Highly professional users always seek the highest level of performance and speed. Of course, they also demand the same at a reasonable price. As a result, they will have more in-depth reviews of the products. Meanwhile, brands such as Patriot, Corsair, GSkill, Idita and others that offer high-performance high-end customers have more customers and fans.

There are a number of users in the computer world who are looking for more complete and detailed information about hardware. They need to identify the factors and terms of the hardware world to make better choices. Such information can also identify the difference in performance, price and cost-effectiveness of the products. In this post, Zoomit explores concepts in the PC world that are useful for better product selection.

The basics

To explore and explain the concepts of memory in the computer world, we focus on a variety of DDR4 memories and related terms. Today, DDR4 has become an industry standard in the memory sector, and in the last four or five years we have generally seen memories of this kind. Of course, most of the terms used in the present article also apply to past generations. However, today’s user generally works with DDR4 memory unless he has a very old computer. The following is an introduction to the basic terms of computer memory:

DIMM :

This is the abbreviation for Dual Inline Memory Module or “Dual Inline Memory Module”. DIMM variants today come with two 2-bit data interfaces on either side of the module. In terms of sales, there are two categories of UDIMM modules for desktops and SODIMM for laptops. However, some compact desktop motherboards also use SODIMM. This solution is used to accommodate four modules on such a motherboard, otherwise the motherboard will probably only support two modules.

SDRAM :

stands for Synchronous Dynamic Random-Access Memory or “Dynamic Random Access Memory”. The structure of these memories has rows and columns of data storage cells, like spreadsheets. Random access memory can access the data of each cell according to the command issued by the memory controller. The term “random” in the above definition means that the controller does not need to read the data of the entire row to access the data of a particular cell and go directly to the address in question. The term “dynamic” means that each cell must be regularly refreshed so that its data is not lost. In contrast, there are static memories that generally perform slower. All the available memory in the system is also synchronized by the external clock generator with the desired frequency.

Data Rate :

The data rate is the number of times the module sends and receives data in seconds. The clock signals generate a square wave. Here is a term called Double Data Rate, which means both data transfer and signal loss. Duplicating the data transfer rate means that a 1MHz wave will be able to move data at a rate of 2 times per second. So we now know that the DDR data frequency is double the clock frequency. The term MT / s is used for the above concept.

In the DDR memory of each signal, two data are transferred

DDR4 : The fourth generation of Double Data Rate memory that is nowadays a comprehensive standard among PC memory products. In each frequency generation, capacity and some other characteristics of memory have improved steadily.

IC : Perhaps the most familiar term in this series can be called integrated circuit. Integrated circuits are generally known by end-users as chips. IC in DRAM generally has an 8-bit interface, but there are 8-bit examples.

Data Rate: The more, the better

By definition, we know that a higher data transfer rate means more data transfer per unit of time. Of course, there are some limitations in memory controller support. Most of today’s professional processors are capable of working with 2MHz (DDR4-3600) DDR4 memory. Of course, some speed limits have been artificially changed and increased, with the sole purpose of developing the market. In this situation, however, a chip maker like Intel wants to charge more for a non-overclocked K-series processor on the pretext of supporting more memory speeds, which means it needs a more professional motherboard. Here are some examples of this artificial increase.

AMD’s family of Series 4 processors are capable of working with memory faster than the DDR4-3600, but the manufacturer in the main firmware has some limitations that cause the DDR4-3600 to exceed half the speed of the controller if it exceeds the defined frequency range of the DDR4-3600. And the deceleration is proportional to the other I / O units on the mainframe.

Practical tests on a sample CPU with G.Skill Trident Z RGB DDR4-3600 memory module confirm the same reduction in performance if CPU limitations are enabled.

In the first experiment,

the defined constraints caused the mentioned parameters to drop at a frequency of 1 MHz, but when the constraints were lifted, the performance level improved at a speed of 2 MHz in the memory module.

The earliest members of the Series 6 family of adopters accept the DDR4-3467 module frequency without functional drawbacks, but higher frequencies will definitely interfere with performance and induce noise (especially in cross-talk signal). In addition, the connection paths between the CPU socket and the DIMM on some motherboards are not ready for such an increase in frequency. However, if you are running on such processors with a lower motherboard than the X470, a data rate increase of more than DDR4-2933 is not recommended.

More data transfer rate means more data transfer per clock signal per unit time

The Intel LGA 1151 family of processors have memory controllers that perform well above the DDR4-3600. Intel, of course, designed non-Z chips to limit any processor (even the K series) to higher rates. Experimental experiments with the Z370 Core i3-8350K Core i3-8350K processor also showed higher limitations.

Finally, the easiest way to break the DDR4-2666 limit is to use Z Series chipsets with Core i5 processors or higher.

The Intel Memory Controller operates at a frequency of 1 or 2 MHz and occasionally reaches higher coefficients of 1 or 2.3 MHz. The lower the memory coefficients, the more stable they are and the more stable they are on older platforms.

For example, the Z270’s designs are more stable. This makes the DDR4-3467 13x 266.6 MHz more stable than the 17x 200 MHz DDR4-3400 and it performs better.

