Relevant information comes from Zhongguancun Online and IT 168.
The first is the CPU article
interface type
We know that the CPU needs to be connected to the motherboard through an interface to work. After so many years of development, the interface methods adopted by CPU are pin, card, contact and pin. At present, the interfaces of CPU are all pin interfaces, and the corresponding motherboards have corresponding slot types. Different types of CPU interfaces have different jacks in number, volume and shape, so they cannot be plugged into each other.
Socket 775
Socket 775, also known as Socket T, is the corresponding interface of CPU encapsulated by Intel LGA775. At present, the CPU packaged by LGA775 includes Pentium 4, Pentium 4 EE, Celeron D and so on. Different from the previous Socket 478 interface CPU, the bottom of Socket 775 interface CPU has no traditional pins, but 775 contacts, that is, it is not pin-type but contact-type, and transmits signals by contacting with the 775 contact pins in the corresponding Socket 775 slots. Socket 775 interface can not only effectively improve the signal strength and frequency of the processor, but also improve the yield of the processor and reduce the production cost. With the gradual fading out of Socket 478, Socket 775 will become the standard interface for all Intel desktop CPUs in the future.
Socket 754
When the AMD64-bit desktop platform was first released in September 2003, Socket 754 was the CPU interface. At present, there are low-end Athlon 64 and high-end Sempron with 754 CPU pins. With the popularity of Socket 939, Socket 754 will eventually fade out.
Socket 939
Socket 939 is a 64-bit desktop interface standard introduced by AMD in June 2004. At present, there are high-end Athlon 64 and Athlon 64 FX with 939 CPU pins. Socket 939 processor and Socket 940 slot cannot be mixed, but Socket 939 still uses the same CPU fan system mode, so the fans used by Socket 940 and Socket 754 can also be used for Socket 939 processor.
Socket 940
Socket 940 is the earliest published AMD64-bit interface standard with 940 CPU pins. Currently, servers/workstations use Opteron and Athlon 64 FX as this interface. With the new Athlon 64 FX switching to Socket 939 interface, Socket 940 will become Opteron's special interface.
Socket 603
Socket 603 is used professionally and applied to Intel's high-end server/workstation platform. The CPU using this interface is Xeon MP and early Xeon, with 603 CPU pins. The CPU of Socket 603 interface can be compatible with Socket 604 slot.
Socket 604
Similar to Socket 603, Socket 604 is still a high-end server/workstation platform for Intel. The CPU using this interface is Xeon of 533MHz and 800MHz FSB. The CPU of Socket 604 interface is not compatible with Socket 603 slot.
Socket 478
Socket 478 interface is the interface type adopted by Pentium 4 series processors at present, with 478 pins. The Pentium 4 processor in the socket 478 has a small area and the pins are arranged very closely. Intel's Pentium 4 series and P4 Celeron series all use this interface.
Socket a
Socket A interface, also called Socket 462, is the socket interface of AMD's Athlon XP and Duron processors. Socket A interface has 462 slots and can support 133MHz external frequency.
Socket 423
The socket 423 was the standard interface of the original Pentium 4 processor. The shape of the 423 socket is similar to that of the previous socket, and the corresponding number of CPU pins is 423. Socket 423 slots are mostly based on the motherboard of Intel 850 chipset, and support Pentium 4 processors of 1.3 GHz ~ 1.8 GHz. However, with the popularity of DDR memory, Intel developed the i845 chipset supporting SDRAM and DDR memory, changed the CPU slot to Socket 478, and the Socket 423 interface disappeared.
Socket 370
Socket 370 architecture is developed by Intel, not a socket architecture. It looks very similar to Socket 7, and it also uses zero-plug slots, and the corresponding CPU is 370 pins. Intel's famous "Copper Mine" and "tualatin" series CPU all use this interface.
Slot 1
Slot 1 is a patented CPU interface developed by Intel Corporation to replace Socket 7. In this way, other manufacturers can't produce products with slot 1 interface. The CPU of SLOT 1 interface is no longer a familiar square, but a flat cuboid, and the interface has also become a golden finger, no longer in the form of a pin.
Slot 1 is a slot designed by Intel Corporation for Pentium II series CPU. Pentium II CPU and its related control circuits and secondary cache are all on a daughter card, and most slots 1 motherboards use 100MHz external frequency. Slot 1 has advanced technical structure, which can provide greater internal transmission bandwidth and CPU performance. This interface has been eliminated, and there is no such interface product on the market.
