In the last article of this series, we reviewed the basic principles and algorithms for overclocking video cards. These simple manipulations provide a significant increase in speed, but the positive effect of a fast video card can only be appreciated in 3D applications. To increase the performance of the system as a whole, you should go to the next stage of overclocking - test the central processor.

Unbreakable bonds

In a computer, all components are interconnected using the motherboard. By editing its parameters, we also change the mode of operation of installed devices. This rule fully applies to the central processor.

The final CPU frequency from Intel is equal to the product of the system bus frequency (Front Side Bus, FSB) and the processor multiplier (multiplier, CPU Ratio). Note that the traditional FSB frequency (200 MHz, 333 MHz) actually means the reference frequency of the clock generator. The effective indicator is four times higher. Therefore, in the specifications for motherboards, we see the values ​​​​of 800 MHz, 1066 MHz, 1333 MHz. In the case of AMD processors, the resulting frequency is the product of the multiplier and the frequency of the clock generator (HTT).

The multiplier shows the number of cycles that the processor performs in one clock cycle of the system bus. This is usually an integer, although it is possible to find processors with a step of 0.5. In ancient times, the multiplier was free to change, which provided overclockers with a wide field for experimentation. Today, one can only reduce its value, i.e. the only way to increase the processor frequency is overclocking on the system bus. The floating multiplier is now found only on processors from the series Intel Core 2 Extreme and AMD Athlon 64FX.

Getting ready for accomplishments

Before proceeding directly to overclocking, we traditionally ask ourselves: does this make sense? In the case of really old and weak processors, the answer is "no". Adequate performance will not be achieved, it is better to think about getting something more powerful. A cheap motherboard or low-quality power supply can become unstable and become an insurmountable obstacle to successful overclocking. The last argument against: overclocking reduces the life of the processor. However, even taking into account the wear and tear, the CPU will work for at least 5-7 years and during this time it will become obsolete.

Now let's get ready. First of all, you need to read the instructions for the motherboard. We pay attention to the section on BIOS setup - our main overclocking tool. Here is a list of parameters to look for: system bus frequency, memory frequency and its timings settings, processor voltage, memory voltage, and chipset northbridge voltage.

Unfortunately, there is no single interface for the BIOS. On the contrary, each manufacturer tries to show maximum ingenuity in this matter. Therefore, the BIOS shells of two different motherboards with an identical set of functions can differ like heaven and earth. Not only the names of the parameters and their location differ, but also the method of modification. In one case, the “Page Up” and “Page Down” buttons are used to change the value, in the other, “plus” and “minus” or “up” and “down” buttons.

The next step on the way to future achievements is the collection of information about the system and its testing in the nominal mode. You need to make sure that under full load it works stably, in addition, it will not hurt to evaluate the performance and peak value of the processor temperature.

The utility will provide detailed information about the CPU CPU-Z. You should write down the value of the processor voltage, it will still come in handy. CPU speed is measured by the program super pi. This utility calculates pi to 33.5 million decimal places and puts a lot of stress on the system. The difference in values ​​before and after overclocking is used to estimate the increase in performance. Synthetic tests are also suitable for this purpose. futuremark PCMark05, Everest Ultimate Edition other.

Programs will tell about the temperature of the processor coretemp, S&M or speedfan. The latter, by the way, allows you to control the fan speed on the CPU cooler. In addition, monitoring utilities are supplied with the motherboard. The stability of the “processor and memory” bundle is best checked by the program S&M. If errors are observed even at the nominal frequency, then overclocking is out of the question.

We advise you to find out the temperature limit for your processor. This value is indicated either on the packaging (if you have the Box version) or on the manufacturer's website. Exceeding the maximum temperature is strictly not recommended.

Finally, we remind you that many factors play a role when overclocking a processor. A clear understanding of all the actions performed is required. Lack of caution or attention is unacceptable, because. both can lead to irreversible consequences.

The educational program is over, let's start overclocking.

Meticulous digging into the BIOS is just one of the ways to overclock the processor. There are programs that can adjust the frequency of the motherboard clock generator. Such programs are often bundled with the system board. In any case, they can be replaced by universal packages like ClockGen.

When programmatically changing frequencies, one cannot count on outstanding results. The utilities are useful only to those users who feel like newbies in overclocking and want to experiment a little. For those who want maximum results, the only way out is BIOS setup.

CPU overclocking

The first step is to enter the BIOS. To do this, immediately after turning on the computer, we hold down the "Del" button and wait for the cherished blue menu to appear. Sometimes, to get into the BIOS, you need to press some other key. In this case, you should read the instructions for the motherboard.

Next, you should find and fix the frequencies of the PCI Express, PCI, AGP, SATA, etc. buses, since they are usually proportional to the speed of the FSB. This case must be stopped by setting all tires to fixed values. Otherwise, after increasing the frequency of the system bus by 15-20 percent, the system will no longer see the devices. In addition, there is a tiny chance that components from such doping will go to another world. The nominal frequencies are as follows: PCI - 33.3 MHz, AGP - 66.6 MHz, SATA and PCI Express - 100 MHz. We set the memory frequency to the minimum, otherwise it will be a limiting factor during overclocking.

The next items that we take control of are operating voltages. We set the processor to the value shown in CPU-Z. For DDR memory, the nominal voltage is 2.5 V, for DDR2 - 1.8 V. If possible, you should fix the voltage on the north bridge of the chipset (you can find a specific value in the instructions for the board or the Everest utility). Important note: change the voltage only when you are one hundred percent sure that the value is correct.

For AMD processors, it will be useful to reduce the bus frequency by about 1.5 times HyperTransport, acting as a link between the processor and the chipset. Usually it is set as a multiplier to the frequency of the system bus (clock generator). During overclocking, the HyperTransport frequency should not exceed the nominal value. Otherwise, this bus will cause the system to become unstable.

Now we find the line responsible for the frequency of the system buses, and begin to increase the parameter. Let's call the optimal step such a change at which the processor frequency increases by about 100 MHz. In other words, the FSB frequency should be increased by a value equal to "100/multiplier". After calculating the step and changing the speed of the system bus, we save the results (usually the F10 key) and go to Windows. The testing phase begins.

With a performance check, everything is simple: just run a half-hour processor test in the S&M program. If no errors are found, increase the FSB frequency by the same step and restart the test. Do not forget about the CPU temperature - if the peak value under load approaches the maximum allowable, then it is better to stop overclocking. It is advisable to leave a safety margin of 3-4 degrees.

