PC Hardware Tuning & Acceleration

  < Day Day Up > 

(Translated from http://www.fcenter.ru/forprint.shtml?processors/5798# and http://www.fcenter.ru/forprint.shtml?processors/5814# with the permission of http://www.fcenter.ru, a Russian-language Web site.)

Contemporary processors with high energy-consumption (including both AMD and Intel processors) can have a cooling problem because they produce considerable thermal power. It is highly desirable that this heat be not only evacuated from the processor core, but also from the system unit case, so that the temperature of the air flowing around the heatsink be as low as possible.

With the release of each new Pentium 4 model, experts renew their requirements for PC manufacturers, especially those related to the temperature modes of the processor. Among these requirements are the maximum temperature of the CPU core, the maximum temperature of the air inside the system unit cover, and the recommended thermal resistance of the installed cooler. Provided that these requirements are observed, Intel guarantees stable operation of its processors. AMD also specifies similar requirements for PCs equipped with Athlon XP. The main issue is whether or not the PC manufacturer observes these requirements, and if yes, to what extent.

With well-known brands such as Hewlett-Packard or Dell Computer, the manufacturer has carefully tested the temperature characteristics of each component chosen for assembling the computer, including the case and cooler. However, if you purchased your computer from a less well-known vendor or even assembled it yourself, the situation is quite different. Are you sure that all requirements of Intel or AMD for temperature modes have been met? Even the calculations of the component manufacturers might be useless, if you implement modes that differ from the nominal or recommended ones. Overclocking increases the power consumption and heat generation of respective components. The CPU core temperature grows accordingly. This often results in instable operation of the component and the entire computer system.

Instable operation of an overclocked computer is an ordinary occurrence. If an overclocked processor operates slower than expected, the situation is different. Such situations are typical on Pentium 4. At first, they may seem surprising. Such behavior is caused by overheating of the processor chip — to be precise, by the operation of the Thermal Control Circuit. This system controls the processor performance, depending on its temperature. The following experiments confirm this.

Thermal Monitor and Thermal Control Circuit

Intel developers have implemented new technology known as the Thermal Control Circuit in Pentium 4. This technology is aimed at ensuring stable operation and at protecting processors from damage caused by overheating. The architecture of all Pentium 4 processors includes two built-in temperature sensors, which actually are specialized thermal diodes. One of these sensors, integrated into the chip, informs the system BIOS and the hardware-monitoring subsystem of the CPU core temperature. Another sensor located in the "hottest" zone of the core — near the Arithmetic Logic Unit (ALU) — is an integral part of the Thermal Monitor layout.

It is necessary to mention that Athlon XP contains a similar sensor. However, the difference between Athlon XP and Pentium 4 is significant. The thermal diode of Athlon XP informs the motherboard of the temperature of the CPU core. Special logical circuits of the motherboard process the received data and power down the system if the CPU temperature reaches the temperature threshold. Thus, the hardware control circuits prevent the processor from being damaged by overheating. If this happens when the OS is running and data is being processed, all unsaved data probably will be lost.

Temperature monitoring in Pentium 4 is based on different principles. As a result, the computer retains overall stability even when the temperature threshold is reached. The system is powered down only in emergencies. This means that the processor protects itself from overheating and is capable of continuing stable operation of the system and application programs.

To meet this requirement, the Pentium 4 core contains a special integrated circuit that compares the current and threshold temperatures. This circuit, known as the Thermal Monitor, complements Thermal Control Circuit logical boards that control the heat generation of the CPU.

The operating principle of the Thermal Monitor is based on the comparison of two currents: one flowing through the thermal diode, and one taken from a separate source as a predefined reference value. The resistance of the thermal diode depends on its temperature. Consequently, the current flowing through the thermal diode will change with the CPU core temperature. By comparing the reference value of the current passing through the thermal diode, it is possible to determine whether or not the temperature threshold has been reached. The Thermal Monitor is designed to complete a simple task: If the temperature in the hottest location of the CPU exceeds the threshold, the Thermal Monitor generates the PROCHOT# signal. As a result, the Thermal Control Circuit is activated to reduce the generated heat and prevent the temperature from rising.

