It is always nice to get additional speed from the system for free – that is why people are engaged in overclocking. However, first of all, overclocks disperse the processor and video card, since experiments on these components give the greatest speed increase. Memory is usually left for dessert or not touched at all. Some stops the fact that it is difficult to disperse the rally, others – that this process will grant a very insignificant bonus to performance. It even happens that acceleration of memory is visible in benchmarks and some applications and is absolutely not visible in games. But for those who, in any case, want to squeeze out all the juices from their system, “gambling” publishes an educational program to accelerate memory.

Multifaceted

As in the case of other components of the system, the process of dispersing RAM is to change the working parameters of the device. Shamanic dance with three main characteristics help to achieve maximum performance from RAM – frequency, tension and delays (timings).

What can be said about the frequency? The more it is – the better! In fact, its meaning shows how many useful measures can make modules of memory in a second of real time. However, there are nuances here. The fact is that for memory of the type of DDR, which is used in modern computers, there are two different frequencies – real and effective, and the second is exactly twice as high as the first. Modules manufacturers always indicate the effective frequency of their creations, while in various diagnostic utilities, as well as in the BIOS of motherboards, it is the real frequency that is often displayed.

What is the catch? The DDR name is a reduction in the phrase DDR SDRAM, which is deciphered as DOUBLE DATA Rate Synchronous Dynamic Random Access Memory, that is, synchronous dynamic memory with arbitrary access and doubled data transfer speed. Keywords here – doubled speed. Unlike simple SDRAM (DDR predecessors), the memory in question interacts with the data bus not only on the front, but also by decline of the clock signal, that is, two tire tires correspond to two tactes of memory microcircuit. Accordingly, some software developers prefer to consider tires (real frequency), while others indicate the frequency of the chips themselves (effective frequency). So if during the acceleration you suddenly find that the memory frequency is exactly two times lower than it should be, then do not be surprised, this is normal.

The working voltage of the modules has a significant impact on their stability. In accordance with the standards, for DDR2 dies, a voltage of 1.8 V is standard, and for DDR3 – 1.5 V. Slow modules, as a rule, adhere to these values, but overclocking sets almost always work with increased voltages: the accelerated chips do not have enough power, and it has to be increased. Naturally, this leads to more intense heat, but if there are radiators on memory microcircuits, then a slight increase in voltage does not create special problems. Nevertheless, it is better not to cross certain boundaries, otherwise the modules can fail. For DDR2, a voltage of 2.2 V can be considered a reasonable maximum, and for DDR3 – 1.65 V.

The third key parameter of RAM is delay (timings), and this is definitely a topic for a separate chapter.

Without rush

So, delays – or timings. Before explaining what it is, it will not hurt to get acquainted with the architecture of DDR memory.

To store the simplest unit of information (bit) in DDR chips, a cell is used, which is a combination of a transistor and capacitor. Such cells in each memory chip a huge number. They line up in lines and columns, which ultimately form arrays called banks. Since DDR chips belong to the dynamic type of memory, their contents must be periodically updated (recharged), otherwise the information recorded in them will be lost.

The so -called memory controller is engaged in interaction with RAM. Having received a command of data from the processor from the processor with a logical address, he determines in which bank/line/column the desired cell is located and what should be done with it. The problem is that the cell cannot be processed instantly – a certain time (read: the number of tactes of memory) before the desired operation is performed. Delays arising at certain stages of reading/recording bits, and are called timings.

There are a large number of timings, but only a few of them have a key effect on memory performance. Specifically-Cas Latency, Ras-to-Cas Delay, Row Precharge Time and Row Active Time. These are their order by the degree of significance, and it is in this sequence that they are located in the BIOS of the motherboard and in the descriptions of the memory modules. For example, in the technical characteristics of the dicks Kingmax DDR3 2400 MHZ Nano Gaming Ram there is a line “10-11-10-30”-and so, this is timings. The first digit shows the value of Cas Latency, the second-Ras-to-Cas Delay and so on.

To understand what certain delays are responsible for, you should figure out how to read data from cells. First, the memory chip must prepare the desired line and the column in the bank for processing. To do this, they send the appropriate team, after which the process of activation of the line takes place, which takes a certain time. The number of clocks required for “awakening” the line is just called Ras-to-Cas Delay.