A separate article is a check for "throttling" (throttling) - a special processor protection mechanism. The essence of the technology is that when the CPU overheats, it starts skipping cycles in order to reduce the load. As a result, the frequency remains unchanged, but the efficiency drops. You yourself understand that overclocking with “throttling” is a pointless exercise. If the protective mechanism has worked, care must be taken to reduce the temperature (reduce the frequency or change the cooling). We recommend the following programs for tracking throttling: RightMark CPU Clock Utility and Throttle Watch.

No matter how smoothly the overclocking process goes, at a certain stage the processor will still start to produce errors. If the temperature is far from the limit, we try to increase the processor voltage. Since this leads to rapid heating (the dependence of temperature on voltage is non-linear), the initial change in voltage should be minimal. If the errors have disappeared, we continue overclocking, if necessary, raising the voltage. An increase of more than 5-7 percent is highly undesirable, otherwise, with prolonged use, processor degradation is possible. Don't forget about temperature control.

Experiments with the voltage of the north bridge are also not forbidden. True, we must remember that the chipset loses to the processor in terms of cooling quality, and proceed with caution.

When the limit is reached, the CPU temperature is close to dangerous, and errors can no longer be avoided, we lower the processor frequency by 120-150 MHz. As a result, we obtain a value at which the system will be stable. We save the operating frequency FSB and do not touch it anymore.

Often overclocking is not related to practical purposes. For some people, this process has become a kind of hobby. They are ready to spend a lot of money and a lot of time for one goal - to be the very best among their own kind for a couple of days. Ratings of champions are compiled based on the results of test applications from the series 3DMark.There are special statistics servers (for each version of the program) to which you can send your achievements.

It is simply unrealistic for an ordinary user to get to the top of these ratings. After all, extreme overclocking is not only the best equipment, but also non-standard methods. Cooling components with dry ice and liquid nitrogen is considered the norm by extreme sportsmen, and voltmod (changing the configuration of power circuits) is a vital necessity. The computer is assembled for one "run", and the components wear out in a matter of hours.

The speed achieved is amazing, but it is impossible to use this power for practical purposes.

Memory overclocking

With memory overclocking, everything is somewhat more complicated, because only the frequency is not limited here. RAM has such a parameter as timings - delays between sending a memory controller command and its execution. The less delay, the better. They are denoted, as a rule, by the lines CAS Latency (tCL), RAS-to-CAS Delay (tRCD), RAS Precharge (tRP) and Precharge Delay (tRAS).

First, we leave the timings unchanged and proceed to the search for the maximum frequency. If its value is given by a number, then the increment is usually 33 MHz (in the case of a real frequency). Many motherboards, such as the latest Intel chipsets, use dividers. They show the ratio of FSB and memory frequencies (for example, 5:4). In any case, the initial increase in frequency should be minimal.

After increasing the values, we save the results and test the system in S&M (memory test). There are no errors, so we are accelerating the memory again. And so on until the failures show themselves. It will be useful to slightly increase the voltage, no more than 0.2 V. After determining the maximum frequency at which the memory works without errors, we proceed to manipulate the timings.

There are two options: either increase the timings and conquer even higher frequencies, or decrease them, thereby increasing the efficiency of the memory at the current frequency. Which option is better depends largely on the features of the system. It turns out exclusively by experience, i.e. comparing the results of the tests carried out for each of the cases. When the ideal, in your opinion, memory settings are selected, overclocking is considered complete.

During experiments with memory, it sometimes happens that the computer simply refuses to start. There is no need to panic, just reset the BIOS configuration, and the computer will come to life again. To do this, either start the system with the "Insert" key pressed, or switch a special jumper on the motherboard. The last option is to remove the battery for a few seconds. The last two steps must be carried out with the computer turned off. After that, all parameters will be reset to their nominal values, and all values ​​will have to be set manually again.

After overclocking the processor and memory, the average temperature in the system unit will inevitably increase. This can negatively affect the video card if it is running at its limit. It is possible that the frequencies of the graphics core and video memory will have to be slightly reduced.

1. What is HyperTransport Technology?
HyperTransport Technology (formerly known as LDT, Lightning Data Transport, now often referred to simply as "HT") is a peer-to-peer, high-speed, low-latency packet communications bus developed by the HyperTransport Technology consortium (led by AMD). ", which allows chips to transfer data at a maximum speed of up to 41.6 Gb / s (for the 32-bit version of version 3.0). The scalability of its architecture is able to simplify intra-system connections by replacing some of the existing buses and bridges, as well as by reducing the number of bottlenecks and delays within the system.

2. What is the purpose of HyperTransport technology?
HyperTransport can be used in the architecture of personal computers and servers as a replacement for the proprietary version of the system bus (FSB) for processor-to-chipset communication and for processor-to-processor communication in multiprocessor systems - this is a distinctive feature of all AMD processors with the K8 (Athlon64) architecture and beyond.
It can also be used in specialized networking and telecommunications equipment, providing a significantly higher data transfer rate compared to what bus technologies that existed before the advent of HyperTransport allow.
The first real use of HyperTransport was the NVIDIA nForce chipset, which used HyperTransport technology to communicate between the two chips that make up this chipset - the IGP (nForce Integrated Graphics Processor) graphics processor and the MCP (nForce Media and Communications Processor). Since then, more and more nVidia chipsets use this technology for similar purposes (and the variants for AMD processors also actually communicate with the processor).
It can also be used as a peripheral bus for connecting specialized processors that lack the bandwidth or latency of "regular" (PCI-X, PCI-E) buses. For such purposes, the HyperTransport bus has an external design, the corresponding connector is called HTX (Hyper Transport eXtension).

3. What buses and other technologies is HyperTransport compatible with?
Bridges have been created for HyperTransport on the vast majority of data transfer buses that exist in nature, including PCI-Express, AGP, PCI, PCI-X, IEEE-1394, USB 2.0, Gigabit Ethernet, as well as less popular PL-3, SPI-4, Infiniband , SPI-5, 10 Gigabit Ethernet, etc. In traditional bus architectures (such as PCI), multiple devices share a single bus, while in HyperTransport technology, each element gets its own I/O channel. Thus, the number of "bottlenecks" (bottlenecks) in the system is reduced, and its performance is increased.
However, directly at the physical layer, HyperTransport is incompatible with any of the existing buses.

  4. Is HyperTransport technology compatible with existing programs and operating systems?
Yes, HyperTransport technology is compatible with current and future operating systems because it is logically compatible with PCI as an operating system perspective. This has already been demonstrated in the production of systems based on NVIDIA nForce chipsets.

5. Is HyperTransport technology Plug & Play compatible?
Yes, HyperTransport I/O devices are designed to use standard Plug & Play methodology and are compatible with any operating system that supports the PCI standard at the boot, runtime, and driver levels.

