has taken place on SDR (old) memory. The MT/s rate is 2000.
      If a memory’s clock rate is 2000MHz, on each of the clock ticks, two transfers of data has taken place on DDR memory. The MT/s rate is 4000.
      On NO occasions can the memory manufacturers change the clock rate from 2000MHz to a doubled 4000MHz. Not digitally possible. Sorry, Prince, you are incorrect.

      Don’t think you are really interested in reading some facts, but the following is for those who desire a bit more information on XMP.

      XMP doesn’t really *overclock* the memory. Historically, overclocking is applying settings that are not normally intended or allowed by said manufacturers.
      XMP memory is *manufacturer supported* internal values in the onboard memory SPD chip.
      XMP memory defaults to a slow SPD memory value to support all kinds of unknown/untested motherboards for compatibility and stability.
      XMP memory values are merely aggressive profiles that tighten up memory timings, modify voltage and allow the memory to operate at it’s highest *supported and intended* speed for high end motherboard/CPU combinations. This may come with unstability and memory issues, despite the memory manufacturer testing said product as much as they do.

  • bulspirit

    This was quite well explained and managed to finally understand the enigma of mhz vs mt/s, thank you!

    • PC Pro

      Yes, this article was well written, easily explained and accurate. I enjoyed it.