Slot 2
Slot 2 is used professionally for high-end servers and graphics workstation systems. The CPU used is also expensive Xeon series. There are many differences between slot 2 and slot 1. First of all, the Slot 2 slot is longer and the CPU itself is larger. Secondly, Slot 2 is competent for more demanding multi-purpose computing, which is the key to enter the high-end enterprise computing market. In the standard server design at that time, general manufacturers could only use two Pentium II processors in the system at the same time. With the design of slot 2, a server can use eight processors at the same time. Moreover, Pentium II CPU with Slot 2 interface adopted the most advanced 0.25 micron manufacturing technology at that time. Motherboard chipsets supporting SLOT 2 interface include 440GX and 450NX.
Slot a
SLOT A interface is similar to Intel's SLOT 1 interface, which is used by AMD's K7 Athlon. In terms of technology and performance, SLOT A motherboard is fully compatible with all kinds of original peripheral expansion cards. It does not use Intel's P6 GTL+ bus protocol, but uses Digital's Alpha bus protocol EV6. EV6 architecture is an advanced architecture, which adopts multi-thread point-to-point topology and supports 200MHz bus frequency.
- .
Fourth, the number of needles.
At present, CPUs are all connected to the motherboard through pin interfaces, but the number of pins of CPUs with different interfaces is different. The naming of CPU interface types is usually expressed by the number of pins. For example, the Socket 478 interface currently used by Pentium 4 series processors has 478 pins. Socket 462 interface adopted by Athlon XP series processors has 462 pins.
Interface type pin number
Socket 775 775
Socket 939 939
Socket 940 940
Socket 754 754
Socket A(462) 462
Socket 478 478
Socket 604 604
Socket 603 603
Socket 423 423
Socket 370 370
- .
Fifth, the main frequency
In electronic technology, pulse signal is a continuous pulse signal with a certain voltage amplitude and a certain time interval. The time interval between pulse signals is called period; The number of pulses generated per unit time (such as 1 sec) is called frequency. Frequency is a measurement name, which describes the number of pulses that appear in a unit time for periodic cyclic signals (including pulse signals). The standard unit of measurement of frequency is hertz. The system clock in the computer is a typical pulse signal generator, and its frequency is quite accurate and stable. Frequency is represented by "f" in mathematical expression, and the corresponding units are: hertz (Hz), kilohertz (kHz), megahertz (MHz) and gigahertz (GHz). Where 1GHz= 1000MHz, 1MHz= 1000kHz, 1kHz= 1000Hz. Calculate the time unit of the pulse signal period and the corresponding conversion relationship: s (seconds), ms (milliseconds), μs (microseconds) and ns (nanoseconds), where: 1s= 1000ms, 1ms = 1000ms,.
The main frequency of the CPU, that is, the CPU clock speed at which the CPU core works. How many megahertz is a certain CPU? This megahertz is the "CPU main frequency". Many people think that the main frequency of CPU is its running speed, but it is not. The main frequency of CPU indicates the oscillation speed of digital pulse signal in CPU, which is not directly related to the actual computing power of CPU. There is a certain relationship between the main frequency and the actual running speed, but there is no definite formula to quantify the numerical relationship between them, because the running speed of CPU depends on the performance indicators of CPU pipeline (cache, instruction set, CPU bits, etc.). ). Because the main frequency does not directly represent the running speed, in some cases, it is likely that the higher the main frequency, the lower the actual running speed of the CPU. For example, most of AMD's AthlonXP series CPU can achieve the CPU performance of Intel's Pentium 4 series CPU at a low frequency, so AthlonXP series CPU is named after PR value. Therefore, the main frequency is only one aspect of CPU performance and does not represent the overall performance of CPU.
The main frequency of CPU does not represent the speed of CPU, but improving the main frequency is very important to improve the running speed of CPU. For example, suppose a CPU executes an operation instruction in a clock cycle, then when the CPU runs at 100MHz, it will be twice as fast as when it runs at 50MHz. Because the clock cycle of 100MHz is half of that of 50MHz, that is, the CPU working at 100MHz only needs 10ns to execute an operation instruction, which is half of that of 20ns working at 50MHz, and the natural operation speed is doubled. However, the overall running speed of the computer depends not only on the running speed of the CPU, but also on the running speed of other subsystems. Only when the main frequency is improved, the running speed of each subsystem and the data transmission speed between subsystems can be improved, and the overall running speed of the computer can be really improved.
Improving the working frequency of CPU is mainly limited by the production process. Since CPU is made on a semiconductor silicon wafer, the components on the silicon wafer need to be connected by wires. Because the wire is required to be as thin as possible at high frequency, it can reduce stray interference such as distributed capacitance of the wire and ensure the correct operation of CPU. Therefore, the limitation of manufacturing process is one of the biggest obstacles to the development of CPU main frequency.
