FAQ: Processors Overclocking
IntroductionCPU (Central Processor Unit) or simply the central processing unit is the main computing unit (brain) of the system. The word "central" in its name is not accidental: in any scheme, the processor is really the center, communicating all devices in the system with each other, as well as their connection with the software shell. The core of the central processor is computing cores. All PC work is a computation, that is, any task that a computer performs is actually nothing more than a mathematical algorithm. However, the algorithm itself is just a function that makes sense in the presence of variables. Variables are data that must be stored somewhere, so the processor also has a cache memory. Besides, To implement the function of the "central link" of the system, the processor also needs "means of communication" with other devices, called interfaces. Among them are usually memory controllers, buses for communication with hubs (bridges), a PCI-E controller.
The processor itself has probably been seen by many: a square green printed circuit board, and a metal cap on top, but the most interesting is hidden inside.
There is a silicon crystal under the lid. It's only a couple of quadrant centimeters in size, but it contains billions of transistors. Let's take a deeper look.
The figure shows the internal structure of the processor. We can see here the previously mentioned nodes, and now it's time to take a deeper look at the principle of their operation and purpose.
Despite the fact that the computational cores (Core) are usually considered as a single module, many nodes are hidden inside the core: this is a level 1 and 2 cache memory, various blocks for calculations according to different algorithms, as well as its own input / output interface.
In the center, we can see a fairly large space occupied by the level 3 cache. Unlike levels 1 and 2, level 3 memory is common for all cores, which allows them to exchange information in parallel computations, as well as accept new tasks faster after completing previous ones.
On the right and at the bottom, we can see the controllers of the main interfaces. In order to understand why they are needed, let's look at the scheme of interaction of the CPU with other basic elements of the PC.
The picture shows a block diagram of the Z77 chipset. A chipset is a set of system logic that is designed to work with a specific processor family. Includes access buses and interfaces for connecting other devices. The chipset is the backbone of any motherboard.
The image clearly shows that the centers of the circuit are the processor and the PCH (Platform Controller Hub - a connecting node for secondary PC devices, providing them with a communication channel with the processor).
Now let's look at the main processor controllers.
IMC (Integrated memory controller), or memory controller. Provides access and management of RAM. Typically, the memory controller is multi-channel, which means that it can work with several memory modules at the same time. In our example, the controller has 2 channels, which allows it to work simultaneously with 2 DDR3 modules. Simultaneous work allows you to get a multiple increase in bandwidth.
PCI-E controller supports PCI-Express interface, which is the fastest of all available in the system. This interface is used to connect a wide range of devices such as a video card, sound card, RAID controller, and others. The main characteristic of a PCI-E controller is the number of supported lanes. One PCI-E lane (or bandwidth) provides up to 16Gbit (version 3.0) bandwidth with a processor. One slot can be supplied with multiple channels, up to 16.
DMI. DMI is one example of a PCH communication interface. Different systems use different bridges and different buses, but the general meaning is the same: to provide a universal communication interface to the processor for a lot of different devices. Most of the devices in the system, such as controllers USB, SATA, Ethernet, Audio and many others, do not require extra high bandwidth, but due to their number and frequency of access to them, the processor cannot contain enough logic to communicate with everyone directly. For this purpose, there is a special node through which all these devices communicate with the processor.
Now that we have looked at the processor design, it's time to get acquainted with the principle of its operation.
Like any digital device, the processor operates at a specific frequency. Frequency is the number of operations performed per second. As it is not difficult to guess, the more operations per second the processor performs, the more efficient it is. Most modern processors operate at frequencies between 2 and 4 GHz. Since there are a lot of different devices in the system, and they all work with a certain, at the same time, different frequencies, they must be synchronized. To do this, there is a main frequency generator (usually called BCLK), which usually generates a frequency of 100 MHz, and all other frequencies, including the frequency of the processor cores, are obtained by multiplying the BCLK.
In order to better understand how the processor works, and what it is for, consider its work with a specific example. What could be closer to us than computer games? Let's take them as an example.