Erroneous opinions related to the operation of the Thermal Control Circuit are common. One of the most frequent fallacies states that Pentium 4 reduces its clock frequency when overheated. Suppose that the processor operated at 2.2 GHz when cooled adequately. If it became overheated, the Thermal Control Circuit would reduce its clock frequency to 1.8 GHz or even lower. To understand this, you must know how the operating frequency is generated in the CPU.

Figure 8.27: Generation of the operating frequency in Pentium 4

Suppose that the motherboard supplies a frequency of 133 MHz via the FSB to the processor. In the processor, this frequency is multiplied by the multiplier value. (Pentium 4 2.8 GHz has a multiplier of 21x). The resulting frequency (2.8 GHz, in this case) of the generated signal used by internal components is the value marked on the CPU. This value is read by programs such as WCPUID. It is this high-frequency signal that determines the CPU clock speed and operation of the ALU. This signal is controlled by the Thermal Control Circuit.

Thus, under normal temperatures, clock pulses supplied to the arithmetic units of Intel Pentium 4 2.8 GHz have a frequency of 2.8 GHz. However, when the processor temperature reaches the threshold, the Thermal Monitor issues the PROCHOT# signal. As a result, the Thermal Control Circuit is activated. The Thermal Control Circuit, in turn, modulates the signal supplied to the processor, and determines how many pulses must be discarded to decrease the heat generated by the CPU. Modification of the clock signal is illustrated in Fig. 8.28.

Figure 8.28: Modification of the clock signal supplied to Pentium 4 by the thermal control system

A portion of the clock pulses generated by the CPU's multiplying unit is discarded by the thermal control system. As a result, the operating intensity of the calculation units of the processor, which idle during idle clocks, is decreased. Consequently, the processor performance and heat generation decrease, even though both the motherboard and the internal clock pulse generator specify a value of 2.8 GHz.

According to the data provided by Intel, the resulting frequency can be lower than the nominal by between 30% and 50%, depending on the processor model.

As the core temperature decreases, the Thermal Control Circuit logical unit gradually returns the processor to the normal mode of operation, reducing the number of discarded clock cycles and increasing the intensity of operation for internal CPU circuitry. When the CPU core temperature drops approximately 1°C (33.8°F) below the threshold (the so-called value of temperature hysteresis), the Thermal Monitor will stop issuing the PROCHOT# signal. After that, the idle cycles enforced by the Thermal Control Circuit will disappear, and the internal CPU circuits will run at the specified clock frequency (in this example, 2.8 GHz).

By default, the Thermal Control Circuit is disabled for all processors of the Pentium 4 family. To initialize it, you must include an appropriate functionality in the motherboard BIOS. The Thermal Control Circuit is enabled when the computer is powered on, or later at the operating system boot (via special drivers or system software).

What benefits are provided by the Thermal Control Circuit technology? To answer this question, it is sufficient to imagine a situation in which a Pentium 4 processor is inadequately cooled. Such a situation can arise if the computer is equipped with an inferior cooler, if there is no thermal paste between the CPU surface and heatsink, or if the case of the system unit is stuffed with electronic components and has no extra coolers. As a result, the processor overheats, and its performance will be much lower than the performance of a CPU that operates in an optimal temperature mode.

Another common situation is related to an overclocked processor. Such a processor generates more heat than a CPU that runs at the nominal frequency. An overclocked processor may demonstrate performance even lower than that of a processor running at the nominal frequency. This is possible because, under conditions of inadequate cooling, a core that operates at a high intensity can reach the temperature threshold, which would cause the Thermal Monitor to activate. After the Thermal Monitor was activated, the Thermal Control Circuit would force the processor to discard clock cycles. Consequently, at the computer boot time (at the stage of the POST routine) and later, when the programs such as WCPUID start, the user would be shown the clock frequency to which the processor was overclocked. However, the real performance of that processor might be much lower.