6. At what clock speeds does HyperTransport operate?

7. What is the width of the HyperTransport I/O bus?
I/O in HyperTransport technology is designed to provide the greatest design flexibility, allowing bus widths of 2, 4, 8, 16, or 32 bits in each direction. During the initialization process, the devices automatically recognize the bus width and then function accordingly.

Questions

What is a computer bus?

A computer bus is used to transfer data between individual functional blocks of a computer and is a collection of signal lines that have certain electrical characteristics and information transfer protocols. Buses can differ in bit width, signal transmission method (serial or parallel, synchronous or asynchronous), bandwidth, number and types of supported devices, operation protocol, purpose (internal or interface).

What is QPB?

The 64-bit QPB (Quad-Pumped Bus) processor bus provides communication between Intel processors and the northbridge of the chipset. Its characteristic feature is the transfer of four blocks of data (and two addresses) per cycle. Thus, for an FSB frequency of 200 MHz, the effective data rate would be equivalent to 800 MHz (4 x 200 MHz).

What is HyperTransport?

The serial bidirectional HyperTransport (HT) bus was developed by a consortium of companies led by AMD and serves to connect AMD K8 family processors with each other, as well as with the chipset. In addition, many modern chipsets use NT to communicate between bridges, and it has found a place in high-performance network devices - routers and switches. A characteristic feature of the NT bus is its organization according to the Peer-to-Peer (point-to-point) scheme, which provides high data exchange speed with low latency, as well as wide scalability - buses are supported from 2 to 32 bits wide in each direction (each line - of two conductors), and the "width" of the directions, unlike PCI Express, does not have to be the same. For example, it is possible to use two HT lines for reception and 32 for transmission.

The "base" clock frequency of the HT bus is 200 MHz, all subsequent clock frequencies are defined as multiples of this - 400 MHz, 600 MHz, 800 MHz and 1000 MHz. The clock speeds and data transfer rates of the HyperTransport version 1.1 bus are shown in the table:

At the moment, the HyperTransport consortium has already developed the third version of the HT specification, according to which the HyperTransport 3.0 bus allows the possibility of "hot" connection and disconnection of devices; can operate at frequencies up to 2.6 GHz, which allows you to bring the data transfer rate up to 20800 Mb / s (in the case of a 32-bit bus) in each direction, being by far the fastest bus among its kind.

What is PCI?

The PCI (Peripheral Component Interconnect) bus, despite its more than solid (by computer standards) age, is still the main bus for connecting a wide variety of peripheral devices to a computer motherboard. The 32-bit PCI bus allows dynamic configuration of connected devices and operates at 33.3 MHz (133 Mbps peak bandwidth).

Servers use its extended versions PCI66 and PCI64 (32 bit/66 MHz and 64 bit/33 MHz respectively), as well as PCI-X, a 64-bit bus accelerated to 133 MHz.

Other options for the PCI bus are the recently popular AGP graphics bus and a pair of interfaces for mobile computers: an internal mini-PCI bus and PCMCIA/Card Bus (16/32-bit external device interface options that allow hot-plugging of peripherals). Despite its widespread use, the time of the PCI bus (and its derivatives) is coming to an end - they are being replaced (albeit not as fast as its developers would like) by the modern high-performance PCI-Express bus.

What is PCI-Express?

PCI-Express is a serial interface developed by the PCI-SIG organization led by Intel and designed to be used as a local bus instead of PCI. A characteristic feature of PCI-Express is its point-to-point organization, which eliminates bus arbitration and, thus, resource shuffling.

The connection between PCI-Express devices is called links (link) and consists of one (called 1x) or several (2x, 4x, 8x, 12x, 16x or 32x) bidirectional serial lines (lane). The bandwidth of the modern PCI-Express version 1.1 bus with a different number of lines is shown in the table:

However, this year the new PCI-Express 2.0 specification will become widespread, in which the bandwidth of each link has increased to 0.5 Gb / s in each direction (while maintaining compatibility with PCI-Express 1.1). In addition, PCI-Express 2.0 doubled the power supplied via the bus - 150 W against 75 W in the first version of the standard; and, like HT 3.0, it provides the potential for hot-swappable interface cards (announced but not implemented in version 1.1).

Introduction | Overclocking Basics

Of course, our readers know all about overclocking. In fact, many reviews of processors and video cards would not be complete enough without looking at the potential for overclocking. Articles like our series "Assembling a computer for a gamer" have long specialized in evaluating performance achieved after overclocking, and not in normal mode.

If you consider yourself an enthusiast, forgive us for some background information - we'll get to the technical details shortly.

What is overclocking? At its core, the term is used to describe a component that runs at higher speeds than its specifications to increase performance. You can overclock various computer components, including the processor, memory and video card. And the level of overclocking can be completely different, from a simple increase in performance for inexpensive components to an increase in performance to an outrageous level that is normally unattainable for retail products.

In this guide, we will be focusing on overclocking modern AMD processors to get the best possible performance based on your chosen cooling solution.

Choosing the right accessories

The level of overclocking success is highly dependent on the components of the system. To begin with, you need a processor with good overclocking potential, capable of operating at higher frequencies than the manufacturer specifies. AMD is currently selling several processors that have fairly good overclocking potential, with the "Black Edition" line of processors directly aimed at enthusiasts and overclockers due to the unlocked multiplier. We tested four processors from different families of the company to illustrate the process of overclocking each of them.

For overclocking the processor, it is important that other components are also selected with this task in mind. The choice of a motherboard with an overclocking-friendly BIOS is quite critical. We took a couple of Asus M3A78-T (790GX + 750SB) motherboards, which not only provide a fairly large set of BIOS features, including Advanced Clock Calibration (ACC) support, but also work great with the AMD OverDrive utility, which is important for getting the most out of Phenom processors.

Selecting the right memory is also important if you want to achieve maximum performance after overclocking. Where possible, we recommend installing high-performance DDR2 memory that is capable of clock speeds above 1066MHz on AM2+ motherboards with 45nm or 65nm Phenom processors that support DDR2-1066.

During acceleration, frequencies and voltages increase, which leads to an increase in heat dissipation. Therefore, it is better if your system will run a proprietary power supply that provides stable voltage levels and sufficient current to cope with the increased demands of an overclocked computer. A weak or outdated power supply, loaded "to the eyeballs", can spoil all the efforts of an overclocker.

Increasing frequencies, voltages and power consumption, of course, will lead to an increase in heat dissipation levels, so the cooling of the processor and case also has a significant effect on overclocking results. We didn't want to hit any overclocking or performance records with this article, so we ended up with rather modest $20-25 coolers.

This guide is intended to help those users who are less experienced with overclocking processors to enjoy the performance benefits of overclocking their Phenom II, Phenom or Athlon X2. Let's hope that our advice will help novice overclockers in this difficult but interesting business.