When we play games, the computer performs a lot of operations: sound, graphics, physics, input / output of information to peripheral devices, and much more. However, as we have already said, all these operations are nothing more than a mathematical algorithm for a computer. The game is a kind of application. First of all, you need to download it. To do this, the operating system kernel generates a request to read data from memory (hard disk). The processor sends a corresponding request to the PCH, from where it goes to the SATA controller, and then to the hard disk, which in turn reads the necessary information. After that, the information enters the RAM, which is also connected to the processor. After that, it becomes possible to run the algorithm. The calculations of the program itself begin, of what is happening in the game. Now you need to show the picture on the monitor, the video card is engaged in drawing the image and displaying it on the monitor, but it does not know anything about what it needs to draw. At the beginning, the processor takes textures from RAM and sends them to the memory of the video card. Then it launches the graphics engine, with which it starts creating the scene. After the scene is created and the position of all objects on it is known, the scene is sent for rendering to the video card. It is also necessary to play the sound: for this, the processor starts the audio engine, and starts monitoring the events that should lead to the playback of the sound. As soon as such an event occurs, the processor tells the sound card which audio file to play. And finally, the most important thing: the processor begins to receive input information from the user: these are mouse movements, pressing buttons on the keyboard, or gamepad, or a command from any other controller. Each new action of the player leads to a change in what is happening in the game, which launches new algorithms, and so it continues until you get bored.
Summing up, we can say that the processor is an extremely multifunctional device that monitors everything that happens in the system, not a single operation can do without it, and therefore it is rightfully called the central one. No device will start working until it receives a command from it, which means that its performance is the main performance parameter of the entire system.
Work on material: DartMaul
General FAQ:
Q) How to choose the right processor? What should you pay attention to?
A) The main characteristics of the processor are the number of cores and frequency. However, the choice of the processor is inextricably linked with the choice of the platform (chipset), which determines a lot of other parameters of the computer. The issue of choosing a chipset will be discussed in the corresponding section on motherboards.
Q) What is socket?
A) Socket is a special socket on the motherboard where the processor is installed. To install a processor on a motherboard, their socket versions must be the same.
Q) What is cache memory?
A) Cache memory is special memory located inside the processor. With its help, a buffer is created through which the processor loads information and exchanges it between the cores. Most modern processors have the same amount of L1 and L2 caches, and the amount of L3 cache usually depends on the number of cores. In general, the amount of cache memory does not have a significant impact on performance.
Q) Does more cores always give better performance?
A) No. A larger number of cores, just like a higher frequency, give a linear performance gain only relative to a processor from the same family. There is a conventional concept of "performance per MHz core". This term means the number of operations (usually measured in flops (floating point operations per second)) that one processor core can perform in 1 million clock cycles (1 MHz). The final processor performance is calculated using the formula “performance per MHz core * number of cores * frequency in MHzâ€. Processors from different families have different performance per MHz core.
Q) What is the correct way to compare the performance of different processors from different families / manufacturers?
A) To measure the performance of the processor, special programs are used, called benchmarks. These programs run a calculation algorithm of a fixed, known in advance, and measure the time it takes for the processor to execute the algorithm. Benchmarks can use different algorithms and different instructions. For a correct comparison, it is worth using the results of at least two tests, one of which uses a parallel algorithm, the other a sequential one. It is worth noting that sequential performance (where only one core is used) is critical for most applications, so processors with better sequential (single-threaded) benchmarks are preferred.
Q) What is sequential and parallel computing?
A) All computational algorithms are divided into sequential (single-threaded) and parallel (multi-threaded). Sequential calculations always require the previous action to be performed before the next one is performed. As an example, let's take the expression a + b = c c + d = e, in which the variables “aâ€, “b†and “d†are known, but the variables “c†and “e†are not known. To calculate such an algorithm, it is necessary to perform 2 actions, but they can only be performed sequentially, since to find the variable "e", you must first calculate the variable "c". In parallel computations, there is no rigid sequence of actions, which makes it possible to split the algorithm into parts and execute them simultaneously.
Q) What is the difference between parallel and sequential computing from a processor's point of view?