When considering the Thermal Control Circuit technology, it is necessary to mention that its implementation allows you to prevent processor overheating, but it won't preserve system stability (for example, if the cooler fails). To prevent processor damage in such emergencies, another temperature sensor has been integrated with the core. This sensor tracks another temperature value — the one that prompts the THERMTRIP# signal, which initiates system shutdown. This value is lower than the level at which CPU semiconductor circuits are destroyed. Even if the fan of the processor cooler fails, the computer will be powered down before it reaches the fatal core temperature. The temperature measurement interval is no more than several dozen nanoseconds; therefore, the Thermal Control Circuit will save the overheated processor even if you physically remove the CPU cooler. The temperature that prompts the THERMTRIP# signal is about 135°C (275°F), according to the manufacturer.

Testing the Temperature Control Facilities

The primary goals of this test were to determine the threshold that activates thermal protection and to reveal the dependence between the Pentium 4 processor and its core temperature.

Configuration of the Test System

To investigate the influence of Thermal Monitor and Thermal Control Circuit operation on the processor performance, it was necessary to ensure a smooth increase in its temperature. Reaching the temperature threshold was determined by the change of performance, which had to be controlled constantly.

Unfortunately, the fan speed control using Zalman FanMate was insufficient.

Figure 8.29: Zalman FanMate regulator

First, it was impossible to smooth the increase in CPU temperature. Furthermore, the lower the fan speed, the lower the cooler efficiency. At low rotation speed, the dependence of the cooling efficiency on the rotation speed was nonlinear. Because of this, another method was chosen.

The processor temperature at a constant workload, if the same cooler is used, will depend almost linearly on the temperature of the air within the system unit case, which, in turn, depends on the ambient temperature. By smoothly increasing the ambient temperature, it is possible to smoothly increase the processor temperature.

Measuring the room temperature with a precision of 1°C is difficult; however, this precision was desirable for testing purposes. This was achieved using the MIR-253 thermally isolated chamber from Sanyo. This device has a case with the dimensions of 162×50×70 cm. The ATX system unit can be placed within it. A special outlet in the case of the thermal camera allows the connection of the monitor, keyboard, mouse, and power cable. The computer case is placed into the thermally isolated medium, the temperature of which doesn't depend on the ambient temperature and can be set manually.

Figure 8.30: MIR-253 thermally isolated chamber from Sanyo

MIR-253 includes a heating element and cooling system similar to the ones used in refrigerators. The incubator consumes 220 W and is capable of supporting an internal temperature ranging from -10°C (+14°F) to +50°C (+122°F), with a precision of 0.1°C (32.18°F). However, such measurement precision could not be achieved, because a computer is a constant and powerful source of heat. Consequently, the temperature within the incubator constantly changes by 0.5°C (32.9°F) to 1°C (33.8°F).

This precision was sufficient for performing the planned measurements.

Using MIR-253, it became possible to measure the ambient temperature for the computer with a precision of 1°C. This ensured adequate measurements of the CPU temperature.

A decision was made to trace the CPU performance and its dependence on the temperature using a real application, rather than synthetic test. For this purpose, the Unreal Tournament 2003 game and Fraps program were used. The latter displays in real-time the number of frames per second achieved in a Direct3D application. In contrast to the built-in tools of performance measurements implemented in various games, such as Unreal Tournament or Quake III Arena, Fraps displays the result in a large font, which is important when starting the application in a window with low resolution. Generally, Fraps also consumes system resources and influences the displayed results. However, this isn't important, because it is the change of the performance parameters that is of interest in this case. The settings of the Unreal Tournament were changed to decrease the workload on the video adapter as much as possible, making performance more dependent on the processor. The game was started at a screen resolution of 320 × 240 with a 16-bit color depth, in the Instant Action mode, at the DM-Asbestos level. After the game started, a spot was chosen with a constant number of frames per second. At this point, the main personage was stopped. From this moment, the tester touched neither the keyboard nor the mouse. During the investigation, the rendering speed was registered for the same scene depending on the processor temperature. Because the game was started in a window, rather than in full-screen mode, it was possible to constantly monitor the processor temperature using the Asus PC Probe program supplied with the motherboard, without stopping the testing process. All that remained after starting Unreal Tournament was to gradually increase the temperature within the thermally isolated chamber.