Terminology

A variety of terms, often denoting the same thing, can confuse or even frighten an uninitiated user. Therefore, before we go directly to the walkthrough, we will look at the most commonly used terms related to overclocking.

Clock speeds

CPU frequency(CPU speed, CPU frequency, CPU clock speed): The frequency at which the computer's central processing unit (CPU) executes instructions (for example, 3000 MHz or 3.0 GHz). It is this frequency that we plan to increase in order to get a performance boost.

HyperTransport link frequency: interface frequency between the CPU and the northbridge (for example, 1000, 1800 or 2000 MHz). Usually the frequency is equal to (but should not exceed) the northbridge frequency.

Northbridge frequency: frequency of the northbridge chip (for example, 1800 or 2000 MHz). For AM2+ processors, increasing the northbridge frequency will result in better memory controller performance and L3 frequency. The frequency must be at least as high as the HyperTransport link, but it can be increased much higher.

Memory frequency(DRAM frequency and memory speed): The frequency, measured in megahertz (MHz), at which the memory bus operates. Can be specified as either a physical frequency such as 200, 333, 400, and 533 MHz or an effective frequency such as DDR2-400, DDR2-667, DDR2-800, or DDR2-1066.

Base or reference frequency: The default is 200 MHz. As you can see from the AM2+ processors, other clocks are subtracted from the base clock using multipliers and sometimes dividers.

Frequency calculation

Before we move on to the description of frequency calculations, it should be mentioned that most of our guide covers overclocking AM2+ processors such as Phenom II, Phenom or other Athlon 7xxx models based on the K10 core. But we also wanted to cover early AM2 Athlon X2 processors based on the K8 core, such as the 4xxx, 5xxx, and 6xxx lines. Overclocking K8 processors has some differences, which we will mention a little later in our article.

Below are the basic formulas for calculating the frequencies of the AM2+ processors mentioned above.

• CPU clock speed = base frequency * CPU multiplier;
• northbridge frequency = base frequency * northbridge multiplier;
• HyperTransport link frequency = base frequency * HyperTransport multiplier;
• memory frequency = base frequency * memory multiplier.

If we want to overclock the processor (increase its clock speed), then we need to either increase the base frequency or increase the CPU multiplier. To take an example, the Phenom II X4 940 is running at a base frequency of 200 MHz and a CPU multiplier of 15x, resulting in a CPU clock speed of 3000 MHz (200 * 15 = 3000).

We can overclock this processor to 3300 MHz by increasing the multiplier to 16.5 (200 * 16.5 = 3300) or raising the base clock to 220 (220 * 15 = 3300).

But it should be remembered that the other frequencies listed above also depend on the base frequency, so raising it to 220 MHz will also increase (overclock) the frequencies of the north bridge, the HyperTransport channel, as well as the memory frequency. On the contrary, simply increasing the CPU multiplier will only increase the CPU clock speed of AM2+ processors. We'll look at simple multiplier overclocking with AMD's OverDrive utility below, and then move into the BIOS for more advanced base clock overclocking.

Depending on the manufacturer of the motherboard, the BIOS options for the processor frequency and northbridge sometimes use not just a multiplier, but the ratio of FID (Frequency ID) and DID (Divisor ID). In this case, the formulas will be as follows.

• Processor clock speed = base frequency * FID (multiplier) / DID (divider);
• northbridge frequency = base frequency * NB FID (multiplier) / NB DID (divider).

By keeping the DID at level 1, you'll get to the simple multiplier formula we discussed above, meaning you can increase CPU multipliers in 0.5 increments: 8.5, 9, 9.5, 10, etc. But if you set the DID to 2 or 4, you can increase the multiplier in smaller increments. To complicate matters, values ​​can be specified as frequencies, such as 1800 MHz, or as multipliers, such as 9, and you may have to enter hexadecimal numbers. In any case, refer to your motherboard manual or look online for hex values ​​for different CPU and Northbridge FIDs.

There are other exceptions, for example, it may not be possible to set multipliers. So, in some cases, the memory frequency is set directly in the BIOS: DDR2-400, DDR2-533, DDR2-800 or DDR2-1066 instead of choosing a memory multiplier or divider. In addition, the frequencies of the north bridge and the HyperTransport link can also be set directly, and not through a multiplier. In general, we do not recommend worrying too much about such differences, but we recommend that you return to this part of the article if the need arises.

Test hardware and BIOS settings

Useful utilities.

• : overclocking utility;
• CPU-Z: system information utility;
• Prime95: stability test;
• Memtest86: memory test (bootable CD).
• hardware monitoring: Hardware Monitor, Core Temp, Asus Probe II, other utilities included with the motherboard.
• performance testing: W Prime, Super Pi Mod, Cinebench, 3DMark 2006 CPU test, 3DMark Vantage CPU test

• Disable Cool "n" Quiet (disable Cool "n" Quiet);
• Disable C1E (disable C1E);
• Disable Spread Spectrum (disable Spread Spectrum);
• Disable Smart CPU Fan Control (disable Smart CPU Fan Control);
• manually adjust Memory Timings (memory delays);
• Windows Power Plan: High Performance.

A warning.

Remember that you are exceeding the manufacturer's specifications. Overclocking is done at your own risk. Most hardware manufacturers, including AMD, do not offer a warranty for damage caused by overclocking, even if you use AMD's utility. the site or the author is not responsible for damage that may occur during overclocking.

Introduction to AMD OverDrive

AMD OverDrive is a powerful all-in-one overclocking, monitoring and testing utility for AMD 700 series motherboards. Many overclockers don't like to use a software utility under the operating system, so they prefer to change the values ​​directly in the BIOS. I also usually avoid the utilities that come with motherboards. But after testing the latest versions of the AMD OverDrive utility on our systems, it became clear that the utility is quite valuable.

We'll start by taking a look at the AMD OverDrive utility menu, highlighting interesting features as well as unlocking the advanced features we'll need. After launching the OverDrive utility, you are greeted with a warning message that clearly states that you are using the utility at your own peril and risk.

When you agree, pressing the "OK" key will take you to the "Basic System Information" tab, which displays information about the CPU and memory.

The "Status Monitor" tab is very useful during overclocking because it allows you to monitor the processor clock speed, multiplier, voltage, temperature and load level.

To unlock the advanced frequency setting, go to the "Preference/Settings" tab and select "Advanced Mode".

The "Memory" tab displays a lot of information about memory and allows you to adjust delays.

The utility also contains tests that load the system to check the stability of the system.

Now that you are familiar with the AMD OverDrive utility and have switched it to Advanced mode, let's move on to overclocking.