A) Sequential computations, unlike parallel ones, can use only one processor core. Therefore, the number of processor cores does not affect the performance of such calculations.
Q) In the description of the processor, the number of cores and the number of threads are indicated separately. What is the difference? What is “physical†and “logical†core?
A) The number of cores (physical cores) is the number of computational units, and the number of threads (logical cores) is the queue units. Some processors have 2 queue blocks per core. The Hyper-threading technology works on this principle. The point is that sometimes the processor is waiting for data to be loaded into it, and at this time it is idle. The presence of the second queue allows him to perform some tasks while waiting for data to load to perform others. A processor with 2 queue blocks per core is able to demonstrate parallel computing performance up to 30% higher than a similar processor with 1 block.
Q) What is TDP?
A) TDP (thermal design power) translates to heat sink requirement. In fact, this is the value of the processor power. The amount (peak) electricity it spends while working and the amount of heat it emits. It is important to consider this indicator (as well as the TDP of other devices in the system) when choosing a cooler, motherboard and power supply.
Q) What is a processor instruction set?
A) Processor instructions are mathematical operations that the processor can perceive in hardware. Most of them are connected by the memory register, and the commands to access it. For ordinary users, the set of instructions of the processor does not matter, since all processors have a sufficient set of them for most tasks.
Q) What is tech. process?
A) Tech. the process by which the processor die is made is the width of the conductive track + the width of the insulator. As technology advances, the process becomes smaller, allowing more items to be placed in smaller areas. This parameter does not matter for end users, although by it you can roughly determine which processor model is new.
Work on material: DartMaul
Overclocking the processor
Attention !!! The author of this material and the PG administration do not bear any responsibility for any loss or damage arising as a result of overclocking the processor by you based on this material! Pay attention to the risks associated with overclocking, as this can cause permanent damage to the processor. Overclocking any processor will void the warranty, so please keep this in mind before attempting to conquer frequencies!
General concepts
Overclocking is an increase in the clock frequency of a node, leading to an increase in the number of operations per unit of time.
This article describes the basic theoretical foundations of overclocking a central processing unit (CPU). Please keep in mind that each platform has its own overclocking features, which should be familiarized with before attempting to overclock.
In addition to increasing CPU performance, overclocking also increases the power consumption and heat dissipation of the processor, so before starting, you should make sure that your power supply and processor cooling are available.
First, a little theory
The CPU clock is the result of the product of the central bus frequency and the multiplier of the processor cores. The goal of overclocking is to increase the final processor frequency, which in turn requires increasing at least one of its components.
The processor multiplier is the value by which the frequency of the master oscillator (main system bus) is multiplied. The PLL (Phase locked loop) is responsible for this action.
The advantage of multiplier overclocking is that when using this method, the processor frequency changes regardless of the frequencies of other nodes. However, most processors have an artificially limited multiplier, which usually allows you to set the multiplier 1-2 values ​​higher than the standard. Significant multiplier overclocking cannot be achieved in such a situation.
FSB (Front-Side Bus) frequency or BCLK (Base Clock) is the frequency of the main oscillator / center bus, which synchronizes the frequencies of all nodes in the system. Increasing this frequency can also be used to overclock the processor, but changing it also leads to overclocking other nodes, such as RAM, the bus between the CPU and NB (QPI) or PCH (DMI), or sometimes the PCI-E bus, memory controller or cache memory. When using the main frequency to overclock the processor, it is necessary to monitor the resulting frequencies of other nodes and, if necessary, reduce their multipliers (or, if this is not possible, increase the corresponding voltages as necessary) to ensure the stability of these nodes.
Get to the point
It’s probably no secret to anyone that overclocking is associated with an increase in voltage. However, not everyone understands why it is necessary, and what side effects it carries.
The processor is an extremely complex device, to understand the principle of its operation, knowledge of physics alone at the level of the school curriculum is not enough. However, what is resistance, and how it affects the electric current passing through a conductor, is not a secret for a common person. The higher the resistance, the more the voltage drops through the conductor, and the more heat the conductor generates. It is also no secret that resistance is directly related to the temperature of the conductor.