Next, the controller sends the desired sequence of cells (its length depends on the type of memory and additional settings) the command for reading, but the first portion of the information does not immediately enter the data bus, but after a few beats – this delay is called Cas Latency and is considered the key for memory modules. After all the necessary data is recorded, the controller is given a command to close and recharging the line.

And where are the two other timings? The first, Row Precharge Time, comes into force immediately after the line closes. The fact is that the subsequent access to this line does not become possible immediately, but only after recharging, which takes a certain number of measures – for this interval and Row Precharge Time is responsible for this interval. Well, the Timing Row Active Time shows the period of line activity, that is, the number of tacts that have passed from the moment of its activation to the moment of receipt of the recharge command. In fact, this delay depends on the parameters of the Ras-to-Cas Delay, Cas Latency and the length of the counted line, however, its value is usually selected by a simple addition of three other timings. This is not entirely correct, but allows guaranteed to avoid problems with the stability of work with minimal performance losses.

The data of data in memory cells is made in a similar way, so we will not consider this process in detail. Also, we will not focus on additional memory settings like the length of the line and secondary timings – their influence on the overall speed of the system is too insignificant. These parameters will be interesting to overklockers going to the record, and not simple users at all.

Stuck – and order?

Many novice collectors often make the following mistake: trying to arm the systematic system to the maximum, they set the DDR3 modules with an prohibitive operating frequency (say 2400 MHz) in the motherboard and remain in the happy confidence that the memory in their computer is already working at the declared speed. However, without additional manipulations from the user, such dies will work in the same mode as their cheap brothers. This is explained by the fact that the basic memory settings are drawn from a special SPD chip (Serial Pressence Detect), with which each DDR module is mandatory. The frequencies and timings prescribed in SPD, as a rule, are far from the most possible – this is done so that the modules can start even in a very weak system. Accordingly, such a memory has to be additionally dispersed.

Fortunately, sometimes this process can be greatly facilitated. So, Intel has been promoting a special expansion for the SPD chip for several years, known as XMP (Extreme Memory Profiles). It writes in memory modules information about additional settings of the system, which can be considered maternal boards with support for this technology. If the motherboard manages to pick up the desired XMP profile (it is selected through BIOS), then it will automatically set the memory frequency stated in it, adjusting the other parameters of the system for this, automatic acceleration will occur. True, it is extremely desirable that the memory is certified for the platform on which it is installed, otherwise the profile will either not work or work, but not as it should. In addition, it is never superfluous to double -check the values ​​set by automation, since some memory manufacturers manage to prescribe in the XMP profile such settings from which the system can die suddenly. In general, this technology is very useful, but it is friends only with Intel processors.

It is worth noting that even before the appearance of the XMP company NVIDIA And Corsair promoted a similar development, known as EPP (Enhanced Performance Profiles), but she did not take root.

Juicer

How RAM works, we figured out. Now it remains to understand how to get greater performance from her – and this is not just with this. There are two different ways to accelerate memory. The first implies an increase in the frequency of modules, the second is a decrease in timings. In other words: you can either increase the number of strokes per second, or make the clocks themselves more productive. Ideally, of course, both methods should be used at the same time, but the improvement of one parameter always leads to a deterioration in another, and it is not easy to choose the optimal balance. This is not to say in advance what will be more useful than your system – high -frequency memory with weakened timings or modules that function at a lower frequency, but have minimal delays.

If you are ready to fight for every extra score in some PCMark, then we recommend trying several different frequency and timings ratios and choose the one that gives the best result specifically for your system. Otherwise, it will be more reasonable to first increase timings, then find a frequency ceiling for the memory modules used, and then try to reduce delays again – as practice shows, this approach is more often winning. At the same time, throughout the way, you should not deviate much from the basic ratio of timings: the first three delays should be approximately the same, and for the fourth it is advisable to set the value equal to the sum of these timings or just below.