Overclocking through the multiplier

With the 790GX motherboard and the Black Edition processors we used, overclocking with the AMD OverDrive utility is pretty easy. If your processor does not belong to the Black Edition line, then you will not be able to raise the multiplier.

Let's take a look at the normal operation of our Phenom II X4 940 processor. The base frequency of the motherboard varies from 200.5 to 200.6 MHz in our system, which gives a core frequency between 3007 and 3008 MHz.

It is useful to run some performance tests at the stock clock speed to compare the results of the overclocked system with them later (you can use the tests and utilities we suggested above). Benchmarks allow you to evaluate performance gains and losses after changing settings.

To overclock the Black Edition processor, check the "Select All Cores" checkbox on the "Clock/Voltage" tab, then start increasing the CPU multiplier in small steps. By the way, if you do not check the box, then you can overclock the processor cores separately. As you overclock, don't forget to look at temperatures and constantly run stability tests. In addition, we recommend that you make notes regarding each change, where you will describe the results.

After stress testing Prime 95 for 15 minutes without a single error, we decided to raise the multiplier further. Accordingly, the next multiplier of 16.5 gave a frequency of 3300 MHz. And at this core frequency, our Phenom II passed through the stability tests without any problems.

A multiplier of 17 gives a clock speed of 3400 MHz, and again stability tests were performed without a single error.

At 3.5 GHz (17.5*200) we successfully passed a one-hour stability test under AOD, but after about eight minutes in the "heavier" Prime95 application, we got a blue screen and the system rebooted. We were able to run all benchmark tests on these settings without crashing, but we still wanted our system to get through the 30-60 minute Prime95 test without crashing. Therefore, the maximum overclocking level of our processor at a nominal voltage of 1.35 V is between 3.4 and 3.5 GHz. If you do not want to raise the tension, then you can stop there. Or you can try to find the maximum stable CPU frequency at a given voltage by increasing the base frequency in one megahertz steps, which for a multiplier of 17 will give 17 MHz in each step.

If you are not averse to raising the voltage, then it is better to do this in small steps of 0.025-0.05 V, while you need to monitor the temperatures. We kept the CPU temperatures low and we started raising the CPU voltage little by little, with a slight rise to 1.375V causing the Prime95 benchmarks to run at 3.5GHz quite steadily.

It took 1.400V to run stable at a multiplier of 18 at 3.6GHz. It took 1.4875V to be stable at 3.7GHz, which is more than the default AOD allows. Not every system will be able to provide sufficient cooling at this voltage. To increase the default AOD limit, edit the AOD .xml settings file in Notepad to increase the limit to 1.55V.

We had to bump the voltage up to 1,500V to get the system stable in the 3.8GHz 18 multiplier tests, but even bumping it up to 1.55V didn't make the Prime95 stress test stable. The core temperature during the Prime95 tests was somewhere in the 55 degrees Celsius region, which means we hardly needed better cooling.

We rolled back to a 3.7 GHz overclock, with the Prime95 test successfully running for an hour, which means the stability of the system was checked. Then we began to increase the base frequency in 1 MHz steps, while the maximum overclocking level was 3765 MHz (203 * 18.5).

It is important to remember that the frequencies that can be obtained through overclocking, as well as the voltage values ​​\u200b\u200bfor this, change from one processor sample to another, so in your case everything may be different. It is important to increase the frequencies and voltages in small increments while performing stability tests and monitoring the temperature throughout the process. With these CPU models, increasing the voltage does not always help, and processors can even become unstable if the voltage is increased too much. Sometimes for better overclocking it is enough just to strengthen the cooling system. For optimal results, we recommend keeping the CPU core temperature below 50 degrees Celsius under load.

Although we were unable to increase the processor frequency above 3765 MHz, there are still ways to further improve system performance. Raising the frequency of the northbridge, for example, can have a significant impact on application performance, as it increases the speed of the memory controller and L3 cache. The northbridge multiplier cannot be changed from the AOD utility, but it can be done in the BIOS.

The only way to increase the northbridge clock speed under AOD without rebooting is to experiment with a CPU clock speed with a low multiplier and a high base frequency. However, this will increase both HyperTransport speed and memory frequency. We'll take a closer look at this issue in our guide, but for now, let me show you the overclocking results of three other Black Edition processors.

The other two AM2+ processors overclock exactly like the Phenom II, except for one more step - enabling Advanced Clock Calibration (ACC). The ACC feature is only available on AMD SB750 southbridge motherboards, such as our ASUS 790GX model. ACC can be enabled in both AOD and BIOS, but both require a reboot.

For 45nm Phenom II processors, it's best to disable ACC, as AMD states that this feature is already present in the Phenom II die. But with 65nm K10 Phenom and Athlon processors, it is better to set ACC to Auto, +2% or +4%, which can increase the maximum achievable processor frequency.

The above screenshots show our Phenom X4 9950 overclocked at stock 2.6GHz with 13x multiplier and 1.25V CPU voltage. used for overclocking. The multiplier was increased to 15x, which gave a 400-MHz overclock at stock voltage. The voltage was increased to 1.45 V, then we tried the ACC setting in Auto, +2%, and +4%, but Prime95 could only work for 12-15 minutes. Interestingly, with ACC in Auto mode, a multiplier of 16.5x and a voltage of 1.425V, we were able to increase the base frequency to 208 MHz, which gave a higher stable overclock.

Our Athlon X2 7750 runs at stock 2700 MHz and 1.325 V. Without a voltage boost, we were able to increase the multiplier to 16x, resulting in a stable 3200 MHz. The system was also stable at 3300 MHz when we increased the voltage slightly to 1.35 V. With ACC disabled, we increased the CPU voltage to 1.45 V in 0.025 V steps, but the system was not able to work consistently with a 17x multiplier. She "flew" even before stress testing. Setting ACC for all cores to +2% allowed us to achieve an hour of stable operation of Prime95 at 1.425 V. The processor did not respond very well to voltage rises above 1.425 V, so we were able to get a maximum stable frequency of 3417 MHz.

The benefits of enabling ACC, as well as overclocking results in general, vary significantly from one processor to another. However, it's still nice to get such an option at your disposal, and you can spend time fine-tuning the overclocking of each core. We didn't get a huge boost in overclocking from enabling ACC on both processors, but we still recommend checking out 790GX review, where we took a closer look at ACC, and there this feature more seriously affected the overclocking potential of the Phenom X4 9850.

K8 processor overclocking

When overclocking K8 processors, like our Athlon 64 X2 5400+, there are some differences. Let's start with the fact that the ACC option cannot be used with K8 processors, so it is not available in the BIOS. Secondly, there is no northbridge speed control here, so there is nothing to worry about, there are no corresponding items in AOD and CPU-Z.