- .
Six, packaging technology
The so-called "packaging technology" is the technology of packaging integrated circuits with insulating plastics or ceramic materials. Taking CPU as an example, what we actually see is not the size and appearance of the real CPU core, but the packaged products of the CPU core and other components.
Packaging is necessary and critical for the chip. Because the chip must be isolated from the outside world, to prevent impurities in the air from corroding the chip circuit, resulting in the decline of electrical performance. On the other hand, the packaged chip is also more convenient to install and transport. Because the quality of packaging technology also directly affects the performance of the chip itself and the design and manufacture of the PCB (printed circuit board) connected to it, it is very important. Packaging can also be said to refer to the shell for mounting semiconductor integrated circuit chips. It not only plays the role of placing, fixing, sealing, protecting the chip and enhancing thermal conductivity, but also plays the role of a bridge connecting the internal world of the chip and external circuits-the contacts on the chip are connected with the pins of the package shell through wires, and these pins are connected with other devices through wires on the printed circuit board. Therefore, for many integrated circuit products, packaging technology is a very key link.
At present, most CPU packages are made of insulating plastic or ceramic materials, which can seal and improve the electrothermal performance of the chip. Because the internal frequency of the processor chip is getting higher and higher, the function is getting stronger and stronger, the number of pins is increasing, and the package shape is constantly changing. Main factors to be considered when packaging:
The ratio of chip area to package area improves the package efficiency and is as close as possible to 1: 1.
Pins should be as short as possible to reduce delay, and the distance between pins should be as far as possible to ensure mutual interference and improve performance.
Based on the requirements of heat dissipation, the thinner the package, the better.
As an important part of computer, the performance of CPU directly affects the overall performance of computer. The last and most critical step in the manufacturing process of CPU is the packaging technology of CPU. There is a big difference in CPU performance between different packaging processes. Only high-quality packaging technology can produce perfect CPU products.
Packaging technology of CPU chip;
Impregnation technology
QFP technology
PFP technology
PGA technology
BGA technology
At present, the common packaging forms are:
OPGA package
MPGA package
CPGA package
FC-PGA package
FC-PGA2 package
OOI package
PPGA package
South complete sets of equipment
South ECC C2 packet
South environmental protection package
PLGA package
CuPGA package
- .
Seven, the core type
Die, also called kernel, is the most important part of CPU. The chip protruding from the center of CPU is the core, which is made of monocrystalline silicon through a certain manufacturing process. All calculations, receiving/storing commands and processing data of CPU are executed by the kernel. All kinds of CPU cores have a fixed logical structure, and logical units such as first-level cache, second-level cache, execution unit, instruction-level unit and bus interface will have a scientific layout.
In order to manage the design, production and sales of CPU, CPU manufacturers will give corresponding codes to various CPU cores, which is the so-called CPU core type.
Different CPU (different series or the same series) will have different core types (such as Northwood of Pentium 4, Willamette, CXT of K6-2, ST-50 of K6-2+ and so on). ), even the same core will have different versions (for example, Northwood core is divided into B0 and C 1, etc. ). The core version is changed to correct some errors in the previous version. Each core type has its corresponding manufacturing process (such as 0.25um, 0. 18um, 0. 13um, 0.09um, etc.). ), core area (which is the key factor to determine the CPU cost, and the cost is basically proportional to the core area), core voltage, current, number of transistors, cache size at all levels, main frequency range, pipeline architecture and supported instruction sets (these two are the key factors to determine the actual performance and work efficiency of CPU), power consumption and calorific value, and packaging methods (such as S.E.P, PGA, FC-PGA, FC-PGA). ), and interface types (such as Socket3772) Socket A, Socket 478, Socket T, Slot 1, Socket 940, etc. ), FSB, etc. So the core type determines the performance of CPU to some extent.
Generally speaking, the new core type often has better performance than the old core type (for example, the performance of Pentium 4 1.8GHz of Northwood core with the same frequency is higher than that of Pentium 4 1.8 GHz of Willamette core), but this is not absolute. This usually happens when a new core type has just been introduced, which may lead to a new core type due to imperfect technology or immature new architecture and manufacturing technology. For example, the actual performance of Pentium 4 with Willamette Socket 423 interface in the early days is not as good as Pentium III and Celeron with Tualatin Socket 370 interface, and the actual performance of Pentium 4 with low frequency Prescott kernel is not as good as that with high frequency. However, with the progress of technology and the continuous improvement and perfection of the new core by CPU manufacturers, the performance of the new core product will inevitably surpass that of the old core product.