Further, for clarity, let's consider a graph of the interdependence of these 3 critical indicators for overclocking: processor frequency, voltage and temperature:
This chart is based on the author's general experience and is not supported by specific test results. These values ​​will be unique for each specific processor instance, and therefore this graph is presented as an illustration of general trends and cannot be used to find the target values ​​for processor overclocking.
The above graph shows the frequency versus voltage. On the graph you can see 3 curved lines, each of which corresponds to a given temperature (we are talking about the ambient temperature). It is worth explaining the meaning of these curves a little: the first thing worth noting is the exponential nature. In other words, each subsequent 100 MHz "costs" more than the previous ones. This is due to the fact that the amount of energy consumed by the processor (and, consequently, its heat release) grows, and the area of ​​the dissipation of this heat (the area of ​​the processor crystal itself) remains unchanged. This leads to an increase in the core temperature, as a result of an increase in resistance, and an even greater voltage drop. In addition, the graph shows the "break" of the curves, which occurs in the region of 4.5 - 4.6 GHz. This tipping point is usually is the optimal target for daily overclocking, since this frequency can be achieved with "little blood". Further overclocking greatly increases the requirements, first of all, to the cooling system, as well as to the batteries located on the motherboard.
Now that we know frequency is good, heating is bad, voltage is a necessary evil, and performance gains are worth every cost, it's time to get started. The first question to be answered is what is the overclocking potential of our processor, and how much it makes sense to overclock it at all. The answer to this question depends on a number of indicators, most of which are revealed empirically. To begin with, you should get a couple of useful programs and find out the main indicators of the processor: its factory frequency (although this is most likely already known), factory voltage (also known as VID), and the temperature to which it heats up under load. For these purposes, it is recommended to use the following software:
CPU-Z - Application for monitoring the current frequency, voltage and other characteristics of the processor, memory, and other system nodes.
RealTemp is a handy application for monitoring the temperature of CPU cores.
Prime95 is a stress test with high load levels for CPU, cache and RAM.
A version of Prime95 is available here without support for the AVX2 instruction set.
Having launched all 3 programs, we can get the necessary data: to create a load, we run Prime95, we take the frequency and VID of the processor from CPU-Z, the temperature of the cores from RealTemp.
SpoilerTo get correct initial results, the test frequency should coincide with the factory one (in the CPU-Z program, the Core speed should match the frequency specified in the Specifications section). Some processors equipped with Turbo Boost technology or equivalent will run at a higher frequency. In order to disable it during testing, in the RealTemp program go to the Settings and check the box next to the Disable Turbo option.
As a result, we should get the following.
In the picture, we see a processor running at the factory frequency of 3.4GHz, with a voltage of 1.06v. The core temperature does not exceed 50 degrees.
Since in this article we are considering overclocking for daily use, the goal of which is to get additional performance without compromising on components, the main limitations of such overclocking will be cooling, and, of course, the capabilities of the processor itself. There are several ways to test, but in this article we will look at the fastest and easiest:
First you need to set the maximum "comfortable" voltage for the processor. It is different for each processor family, but as a rule this value is approximately equal to VID + 0.3v. You can also rely on the motherboard, which colors the value of a given voltage in different colors depending on the range. Usually the sequence is as follows: white, yellow, purple, red. In such cases, you need to find the maximum value that falls within the yellow range. During testing, the voltage must be set in manual (forced) mode. Having decided on the voltage, we set the multiplier (or a combination of the multiplier and the bus frequency). The frequency should be taken a deliberately overestimated one in order to further reduce it. For example, you can start at 4.8GHz. Usually, to operate at this frequency, a voltage of at least 1.4v is required,
After setting these 2 values, you need to boot into the OS and run a stress test along with monitoring temperature and voltage.
There can be several scenarios for the development of an event:
1) The system does not even start to boot - this means that you are far enough from the operating frequency. In such cases, you need to reset the BIOS settings and set the frequency to 200MHz lower, at the same voltage.
2) The system started loading the OS, but in the process it hung up - reduce the frequency by 100MHz.
3) The system booted, but when starting the stress test it immediately hung / crashed into a blue screen - reduce the frequency by 100MHz.