The third and biggest difference is due to the fact that the overclocking of the Black Edition line through a multiplier is associated with a change in the memory frequency. Unlike the K10 chips, where it is set through the base frequency and multiplier, in this case the memory frequency depends on the CPU frequency. This means that as the multiplier increases, we will change the frequency of the memory.

Officially, processors support frequencies up to DDR2-800, so the CPU frequency will be divided so that the memory frequency is less than or equal to 400 MHz (DDR2-800). This means that chips with even multipliers can run DDR2-800 memory, while chips with odd or half multipliers will run slower than 400 MHz.

Our X2 5400+ uses a 14x multiplier, which results in a processor speed of 2800 MHz. The memory in the BIOS is set to DDR2-800, while the memory frequency will be taken from the CPU frequency by dividing by 7 (half the CPU multiplier), so it will run at 400 MHz (DDR2-800). Increasing the CPU multiplier to 14.5x will give a clock speed of 2900 MHz, and since the memory cannot be clocked above 400 MHz, the memory divisor will be increased by 8 (the next integer), which will give a memory clock speed of only 363 MHz. Increasing CPU multipliers further by half an integer value will continue the trend, and 8 will remain the memory divisor for 15x, 15.5x, and 16x CPU multipliers. Of course, 16x is an even multiplier, so with it the memory will again operate at the full frequency of 400 MHz.

Note that the memory frequency can still be increased by increasing the base frequency.

With all that said, you can overclock your K8 processor using the same steps we tried above. It is important to note that the frequency of the HyperTransport link is lower for K8 processors, so don't expect stability if you seriously overclock the HyperTransport link.

BIOS Options

Our Asus M3A78-T motherboard has been flashed with the latest BIOS to support the new CPUs and also provide the best chance of overclocking success.

First you need to enter the BIOS of the motherboard (usually done by pressing the "Delete" key during the POST boot screen). Check your motherboard manual for how you can clear the CMOS (usually with a jumper) if the system fails the POST boot test. Remember that if this happens, then all previously made changes, such as time / date, turning off the graphics core, boot order, etc. will be lost. If you're new to BIOS setup, pay close attention to the changes you'll make and write down the initial settings if you can't remember them later.

Simply navigating through the BIOS menus is perfectly safe, so if you're new to overclocking, don't be afraid. But make sure that you exit the BIOS without saving the changes you have made if you think that you might accidentally mess something up. This is usually done with the "Esc" key or the corresponding menu option.

Now we have access to the necessary multipliers that can be changed. Please note that in the BIOS the CPU multiplier is changed in increments of 0.5, and the northbridge multiplier in increments of 1. And the HT channel frequency is specified directly, and not through a multiplier. These options vary significantly between different motherboards, for some models they can be set through FID and DID, which we mentioned above.

In the "DRAM Timing Configuration" item, you can set the memory frequency, whether it is DDR2-400, DDR2-533, DDR2-667, DDR2-800 or DDR2-1066, as shown in the photo. In this BIOS version, you do not need to set the memory multiplier/divider. In the "DRAM Timing Mode" item, you can set delays, both automatically and manually. Reducing latency can improve performance. However, if you do not have completely stable memory latency values ​​at hand at different frequencies, then during overclocking it is very reasonable to increase the latency of CL, tRDC, tRP, tRAS, tRC and CR. Also, you can get higher memory frequencies if you increase the tRFC delays to very high values ​​like 127.5 or 135.

Later, all "relaxed" delays can be returned back to squeeze out more performance. The process of reducing one latency per system startup is time-consuming, but well worth the effort to get maximum performance while maintaining stability. When your memory runs outside of specifications, run a stability test with utilities such as the Memtest86 boot CD, as memory instability can lead to data corruption, which is undesirable. With all that said, it's safe to let the motherboard adjust latencies on its own (usually set to quite "relaxed" latencies) and focus on overclocking the CPU.

Advanced overclocking

In this case, the adjective "advanced" is not very appropriate, because, unlike the methods discussed above, we will present here overclocking through the BIOS by increasing the base frequency. The success of such overclocking depends on how well the components of your system can overclock, and to find the capabilities of each of them, we will iterate over them one by one. In principle, no one forces you to follow all the steps given, but finding the maximum for each component can result in higher overclocking, since you will understand why you hit one or another limit.

As we said above, some overclockers prefer direct BIOS overclocking, while others use AOD to save time for testing, since they do not need to reboot every time. The settings can then be manually entered into the BIOS and try to improve them even more. In principle, you can choose any method, since each has its own advantages and disadvantages.

Again, it would be nice to disable the Cool "n" Quiet and C1E power saving options, Spread Spectrum and automatic fan control systems in the BIOS, which reduce its rotation speed. We also disabled the "CPU Tweak" and "Virtualization" options for some of our tests, but did not find a noticeable effect on any of the processors. You can later enable these features if required, and you can check if they affect system performance or your overclocking stability.

Finding the maximum base clock

Now we'll move on to the technique that owners of non-Black Edition processors will have to follow to overclock them (they can't increase the multiplier). Our first step is to find the maximum base frequency (bus frequency) that the processor and motherboard can operate at. You will quickly notice all the confusion in naming the various frequencies and multipliers, which we already mentioned above. For example, the reference clock in AOD is called "Bus Speed" in CPU-Z and "FSB/FSB Frequency" in this BIOS.

If you plan to overclock only through the BIOS, then you should lower the CPU multiplier, northbridge multiplier, HyperTransport multiplier, and memory multiplier. In our BIOS, lowering the northbridge multiplier automatically reduces the available HyperTransport link frequencies to or below the resulting northbridge frequency. You can leave the CPU multiplier at default and then lower it in AOD, which makes it possible to further increase the CPU frequency without rebooting.

For our Phenom X4 9950 processor, we chose an 8x multiplier in the AOD utility, since even a 300 MHz base frequency at this multiplier will be lower than the stock CPU frequency. We then raised the base frequency from 200 MHz to 220 MHz, and then increased it in 10 MHz steps up to 260 MHz. We then moved to a 5 MHz step and increased the frequency to a maximum of 290 MHz. In principle, it is hardly worth increasing this frequency to the limit of stability, so we could easily stop at 275 MHz, since it is unlikely that the northbridge will be able to operate at such a high frequency. Since we were overclocking the base frequency in AOD, we ran AOD stability tests for a few minutes to make sure the system was stable. If we were doing the same in the BIOS, then simply being able to boot under Windows would probably be a good enough test, and then we would run final stability tests at a high base frequency to finally make sure.