The development direction of CPU core is lower voltage, lower power consumption, more advanced manufacturing technology, integration of more transistors, smaller core area (which will reduce the production cost of CPU and ultimately lower the sales price of CPU), more advanced pipeline architecture and more instruction sets, higher front-end bus frequency, and integration of more functions (such as integrated memory controller, etc.). ) and dual-core multi-core (that is, there are two or more CPUs in 1) The most significant thing for ordinary consumers is that they can buy more powerful CPUs at lower prices.
In the long history of CPU, there are many kinds of CPU cores. Let's briefly introduce the mainstream core types of Intel CPU and AMD CPU respectively. Introduction of mainstream core types (desktop CPU only, excluding notebook CPU and server/workstation CPU, excluding older core types).
Intel Core
Tualatin
This is the well-known "tualatin" core, which is the last CPU core of Intel on the Socket 370 architecture. Using 0. 13um manufacturing process and adopting FC-PGA2 and PPGA packaging methods, the core voltage is also reduced to about 1.5V, the main frequency is from 1GHz to 1.4GHz, and the external frequencies are 100MHz (Celeron) and respectively. Secondary cache 5 12KB (Pentium III-S) This is the strongest Socket 370 core, and its performance even exceeds that of the early low-frequency Pentium 4 series CPU.
Willamette
This is the kernel used in the early Pentium 4 and P4 Celeron. Socket 423 interface was adopted at first, and then it was changed to Socket 478 interface (Celeron only has 1.7GHz and 1.8GHz, both of which are Socket 478 interfaces), and the manufacturing process was 0. 18um, and the front-end bus frequency was 400MHz. The main frequency ranges from 1.3GHz to 2.0GHz(Socket 423) and 1.6GHz to 2.0GHz(Socket 478), and the secondary cache is 256KB (Pentium 4) and 128KB (Celeron) respectively. Note that there are also some Pentium 4 models with Socket 423 interface without secondary cache! The core voltage is about 1.75V, and the packaging methods are PPGA INT2, PPGA INT3, OOI 423 pin, 423 socket PPGA FC-PGA2, 478 socket PPGA FC-PGA2, PPGA adopted by Celeron, etc. Willamette core is backward in manufacturing technology, high in calorific value and low in performance, and has been eliminated and replaced by Northwood core.
Northwood
This is the core adopted by the current mainstream Pentium 4 and Celeron. Compared with Willamette core, the biggest improvement is to adopt 0. 13um manufacturing process and Socket 478 interface. The core voltage is about 1.5V, and the secondary cache is 128KB (Celeron) and 5 12KB (Pentium 4) respectively. The front-end bus frequency is 400/533/800MHz (Celeron is only 400MHz), and the main frequency ranges from 2.0GHz to 2.8GHz (Celeron) and 1.6GHz to 2.6GHz(400MHz FSB Pentium 4). 2.26GHz to 3.06GHz(533MHz FSB Pentium 4) and 2.4GHz to 3.4GHz(800MHz FSB Pentium 4), as well as 3.06GHz Pentium 4 and all 800MHz Pentium 4 support hyper-threading technology, and the packaging methods are PPGA FC-PGA2 and PPGA. According to Intel's plan, Northwood core will be replaced by Prescott core soon.
Prescott (man's first and last name)
This is Intel's new CPU core, which was first used on Pentium 4, and now the low-end Celeron D is also widely used. The biggest difference between it and Northwood is that it adopts 0.09um manufacturing technology and more assembly line structures. Socket 478 interface was adopted at first, and then all of them will be switched to LGA 775 interface, with the core voltage of 1.25- 1.525 V, and the front-end bus frequency of 533MHz (not supporting hyper-threading technology) and 800MHz (supporting hyper-threading technology). Compared with Northwood, the L 1 PPGA cache of 2.4GHz, 2.8GHz, 2.8GHz, 3.0GHz, 3.2GHz and 3.4GHz with 533MHz FSB is increased from 8KB to 16KB. According to Intel's plan, Prescott Core will soon replace Northwood Core, and Celeron and Prescott Core 533MHz FSB will be launched soon.
Prescott 2m
Prescott 2m is the core of Intel desktop. Unlike Prescott, Prescott 2M supports EM64T technology, which means it can use more than 4G of memory and belongs to 64-bit CPU. This is Intel's first desktop CPU with 64-bit technology. Prescott 2M core adopts 90nm manufacturing process, which integrates 2M secondary cache and 800 or 1066MHz front-end bus. At present, P4 6 series and P4EE CPU use Prescott 2M core. The performance of Prescott 2M itself is not particularly outstanding, but due to the integration of large-capacity secondary cache and the use of high frequency, the performance has been improved. In addition, Prescott 2M Core supports Enhanced Intel SpeedStep Technology (EIST), which is exactly the same as the energy-saving mechanism in Intel mobile processors. The working frequency of Pentium 4 6 series processors under low load can be reduced, and the working heat and power consumption can be obviously reduced.