When accelerating memory, it is impossible to do without the help of tests that measure the performance of the system – it is they that will allow you to evaluate how great the increase in speed due to your manipulations is and whether it is at all. It may seem paradoxical, but sometimes a decrease in timings or an increase in the frequency of RAM can adversely affect the speed of the computer – such surprises occur infrequently, but you should not brush off from them. In general, without benchmarks nowhere. Which is best to use? We advise a gentleman's set from PCMARK , EVEREST And Winrar (built -in test), but in general, a list of diagnostic utilities for memory is vast – choose what is more like. By the way, benchmarks are also useful because they allow you to check the memory for stability. And after acceleration is considered completed, it will not prevent the computer from further wiping the computer with stress tests like OCCT And S&M, In order to finally make sure of the stability of the system.

When conducting experiments, do not forget about the increase in stress, and this is not only about the modules themselves, but also about the memory controller – it is often that it prevents him from revealing the entire potential of accelerated dies. Earlier on Intel platforms, this important element of the system was located in the northern bridge of the chipset, but recently it has finally moved to the central processors, so on modern platforms the increase in the voltage on the controller has negatively affects the temperature of the CPU. Thus, sometimes for effective acceleration of memory it is necessary to additionally enhance the cooling of the processor, and not the modules themselves. Caunch: do not increase the voltage on the controller by more than a quarter, this can lead to sad consequences.

Finally, it is worth determining in advance how acceleration will be carried out. You can either use a special utility, or change the necessary parameters directly in the BIOS. We strongly recommend that the second option be adopted, since no program is able to disclose all the possibilities provided by the system fee. Accordingly, before conducting experiments it will not prevent you from carefully studying the instructions for the motherboard – this will allow you to understand what exactly is hidden under one or another paragraph in BIOS. It just so happened that each manufacturer seeks to introduce their own designations into everyday life, and even such, it would seem, generally accepted terms, as the titles of timings, can vary from payment to a board.

And one more thing: do not immediately panic if at a certain stage of overclocking the system suddenly completely refuses to start. As a rule, this only means that the motherboard cannot automatically drop the BIOS settings unacceptable for it. This disease is not found so often and it is treated with a banal removal of the battery from the board. But if this does not help, then you can already panic.

Individual approach

When it comes directly to picking in numerous menu, it becomes clear that changing timings is much easier than the memory frequency. This is in the video cards everything is elementary: I pulled the slider to the right in a special utility – received the desired increase in the frequency. With full-fledged DDR modules, everything is much more complicated.

The main problems are related to the fact that the speed of the RAM depends at once on two parameters – the reference frequency (FSB, BCLK) and the multiplier. By changing these values, we get the total RAM frequency. However, a simple increase in the first parameter will almost certainly lead to unforeseen results, because this will invariably affect the performance of other components of the system. Of course, you can not touch the supporting frequency, but to achieve impressive acceleration with the help of only modifications of the multiplier in most cases is impossible.

On different platforms, the change in the support frequency leads to different consequences. In addition, often for the sake of increasing the speed of memory, it is necessary to change the working parameters of other executive blocks of the system. In a word, each platform needs its own approach, so we will try to make out the main nuances for each case. We, of course, will not become possible to consider all possible configurations – we will focus on the desktop platforms that have appeared in the past few years. In all of them, the memory controller is located in the processor, so we can say that acceleration features depend on which piece of silicon is the heart of the system. So, the hit parade of the most relevant processors today ..

Intel Sandy Bridge

The latest processors Intel , represented by a two thousandth ruler Core i3/i5/i7 , will like overclockers-novice. Material acceleration adherents believe that with the advent of Sandy Bridge to disperse the system, it became too boring. The thing is that in these processors the supporting frequency (in Intel it is called BCLK), from which all the main executive blocks are dancing, it is practically impossible to change-it is worth rejecting it for some 6-7 MHz, and the system begins to behave inappropriately. Accordingly, the good old techniques in the case of Sandy Bridge do not work, so the only way to disperse the RAM (as, however, the processor) is to increase the corresponding factor. Fortunately, the memory controller, built into new processors, went out rather nimble, and the frequency of 2133 MHz is submitted to him without problems. Since touching BCLK is strongly not recommended, the total support frequency of memory in any case should be a multiple of 266 MHz, that is, not any DDR3 set can be made at the frequency that is declared by its manufacturer. Say, DDR3-2000 modules, meeting with new Intel processors, will work as DDR3-1866.