Finding the maximum CPU frequency

Since we already lowered the multiplier in AOD, we know the maximum CPU multiplier and now we already know the maximum base frequency that we can use. With the Black Edition processor, we can experiment with any combination within these limits to find the maximum value for other frequencies such as Northbridge frequency, HyperTransport link frequency, and memory frequency. For now, we will continue our overclocking tests as if the CPU multiplier was locked at 13x. We will look for the maximum CPU frequency by increasing the bus frequency by 5 MHz at a time.

Whether overclocking via BIOS or via AOD, we can always go back to the 200MHz base clock and set the multiplier back to 13x, which will give us a stock clock speed of 2600MHz. By the way, in this case, the northbridge multiplier will still remain 4, which gives a frequency of 800 MHz, the HyperTransport channel will operate at 800 MHz, and the memory will operate at 200 MHz (DDR2-400). We will follow the same procedure for increasing the base frequency in small increments, performing stability tests each time. If necessary, we will increase the CPU voltage until we reach the maximum CPU frequency (by turning on the ACC in parallel).

Maximum performance boost

Having found the maximum CPU frequency of our AMD processors, we have taken a significant step towards increasing system performance. But the processor frequency is only part of the overclocking. To squeeze the maximum performance, you can work on other frequencies. If you increase the voltage of the north bridge (NB VID in AMD OverDrive), then its frequency can be increased to 2400-2600 MHz and higher, while you increase the speed of the memory controller and L3 cache. Increasing the frequency and reducing the delays of RAM can also have a positive effect on performance. Even the high performance DDR2-800 memory we used can be overclocked to over 1066 MHz by increasing the voltage and possibly lowering the latency. The HyperTransport link frequency usually does not affect performance above 2000 MHz and can easily lead to instability, but it can also be overclocked. The PCIe frequency can also be slightly overclocked to somewhere around 110 MHz, which can also give a potential performance boost.

As all the mentioned frequencies slowly rise, stability and performance tests should be carried out. Setting different parameters is a lengthy process, perhaps beyond the scope of our guide. But overclocking is always interesting, especially since you will get a significant performance boost.

Conclusion

Let's hope that all our readers who want to overclock an AMD processor now have enough information on hand. Now you can start overclocking using the AMD OverDrive utility or other methods. Keep in mind that the results and the exact sequence of steps will vary from one system to another, so don't blindly copy our settings. Use this manual only as a guide to help you find the potential and limitations of your system on your own. Take your time, don't step up, monitor temperatures, run stability tests, and bump the voltage up a bit if needed. Always carefully feel the limit of safe overclocking, because a sudden increase in frequency and voltage blindly is not only the wrong approach for successful overclocking, but it can also damage your hardware.

Last tip: each motherboard model has its own characteristics, so it doesn’t hurt to get acquainted with the experience of other owners of the same motherboard before overclocking. Tips from experienced users and enthusiasts who have tried this motherboard model in work will help to avoid "pitfalls". Of course, feel free to ask for advice in our "Experts Club" at the link for discussing this article (it is given below).

Addition

We have tested another instance of the AMD Phenom II X4 940 Black Edition processor, provided by the Russian representative office of AMD. It ran successfully at 3.6 GHz when we increased the supply voltage to 1.488 V (CPUZ data). It seems that 3.6 GHz is the threshold for most processors when air-cooled. We successfully overclocked the memory controller to 2.2 GHz.

Pay attention to how the tracks go on the board: a bus goes separately from the CPU to the memory and separately to the north
bridge (AGP tunnel).

After AMD announced the beginning of the transition to 64-bit computing in 1999 and its work on the x86-64 architecture, it became necessary to develop a new technology for transferring information between various nodes of the system, since all existing technologies for connecting chips did not provide the necessary data exchange rate. .

Let's look back

In general, the need to increase the data transfer rate between the elements of the system appeared quite a long time ago. Back in 1997, AMD began work on technology LDT (Lightning Data Transfer- lightning-fast data transfer). In 2000, AMD announces that it has entered into an agreement with Transmeta to license LDT technology. AMD, in turn, gets access to technologies that reduce the power consumption of processors. In February 2001, AMD opens the technology for general licensing, while changing its name to HyperTransport. HT is positioned as a high-speed data transfer bus for personal computers, workstations and servers based on AMD microprocessors, however, the company does not exclude the possibility of using this technology in other parts of the computer, for example, to integrate all intra-system buses, such as PCI, AGP, DRAM, PCI-X, other high-speed ports, use HT in routers and switches. Broadcom, Cisco Systems, Apple Computer, nVidia, and Sun Microsystems were the first to get interested in the technology. Together, they formed a consortium HyperTransport Technology Consortium(http://www.hypertransport.org/). Then, over a short period of time, more than 40 companies joined the alliance.

In 2003, Gabriel Sartori, President of the HyperTransport Technology Consortium, announced the emergence of a new modification of the HyperTransport Technology I / O Link Specification 1.05, and in February 2004 the HyperTransport Release 2.0 Specification was completed.

HT - what kind of animal?

I want to warn you right away that in this article we will not talk about Hyper-Threading technology, throughout the text HT is an abbreviation for HyperTransport. So, HT is a new technology designed to increase the data transfer rate on the system bus, since it has traditionally been a limiting factor in the growth of overall system performance. Due to the increase in the speed of the processor, memory, video system and some other components, it is necessary to make the interaction between them more efficient, that is, to increase the speed of data exchange. This is not a new problem. At one time, the extension bus underwent major changes, which evolved into a general-purpose bus PCI (Peripheral Component Interconnect). Then came the AGP specification, designed specifically to speed up the transfer of graphic data. However, PCI and AGP technologies are becoming obsolete, and can no longer provide sufficient transfer speed. Devices are forced to "compete" for the resources used, and no more than three devices can work on the bus at the same time.

HyperTransport is not just a new system bus, it is a new asynchronous bidirectional communication protocol between devices. Absolutely any device can support HT technology: processors, logic sets, controllers, etc. The components of the system are connected to each other on a peer-to-peer basis, which means that a connection can be easily established between almost any computer nodes, and without any additional bridges (theoretically, of course :)). Information is exchanged in packets at a rate of 0.8 Gbps to 89.6 Gbps (51.2 Gbps in the first version of NT). The bus is bidirectional, that is, it has two connections: one in the forward direction and one in the reverse direction. Data is transmitted on two edges of the strobe pulse (DDR). The resulting speed depends on the bus width (2-32 bits in each direction) and its frequency (200-1400 MHz, in the first version - 200-800).

For example, in the nForce3 chip from nVidia, HT is used to connect the north and south bridges. It uses an 8-bit connection at a clock frequency of 200 MHz. At the same time, the effective bus frequency is 400 MHz, and the bandwidth is 800 MB / s.