Influence of Hyper-Threading on the CPU Temperature

Hyper-Threading technology, first implemented in Intel Pentium 4 3.06 GHz, is one of the newest achievements of Intel. This technology allows two tasks or two code fragments of the same program to be executed simultaneously on one physical processor. Thus, one processor is interpreted by the operating system as two logical devices, which operate in parallel. This functional capability can be enabled or disabled in BIOS Setup.

In the course of analyzing the temperature influence on the CPU performance, there arose the problem of detecting the dependence of the thermal power on the usage of hyperthreading technology.

In experiments conducted at 20°C (68°F), the following testing programs were used: CPUburn, SiSoftware Sandra 2003, and Unreal Tournament 2003. The first program from this list loads the processor with intense calculations. As a result, the processor heats. Other programs are used as tests.

The CPU temperatures achieved as a result of running these programs, as well as the temperature of the CPU in the idle mode, are shown in Fig. 8.31. Temperature measurements were conducted using Asus PC Probe.

Figure 8.31: Influence of hyperthreading technology on the CPU temperature

As is obvious from the illustration, the CPU temperature in the idle mode decreased when hyperthreading technology was enabled. For Unreal Tournament, the CPU temperature without hyperthreading was much higher. This is why the decision was made to disable hyperthreading technology when investigating the operation of the Thermal Monitor and the Thermal Control Circuit. This had to simplify the CPU heating.

Test Results

Investigation of Thermal Monitor and Thermal Control Circuit operation was started at the following parameter values:

In the course of testing, the temperature of the ambient air was increased smoothly, and changes of the following parameters were registered:

This test showed that the speed of the game test didn't change until the CPU temperature approached the threshold of 72°C (161.6°F) (Section I). From this moment, the speed in Unreal Tournament started to decrease rapidly. Before conducting measurements, it seemed possible that the Thermal Control Circuit would decrease the CPU speed with each additional 1°C increase of its temperature. However, in Fig. 8.32, there are regions in which the CPU temperature doesn't change, even though the speed continues to decrease (Sections II and III). This is evidence that the implementation of protective elements allows Pentium 4 to resist the increase of its core temperature and even to support the temperature at a constant level for some time.

Figure 8.32: Influence of the Thermal Monitor and the Thermal Control Circuit on the CPU performance (1 = temperature of the ambient air, 2 = temperature of the air within the computer case, 3 = CPU temperature, and 4 = speed in the game test)

As a result, the ambient temperature in the course of testing has risen from 27°C (80.6°F) to 50°C (122°F). The temperature within the computer case rose from 44°C (111.2°F) to 63°C (145.4°F), and the CPU temperature changed from 69°C (156.2°F) to 85°C (185°F).

It became possible to experimentally determine the threshold at which the Thermal Monitor starts to issue the PROCHOT# signal and the Thermal Control Circuit starts to slow the operation of the internal processor circuitry. For the Pentium 4 3.06 GHz processor being investigated, this was 72°C (161.6°F).

Increasing the CPU temperature 13°C (55.4°F) from this threshold resulted in a more than twofold performance drop. The frames per second in Unreal Tournament decreased from 115 to 49.

Unfortunately, a further increase of the temperature was impossible because of the limited temperature range of MIR-253: the maximum temperature that can be supported by this device is 50°C (122°F). Because of this, the CPU cooler fan had to be switched. As a result, the CPU temperature increased to 94°C (201.2°F) in less than a minute, and the computer was powered down. During this time, the speed measured by Unreal Tournament hardly changed. Therefore, it appears that decreasing performance 2.3 times is the maximum, at least for the used model of Pentium 4 3.06 GHz.

To perform an additional check of this hypothesis, an experiment was conducted using CPU RightMark 2 RC3. This test displays the CPU performance in real-time mode and allows its changes in the course of the test to be traced. Taking into account the relatively short time during which this test created a graph, the testing method was changed. To increase the temperature, the cooler fan was switched off. In such a case, the CPU temperature rises at a rapid rate. This allows data that serve as evidence of the performance decrease to be registered. Still, the CPU temperature increase is not sufficient to cause a system hangup.