Note that the Sandy Bridge processor alone is not enough for the effective dispersal of RAM – you also need a suitable motherboard. The thing is that Intel artificially limited the overclocking capabilities of not only processors (the multiplier can only be increased in models with the “K” index), but also chipsets. So, younger sets of logic do not know how to disperse the memory, therefore, in the system boards based on them, even the most speedy modules will work as DDR3-1333. But the chipset Intel P67 Express , positioning as a solution for enthusiasts, supports the modes up to DDR3-2133, so the choice of the motherboard under Sandy Bridge should be approached with all its thoroughness.

Ready for battle

How to determine whether your specific modules are suitable for acceleration or not? If the dies are initially not related to the overclocking class (that is, their frequency does not exceed the values ​​recommended by the creators), then it is worth starting primarily from their manufacturer, operating voltage and cooling system.

About the manufacturer, we think, it is not worth explaining: eminent companies use proven chips, the capabilities of which, as a rule, are not fully exhausted, but from the Chinese Nonym to expect an outstanding acceleration potential. The working voltage also allows you to determine how close the microcircuits are close to the limit of their capabilities: the less volt is supplied to the chips by default, the stronger it will be possible to increase the voltage independently and the higher the frequency potential will be. Well, high -quality radiators allow you to more effectively remove heat from the chips, which allows you to squeeze a little more performance from the dies.

Intel Bloomfield

Pets of enthusiasts – processors Core i7 nine hundred series – have phenomenal computing power, but with their help it is very difficult to force the memory to work at transcendent frequencies. This is partly compensated by the fact that the Bloomfield memory controller can operate in three channel mode, inaccessible to other platforms under consideration.

When working with Core i7-9xx The capabilities of overclocking modules, as a rule, rest against the insufficient performance of the UNCORE processor unit. The latter consists of a memory controller and L3-cache, and the speed of its operation directly depends on BCLK. At the same time, there is a rule that the frequency of this block should be at least twice as high as the frequency of memory of the memory, that is, for example, for the normal functioning of the dicks in DDR3-1800 mode, UNCORE will have to be started by 3600 MHz. The problem is that this same block turned out to be large and hot. He does not like work in abnormal regime, and the voltage supplied to him must be significantly increased (but not set above 1.4 in!). As a result, even if you do not disperse the processor computing blocks, the Uncore with a frequency of 4000 MHz will warm the crystal so that not every cooler can cope. Therefore, crossing the line of 2000 MHz for memory, without using serious cooling, is extremely difficult. And since it is not very reasonable to accelerate the memory, without increasing the processor frequency, it can be stated that the average computer on the basis of Bloomfield is not needed at all-some DDR3-1600 is more than enough.

It is curious that the models of the Core i7-9xx family provide the user at the disposal of an impressive set of many memory factors-they cover the range from 6 to 16x in steps 2x. For UNCORE, the factor can completely be unscrewed to 42x. Well, since the standard BCLK frequency in Bloomfield is 133 MHz, you can get to the maximum possible frequency values ​​without even touching the clock generator. However, playing both with BCLK and the multiplier, an experienced overkloker in any case will be able to squeeze a little more bonus megagez from the dies.

Intel Lynnfield

Linechek processors Core i7-8xx And Core i5-7xx , Built on the architecture of Lynnfield, is perhaps the best choice for those who want to set a memory frequency record for memory modules. To make sure of this, it is enough to see which processors are used by the current record holders.

The secret of Lynnfield success is that for the stable operation of the UNCORE frequency, these crystals do not have to have twice as many memory frequencies. Intel decided to generally block the multiplier of the block hated by overklockers: for eight hundred -hundredth models of Core i7, it is recorded at the mark of 18x, and for seven -hundreds – at 16x. The maximum memory multipliers for these processors are 12x and 10s, respectively. Thus, UNCORE no longer acts as a bottleneck when accelerating memory, so the “set of heights” is easily and relaxed.

The processor from the Core i7-8xx line can easily squeeze the maximum from any memory set: up to 1600 MHz (133×12) can be reached without touching the BCLK, and then experiments with the support frequency are used. The seven -hundredth Core i7 capabilities are slightly more modest, but their ordinary user should have enough. Of course, with a significant increase in the BCLK, the Uncore unit will warmly warm up (its operating voltage will have to be strengthened), but by that time the modules will already work at the limit of capabilities. In general, in such cases, a powerful processor cooling system is extremely desirable.