4) The stress test started giving errors - reduce the frequency by 100MHz.
5) The stress test started, but the core temperature became too high (above 85C) - Reduce the voltage by 0.05v and the frequency by 100MHz.
6) If, on the first try, you immediately booted into the system and the stress test after working for 30 seconds did not give errors, and the temperature remained normal (not higher than 70C) - congratulations, you have a very successful copy of the processor in your hands, you can try to start increasing the frequency by 100MHz per step to find the limit of the processor's capabilities.
After you have found the maximum stable frequency with a given voltage, you can begin to lower the voltage to find the optimal combination. It is worth lowering it in increments of 0.02v. At this stage of testing, it is recommended to test each setting for 5 minutes. In case of instability detection, return to the last successful combination.
The final settings should be tested within 10 minutes.
Now, having got acquainted with the general algorithm, we will consider additional features.
Voltage: often during a load, the real voltage shown by CPU-Z can drop relative to that set in the BIOS, and sometimes quite significantly. During testing, it is necessary to note the real voltage under load, it will be required in the future to establish optimal settings for daily work.
The phenomenon of voltage drop under load is known as vdroop. To combat it, there is a special option for processor batteries (VRM). The option is called differently by different motherboard manufacturers (usually called LoadLine Calibration for ASUS motherboards, and Digital Level Compensation for MSI). The setting is designed to compensate for the drop under load, which allows you to set a lower voltage value, and not to give out excessive voltage during idle time. This option usually has several levels, for daily overclocking it is recommended to set such a level that provides an increase in voltage under load by 0.02-0.03v. Unfortunately, not all motherboards can provide this level, and in such cases it is worth setting the maximum compensation level.
After changing this setting, it is necessary to lower the processor voltage so that the final real voltage on the processor during load remains the same as during stability testing.
Auto Voltage Matching: All modern processors are equipped with power saving features that reduce real clock speeds and idle voltages. If the processor voltage is set manually (forced), energy-saving technologies will not be able to change it, which also leads to excessive voltage during idle time. In order to avoid this, the option of setting the processor voltage via Offset is used. The meaning of this option is that the system itself regulates the voltage on the processor (as in Auto mode), but at the same time you change the final value by adding or subtracting a certain amount of volts from it. First, you need to set the voltage to Auto, and see what voltage the system will set, then compare in the one that was found empirically during testing,
For example, if the optimal voltage turned out to be 1.32v, and the system automatically sets 1.28v, you need to enable the Offset option, indicate the + symbol and set the value to 0.04v.
Work on material: DartMaul
Overclocking FAQ:
Q: Will overclocking hurt my processor or shorten its lifespan?
A: If the overclocking was performed correctly, and if the voltage and temperature values ​​do not go beyond the permissible limits, the overclocking has practically no effect on the processor's service life.
Q: How can I determine the optimal temperature and voltage values ​​for my processor?
A: The optimal CPU voltage actually depends on the temperature. In other words, virtually any voltage is considered normal as long as the temperature is within reasonable limits. The maximum core temperature set by manufacturers is usually in the range of 95-100C, however, in cases of overclocking, it is not recommended to create conditions in which the temperature can go beyond 80C.
Q: During the stress test, the temperature of my CPU is about 10-15 degrees higher than during real work, why is this?
A: The stress test is specially made in order to give the maximum possible load on the processor. The point of such testing is to identify limit values, including temperatures. Passing such tests is designed to ensure stable operation within acceptable values ​​at any level of system load.
Q: When running a stress test, my processor instantly heats up 50+ degrees above idle temperature. This is normal?
A: Generally, a large difference between idle temperature and stress test temperature indicates poor cooling, or too high voltage. If the voltage does not exceed 1.4v, then the temperature difference between load and idle should not exceed 35 degrees. If this happens it may be worth checking the thermal grease, or considering replacing the cooling system.
Q: Can the new cooling system increase the overclocking potential?
A: Yes. If the difference between the old and new CO is significant, it may be possible to obtain 100-200 additional MHz or reduce the voltage required for stable operation at the current frequency.
Work on material: DartMaul