The results of the conducted experiment are shown in Fig. 8.33. These results can be used to determine the limit at which the CPU speed can be controlled. In this case, the performance decreased 2.7 times.

Figure 8.33: Results of the CPU RightMark 2 RC3 test (Intel Pentium 4 3.06 GHz)

Based on the results of investigations of Thermal Monitor and Thermal Control Circuit operation, several conclusions can be drawn.

The protective tools implemented in Pentium 4 cannot retain CPU stability when the CPU cooler fan is switched off. However, they are capable of significantly widening the operating temperature range. At the same time, Pentium 4 prevents core overheating at the expense of an almost threefold performance drop.

As the testing results have shown, a 3°C increase in the CPU core temperature (from 72°C to 75°C) degraded performance 10% (for Pentium 4 3.06 GHz).

If cooling is inadequate, the operation of Pentium 4 can slow. Because of this, overclocked processors can be even slower than the ones running at nominal clock frequencies: The Thermal Control Circuit will always determine the ratio between performance and temperature independently of the reason that overheating occurred. Replacing the cooler with more powerful one or installing an additional fan into the computer case can solve the problem. However, to choose optimal cooling facilities, it is necessary to know both the thermal power of the CPU and the temperature threshold that activates the protection system.

This investigation relates only to Pentium 4 3.06 GHz. The parameters Thermal Monitor and Thermal Control Circuit of earlier Pentium 4 models are also interesting.

Investigating Earlier Pentium 4 Models

As previously mentioned, there are two thermal sensors in Pentium 4. The first sensor is an integral part of the Thermal Monitor circuit. It is installed at the hottest point of the core and is used by the Thermal Monitor to compare the CPU temperature to the threshold value at which the PROCHOT# signal is issued. This sensor is unavailable to hardware-monitoring programs, and it is impossible to read the values that it registers. The second sensor is also installed in the processor core; however, it is impossible to tell where it resides. The values measured by this sensor are available to the motherboard BIOS and hardware-monitoring programs. These sensors are installed separately. There is some distance between them, and it is possible to guess that the CPU temperature measured by them will be different. However, for the practical purposes, assume that in Pentium 4, they show the same temperature (i.e., the temperature displayed by the hardware-monitoring program corresponds to the temperature of the Thermal Monitor sensor).

Configuration of the Test System

The method of testing was similar to the one previously described and comprised two stages. In the first stage, the Unreal Tournament 2003 game and Fraps program were used for testing. In the second stage, during which the CPU performance was measured under conditions of quick heating, the CPU RightMark 2 RC3 test was used (the CPU overclocking module).

Figure 8.34: Standard cooler supplied with Pentium processors

Intel Pentium 4 1.6 GHz

This processor is based on the Willamette core with 256 KB L2 cache memory operating at a frequency of 100 MHz, which ensures data transmission at a frequency of 400 MHz. The model under consideration was Stepping 2, Revision D0.

Figure 8.35: Parameters of Intel Pentium 4 1.6 GHz

The temperature threshold that activated the Thermal Monitor and the Thermal Control Circuit was 74°C (165.2°F). Starting from this temperature, the performance of the CPU decreased. As can be seen from the results in Fig. 8.36, the temperature in this test grew to 77°C (170.6°F), and the performance dropped from 77 frames per second to 41 frames per second, or about 1.8 times.

Figure 8.36: Influence of the Thermal Monitor and the Thermal Control Circuit on the performance of Pentium 4 1.6 GHz (1 = CPU temperature, and 2 = speed shown by the Unreal Tournament test)

Figure 8.37: Results of the CPU RightMark 2 RC3 test (Intel Pentium 4 1.6 GHz)

During the CPU RightMark 2 RC3 test, the CPU performance smoothly decreased 1.9 times as a result of the core heating. After the CPU cooler was powered up, the performance returned to its initial value.

The temperature threshold of the system power-down was 110°C (230°F).

Intel Pentium 41.8 GHz

This processor has the Northwood core with 512 KB of L2 cache memory operating at a clock frequency of 100 MHz, which ensures data transmission at 400 MHz and a core voltage of 1.5 V. The model under consideration was Stepping 4, Revision BO.