Intel Clarkdale

Intel budget processors with built -in graphics presented by families Core i5-6xx , Core i3 And Pentium G , are badly friends with memory. Alas, in order to save in these models, the memory controller along with the graphic nucleus is put to a separate crystal, which is connected to the calculatory nuclei of the Qpi bus tire. The use of the tire does not affect the performance of the controller well, so from the speed memory in the system with Claarkdale there will be no particular sense.

Acceleration of memory working in tandem with the designated processors is carried out in the most common way: we increase the factor, twist the frequency of BCLK (by default it is 133 MHz). There are no pitfalls, except that with strong acceleration, you will have to lower the QPI factor and increase the voltage supplied to L3-Cash (notorious UNCORE). The senior Clarkdale, as a rule, can start overclocking memory at frequencies of about 2000 MHz, which is not so bad. Another thing is that the increase in the speed of the system from increasing the speed of the dicks will be very meager. As for the maximum multiplier for memory, it depends on the specific model of the processor: for Pentiums it is 8x, and in Core i5-6xx and Core i3-10x. In addition, there is still Core i5-655K , created specifically for acceleration – he supports the multiplier 16x, but only a few motherboards know about his possibilities.

AMD Phenom II/Athlon II

In recent years, every new processor architecture from Intel has been bringing some new features associated with the dispersal. WITH AMD Everything is different – the algorithm for the splitting of these crystals has long been practically not undergoing changes. It is likely that along with the release of processors Llano , Equipped with the built -in graphic https://sister-site.org/all-wins-casino/ core, this stability will end, but for now we will consider how the memory working in tandem with the current AMD solutions accelerates – Phenom II And Athlon II.

The support frequency for memory in this case is the frequency of the system tire (HT Clock in AMD terminology), which is 200 MHz by default. A change in this parameter affects the operation of the processor, memory controller (this unit is usually designated as CPU NB) and Hypertransport Link tire. For this reason, in search of the frequency ceiling of your RAM, the multipliers for the processor and HT Link should be reduced, but the memory controller, on the contrary, should not be jerking off. Its frequency should be at least three times higher than the real frequency of memory (and, accordingly, one and a half times higher than the frequency is effective), otherwise the stability of the system is not guaranteed. However, the more quickly the controller works, the more chances to squeeze out the extra megagerians from the memory modules or lower their timings. You can even slightly lift the CPU Nb voltage to achieve the best result, but you should not get very carried away.

It should be noted that on the AMD platforms, the memory accelerates worse than on Intel platforms and, as a rule, the overpass of 2000 MHz cannot be conquered by the overclockers. Thus, buying ultra -suffering ddr3 stripes for such a system does not make much sense. Keep in mind that the operating modes to DDR3-1600 MHz inclusively can be activated by a change in the multiplier, however, with further overclocking, in any case, you will have to torment the clock generator.

* * *

As you can see, it is almost always possible to change the supporting frequency with a more or less serious acceleration of memory (and if Sandy Bridge had not existed in the world, this statement would be even more categorical). Yes, sometimes serious frequencies can be achieved through only multipliers, however, the step between the frequency values ​​available for activation in this case is too large, therefore, for a more accurate finding of the frequency ceiling, it is still necessary to shaman with a clock generator. Well, this, as you know, leads to a change in the frequency of the processor.

Morality is as follows: if you are seriously overclocking the memory, then it is worth it to disperse and the processor should be dispersed. In fact, why squeeze all the juices from the dies and at the same time try to restrain the working frequency of the processor, even if a slight acceleration of the CPU gives a much greater effect than all the experiments on memory? Thus, before taking up memory acceleration, it will be good to find out what frequencies can conquer your processor. Well, after that you will have to look for a balance between the speed of the crystal and the frequency/timings of the RAM, because it is usually not possible to set the most attractive values ​​of both components at once.

Difficult? Well, no one prevents you from just slightly twist the timings or increase the memory multiplier, and then enjoy the speed that fell out of nowhere, without delving into the further dispersal of the computer. Do not want to disclose the full potential of the system – no need. Well, the gentlemen of enthusiasts we wish good luck in this difficult, but interesting business.