Calculate the data transfer rate for the connection specified in the example:

• The bandwidth in one direction is 8 bits, that is, 1 byte;


• Bus frequency - 200 MHz;


• 200 MHz*2 (since DDR) = 400 MHz effective;


• Transfer rate in one direction - 400 MHz * 1 byte = 400 MB / s;


• Transfer rate in two directions (total bandwidth) - 2 * 400 MB / s = 800 MB / s

Since HT is designed to replace existing buses and bridges used in modern motherboards, motherboards built using HT technology do not have the usual chipset consisting of a northbridge designed for high-speed nodes and a southbridge used for low-speed peripherals. HyperTransport allows you to flexibly customize the system for specific goals and objectives (this is a big plus of the technology). Using HT modules, you can daisy-chain other high-performance buses and ports onto the HyperTransport bus. For example, it is easy for a server to replace a graphics tunnel with a PCI-X bus tunnel, and for a graphics station, it is easy to enable both tunnels at the same time.

Iron

Since HyperTransport technology is designed to standardize and unify the order of data exchange between all computer nodes, its implementation affects all levels of data transfer: physical (pinout for chipsets), connection level (order of initialization and configuration of devices), protocol level (protocol commands and flow control rules data), transaction level (description of control signals) and session level (general commands).

Consider the first, physical level. Here, HyperTransport defines parameters for data lines, control lines, and clock lines. In addition, controllers and electrical signals are standardized. All physical devices involved in technology8 are divided into several types: cave (cave), tunnel (tunnel) and bridge (bridge). Devices of the “cave” type are the last (closing) device in the chain, the “tunnel” is designed for the transit of information between devices, while the “bridge” is the main device that connects to the bus controller (host) and provides connection with devices connected to it.

The northbridge is now to the left, between the CPU and AGP, since there is no need to place it closer to the memory.

In the smallest possible implementation, the HT bus can be as little as 2 bits. This will require 24 pins (8 for data, 4 for clock signals, 4 for control lines, 2 for signal, 4 for ground, 1 for power, 1 for reset). And in a configuration with a 32-bit bus, you will have to use 197 pins. By the way, PCI 2.1 uses “only” 84 pins, while PCI-X has as many as 150.

The HT bus can be up to 61 centimeters (24 inches) long with a throughput of up to 800 Mbps. In this case, the signal level is 1.2 V, and the differential resistance is 100 ohms. The data transfer method that HyperTransport is physically based on is called LVDS (Low Voltage Differential Signaling- low-voltage differential signals).

The clock frequency of the connections can be from 200 to 1400 MHz depending on the requirements.

Data

As already mentioned, HT technology uses packet data transmission. In this case, the packet is always a multiple of 32 bits, and the maximum length of the packet is 64 bytes (including addresses, commands and data). Because the bus is bidirectional, each connection consists of a transmit (Tx) subconnection and a receive (Rx) subconnection. In this case, both work asynchronously. Each connection can be 2, 4, 8, 16, 32, or 64 bits wide in each direction.

Now let's say that we have a processor that needs a high-speed connection - we use two 32-bit connections with a frequency of 800 MHz, thus getting a speed of 6.4 GB / s for receiving and transmitting (the total bandwidth of such a bus will be 12.8 GB /with). If we do not need such a speed, we can use a four-bit bus with a frequency of 200 MHz. Such a bus will provide up to 100 MB / s for reception and the same amount for transmission. That is, the specification assumes the possibility of choosing a frequency and a bus when developing a device. At the same time, devices with different bus widths can connect to the same HyperTransport bus and freely communicate with each other. For example, a device with a 32-bit bus can be connected to an 8-bit device, while the throughput will be due to the smaller bus width.

For those devices that are demanding on bus bandwidth, HT implements virtual channel technology - StreamThru. This technology ensures that high-speed devices get fast access to RAM through a reserved channel.

HT vs PCI Express

As you may have noticed, there is no mention of Intel Corporation anywhere next to HyperTransport. The thing is that Intel is promoting its technology to increase the speed of the peripheral bus: PCI Express. Both buses have several similarities: a similar request generation mechanism, similar prioritization mechanisms, similar scaling capabilities.

The south bridge, in fact, has not changed.

The main difference between the technologies is their original purpose: PCI Express is a new high-speed peripheral bus, and nothing else. It is designed to work with expansion cards, while HyperTransport is a fundamentally new technology for communication and data exchange between all computer nodes. Of course, these nodes can also be expansion cards.

The packet length and control buffers in HT are 64 bytes, while in PCI Express the packet size can be up to 1 kB, the request size can be up to 4 kB, and the buffer size is 16 bytes. Since PCI Express was originally designed for high-performance servers, it has a higher cost, but at the same time it achieves higher speed than HyperTransport.

PCI Express is not compatible with either PCI or AGP, its use requires new BIOS versions and new drivers, while HT is fully compatible with the current PCI software model.

But in fact, all these comparisons can be omitted, since HyperTransport can also be adapted to PCI Express. Simply put, PCI Express devices can be connected via HyperTransport.

HT in action

Let's now take a look at HyperTransport in action and compare it to Intel technologies. The classic motherboard chipset consists of two microcircuits (north and south bridges): one includes the processor bus, memory controller, AGP and south bridge bus, the second contains various I/O controllers and a PCI bus controller. Intel systems use just such a classic system. Processors (or processor in desktop systems) are connected to memory through a memory controller integrated into the northbridge. In HyperTransport technology, all devices are connected to a single host controller. Moreover, it should be noted that AMD began to integrate the memory controller into its processors, which means that it was removed from the chipset, which somewhat accelerated the work with RAM. Thus, each processor was able to have its own memory. This allows up to 16 GB of memory (four gigabytes for each of the four processors).

In addition, AMD has decided to get rid of the restrictions imposed by the scheme with the north and south bridges. The memory controller, as well as some of the AGP (GART) functions are now implemented in the processor. The HyperTransport controller is also located there. Three separate chips were created for AGP, I/O controllers, PCI controller: AGP tunnel, PCI-X I/O Bus Tunnel and I/O Hub controller. This separation allows you to design a system for specific tasks. Only the last controller is needed for operation (you can do without AGP and PCI-X), in server systems you will hardly need an AGP video card, and in desktop systems PCI-X devices are not yet in demand. By the way, nVidia in its nForce3 chipset combined all the controllers into one chip.

Future

In February of this year, a new version of the technology was introduced - HyperTransport Release 2.0 Specification. The new specification supports three new speed implementations: 1 GHz, 1.2 GHz and 1.4 GHz. In addition, compatibility with the PCI Express interface has become an important feature in the HT2.