The temperature threshold that activated the Thermal Monitor and the Thermal Control Circuit was 68°C (154.4°F). At this temperature, the CPU performance started to decrease. In the course of this test, the temperature grew to 72°C (161.6°F), and the performance dropped from 87 frames per second to 38 frames per second, or about 2.28 times.

Figure 8.38: Parameters of Intel Pentium 4 1.8 GHz

Figure 8.39: Influence of the Thermal Monitor and the Thermal Control Circuit on the performance of Pentium 4 1.8 GHz (1 = CPU temperature, and 2 = speed shown in the Unreal Tournament test)

Figure 8.40: Results of the CPU RightMark 2 RC3 test (Intel Pentium 4 1.8 GHz)

In the course of the CPU RightMark 2 RC3 test, the CPU performance smoothly decreased 2.7 times. After the cooler fan was turned on again, the CPU temperature decreased and performance returned to initial value.

The temperature threshold of the complete system power-down was 95°C (203°F).

Pentium 4 2.0 GHz

This processor is based on the Northwood core with 512 KB L2 cache operating at a 100 MHz clock frequency, which ensures data transmission at a frequency of 400 MHz. The core supply voltage is 1.5 V. The model being investigated was Stepping 7, Revision C1.

The temperature threshold that activated the Thermal Monitor and the Thermal Control Circuit was 73°C (163.4°F). Starting from this temperature, the processor performance decreased. During testing, the temperature grew to 77°C (170.6°F), and the performance decreased from 92 frames per second to 41 frames per second, or about 2.2 times.

In the course of testing using CPU RightMark 2 RC3, processor performance smoothly decreased 2.62 times due to heating. After the CPU cooler was powered up again, the performance returned to its initial value.

The temperature threshold at which the system powered down was 98°C (208.4°F).

Figure 8.41: Results of the CPU RightMark 2 RC3 test (Intel Pentium 4 2.0 GHz)

Figure 8.42: Influence of the Thermal Monitor and the Thermal Control Circuit on the performance of Pentium 4 2.0 GHz (1 = CPU temperature, and 2 = speed shown in the Unreal Tournament test)

Figure 8.43: Results of the CPU RightMark 2 RC3 test (Intel Pentium 4 2.0 GHz)

Summary Data

The main parameters of the Pentium 4 models used in these tests and of their built-in Thermal Monitor and Thermal Control Circuit temperature protection tools are provided in Table 8.15.

Table 8.15: Parameters and Test Results of the Processors

Parameter

Processor

Pentium 4 1.6 GHz

Pentium 4 1.8 GHz

Pentium 4 2.0 GHz

Pentium 4 3.06 GHz

Core architecture

Willamette

Northwood

Northwood

Northwood

Voltage (V)

1.75

1.50

1.50

1.60

Stepping

2

4

7

7

Revision

D0

B0

C1

C1

Heat emission (W)

60.8

49.6

54.3

81.8

Thermal Monitor threshold value (°C)

74

68

73

72

Performance drop shown by Unreal Tournament/CPU RightMark (times)

1.8/1.9

2.3/2.7

2.2/2.6

2.3/2.7

CPU power-down temperature (°C)

110

95

98

94

To conclude, it is necessary to mention once again that the CPU core has two built-in thermal sensors. The built-in motherboard hardware-monitoring tools have access to only one of them. The Thermal Monitor and the Thermal Control Circuit use information from the second sensor to control the processor performance. This sensor is located in the core area, for which high heat emission and temperatures are characteristic. These values strongly depend on the calculation intensity and the workload of the units located in this area. It is necessary to point out that the temperatures of the sensors can be different from each other and can change independently of the tasks being solved.

In the experiments described here, the threshold of temperature protection was based on readings of the hardware-monitoring sensor. The readings of this sensor may not coincide with the readings of the thermal protection sensor. Therefore, the threshold values for the temperature protection activation depend on the tasks being solved. Under different conditions, and when solving other tasks, these values can be different from the ones presented in Table 8.15.

  < Day Day Up > 

Related

Категории