In summary, given the above information, we can conclude that the family of Series 6 CPUs support the DDR4-3600 when the motherboard is ready. On the one hand, Core i5 or Core i7 K Series processors can handle such data rates when installed on a suitable Z390 or Z370 board. For coordination and other limitations of CPU and motherboard combinations, you should refer to hardware forums and especially overclocking groups.

The delay, the less, the better

Latency is said to be the amount of time spent in memory to begin executing an operation (such as reading and writing). It is interesting to note that this factor has not changed much since decades. The old PC-100 system and the new, professional DDR4-3200 blend have both CAS delays of 1 nanosecond. Why didn’t there be any progress? To better understand the basic timing of operations, we need to look at the concept of memory cells being row and column. The following statements should therefore be described:

CAS or Column Address Strobe : The number of clock cycles that will be needed to access data in the new column when the correct row is opened.

tRCD or RAS delay to CAS: Minimum cycles the memory controller modules has to wait for a new row to open.

tRP or Row Precharge : Minimum cycles that the memory controller must wait for the current row to open.

tRAS or Row Active Time : The minimum number of cycles the memory controller must wait between opening one row and closing another.

CMD or Command Rate : The number of cycles that instruction is given to memory to ensure that it is read by memory. Generally the values ​​of 1T or 2T mean a clock cycle and 1 cycle clock.

If we assume that the correct row is opened from memory, CAS will be the time we need to access the next bit in memory. And If all rows are closed, a row must first be opened to access a cell and then the correct column activated (tRCD + tCAS). If the wrong row is open (the row that does not contain the desired data), the current row must be closed, the correct row opened, and the correct column found in the new row (tRAS + tRCD + tCAS) to access the data cell. Finally, when the CMD is increased from 1T to 2T, a clock cycle will be required for each memory instruction.

Delay definitions use the time and clock cycle expressions

In the above definitions the term “time” was first used,

but in the definition of delay types we use the concept of clock cycles. In fact, the delay time is defined by the clock cycle. Again, the question is how the old PC-100 and the modern DDR4-3200 are parity delayed? The duration of a 1MHz clock cycle is 2 nanoseconds. As a result, PC-100 CAS 1 (with a CAS delay of one clock cycle) takes about 2 nanoseconds to access data. In contrast, the DDR4-3200 operates at a 1MHz clock, and a 1MHz clock cycle takes only 2.5 nanoseconds. Finally, the DDR4-3200 CAS 16 requires 6x of data access time, again equal to 1 nanosecond delay.

Since clock cycle time is inversely proportional to the clock,

the faster the memory, the more clock cycles will be needed to achieve a standard 5 nanosecond delay. The DDR4-3600 performs this activity in 1 cycle. The DDR4-4000 has a clock delay of 4 cycles. Finally, reducing the time to less than the standard 2 nanosecond requires fewer latency cycles per frequency. As a result, the DDR4-3200 C14 breaks down to a high standard record in 0.5 nanoseconds and the DDR4-3600 C16 delivers 3.5 nanoseconds.

Increase in rank, decrease in delay

In mainframe modules processes, waiting for each write or reading process to finish before running the next process slows down the processes considerably. Interleaving is a term that allows one to start a command while another is running. Users can increase the CPU’s ability to perform this process by increasing the number of ranks per channel from one to two. To do this, two single-channel DIMMs or one double-DIMM per channel must be installed.

The following information will be used for this purpose:

• Most of the memory produced since year 1 uses eight gigabytes of memory ICs.
• Most memory ICs have an 8-bit interface.
• Eight 8-bit ICs will be needed to fill a 4-bit rank.
• The total capacity for eight eight-gigabit ICs is eight gigabytes.
• As a result of the above information, most 1GB memory kits have four ranks.

Structures consisting of four ranks perform the process of adding two ranks per channel to two-channel motherboards. If you look at the information above, the term “mostly” has been repeated many times. As a result, there are some exceptions to these topics that we will consider below:

• Demand for 4GB ICs is high right now, and companies can’t afford to waste these chips on building single-gigabyte modules. As a result, 2GB modules are built using two 8GB IC ranks.

• Consumer-specific 2GB modules use two 1GB IC ranks. As a result, two 1GB DIMMs provide a 2GB dual-channel kit with four ranks.

• Current 4GB modules use four eight-gigabit ICs each with a 2-bit interface. Four of these modules will be needed to build four ranks.

• Older 4GB ICs are not in high demand. However, shoppers who need 4GB of memory to access the four ranks use them. However, such ICs will be difficult to identify.

At the bottom edge of the memory module’s heatsink to detect the presence of eight ICs on each side. Of course, there are exceptions to this that can only be found online resources and forums.

Conclusion

Higher data rates will mean increased modules performance. Of course, the limitations of the CPU and motherboard should always be considered. Low latency also increases performance without increasing data rates. Four ranks perform better than two ranks, resulting in a 2 GB DDR4-3200 performance better than a 1 GB DDR4-3600.

Given the above information, we understand more about them when purchasing PC memory. Of course, there are suggestions from online retailers and experts as well, but in the end you should be careful that the memory you buy is no more than CPU processing and management. In the meantime, you need to be very careful when buying single-color kits, as they will ultimately reduce overall speed and efficiency.