AMD CPU core
The core types of Athlon XP
Athlon XP has four different core types, but they all have * * * similarities: they all use Socket A interface and are marked with PR nominal value.
Palomino
This is the core of the earliest Athlon XP, which adopts 0. 18um manufacturing technology, the core voltage is about 1.75V, the secondary cache is 256KB, the packaging method is OPGA, and the front-end bus frequency is 266MHz.
purebred
This is the first Athlon XP core with 0. 13um manufacturing process, which is divided into thoroughbred -A and thoroughbred -B versions. The core voltage is about 1.65V- 1.75V, the secondary cache is 256KB, the packaging method is OPGA, and the front-end bus frequency is 266MHz and 333MHz.
Salton
The manufacturing process is 0. 13um, the core voltage is about 1.65V, the secondary cache is 256KB, the packaging method is OPGA, and the front-end bus frequency is 333MHz. It can be seen that Barton blocked half of the secondary cache.
farmyard
The manufacturing process is 0. 13um, the core voltage is about 1.65V, the secondary buffer is 5 12KB, the packaging method is OPGA, and the front-end bus frequency is 333MHz and 400MHz.
The core type of new duron
Apple breeding
Using 0. 13um manufacturing process, the core voltage is about 1.5V, the secondary cache is 64KB, the packaging method is OPGA, and the front-end bus frequency is 266MHz. There are three kinds of labels, 1.4GHz, 1.6GHz, 1.8GHz, and the nominal value of PR is not marked.
Core types of Athlon 64 series CPU
sledgehammer
Hammer is the core of AMD server CPU, which is a 64-bit CPU with 940 interface and 0. 13 micron process. Sledgehammer is powerful, integrates three HyperTransprot buses, takes 12 pipeline as the core, 128K first-level cache, and integrates 1M second-level cache, which can be used for one-way to eight-way CPU servers. Sledgehammer integrated memory controller has less delay than the traditional memory controller located in Northbridge, and supports dual-channel DDR memory. Because it is a server CPU, it certainly supports ECC checking.
Grabbing
The manufacturing process is 0. 13um, the core voltage is about 1.5V, the secondary cache is 1MB, the packaging method is mPGA, Hyper Transport bus is adopted, and 1 memory controller with 128bit is built in. Socket 754, Socket 940 and Socket 939 interfaces are adopted.
Newcastle
The main difference between it and Clawhammer is that the secondary cache is reduced to 5 12KB (which is also the result of AMD's relatively low price policy and accelerated promotion of 64-bit CPU for market demand), and other performances are basically the same.
Winchester
Wincheste is a relatively new AMD Athlon 64CPU core, which is a 64-bit CPU with 939 interface and 0.09 micron manufacturing process. This kind of core adopts 200MHz external frequency, supports1Hyypertransport bus and 5 12K secondary cache, which is cost-effective. Wincheste integrates dual-channel memory controller and supports dual-channel DDR memory. Due to the adoption of new technology, Winchester's calorific value is less than that of the old Athlon, and its performance is also improved.
Core types of Feilong series CPU
Paris
Paris core is the successor of Barton core, which is mainly used for AMD's flash dragon, and the early 754 interface flash dragon part used Paris core. Paris adopts 90nm manufacturing technology and supports iSSE2 instruction set, which is generally 256K L2 cache and 200MHz external frequency. Paris core is a 32-bit CPU from K8 core, so it also has a memory control unit. The main advantage of CPU's built-in memory controller is that the memory controller can run at CPU frequency, and the delay is smaller than that of the traditional memory controller located in North Bridge. Compared with Socket A interface Flash CPU, the performance of Flash using Paris core is improved obviously.
Palermo
At present, Palermo core is mainly used for AMD's Flash CPU, with Socket 754 interface, 90nm manufacturing technology, voltage of about 1.4V, 200MHz external frequency, 128K or 256K secondary cache. Palermo kernel comes from Wincheste kernel of K8, but it is 32-bit. In addition to having the same internal architecture as AMD's high-end processors, it also has EVP, cool' n' quiet and HyperTransport, which brings users more "cool" and higher computing power. Because it was born out of ATHLON64 processor, Palermo also has a memory control unit. The main advantage of CPU's built-in memory controller is that the memory controller can run at CPU frequency, and the delay is smaller than that of the traditional memory controller located in North Bridge.