64-bit unstoppable] In the history of personal computers for more than 20 years, AMD seems to have never occupied the advantage of Intel, which seems to be attributed to its X86-64 series 64-bit CPU. We know that the X86 structure has entered the human world with the IBM PC-XT PC with the central processor of IBM PC-XT Personal Computer, which has a processor with excellent performance and low price at the time. It is a truly The 16-bit microprocessor (8086 has 16-bit registers, operational units, and 16-bit input and output bus, its twin brothers 8088 is identical to the 8086 core, but 8-bit input output is used for compatibility. Subsequent 80286 is also a 16-bit X86 structural processor with a 16-bit X86 command system (IS, ISTRCTION SET), but has a more advanced protection mode instruction set. Soon, the development of PC makes Intel feel that the CPU - 80386.80386 based on 32-bit x86 instruction set is a milestone, which is a heavyweight weapon in the world. The technology and market strategies of this processor are simple. For a time, the compatible processors of each household, including Ti, Cyrix, IBM, and AMD, I want to have a piece of tide that has been surging in PC. However, the winner is in all, in order to show his great victory, Intel calls the Intel System 32 (IA-32, Intel Architecture-32) based on the 32-bit X86 instruction system (IA-32, Intel Architecture-32). From this time after more than 10 years, 80486,80586 (Pentium, this word is the "fifth generation" meaning), 80686 (Pentium2), Pentium3 until Pentium4, Intel has always firmly occupying the core position of the personal computer development. The 32-bit X86 architecture has also experienced more than 10 years of wind and rain. In the past 20 years, X86 has become the standard of computing platform with unparalleled performance prices. More than 100 million installed volumes per year, more than 90% market share were widely recognized by users. At the same time, the benign competitive environment of the open platform promotes the technology development of the X86 sent far more than the RISC system. The X86 product has already entered the workstation from the initial PC, and the server field, and the RISC has been replaced, and the situation is calculated in the world. In this process, it hinders X86 to enter the high-end corporate market, that is, X86 is still based on 32-bit technology. There is no power for high-end enterprise servers and workstation applications. With the development of business computing applications, 64-bit applications will become more extensive, and the X86 is imperative to 64-bit expansion, and it has become a hope of unifying 64-bit computing standards. At the same time, individual, business and mobile computing areas, 32-bit calculations have gradually begun to seek strength, regardless of technology or competitive needs, 32-bit era, has gradually begun.
During the X86 from 32-bit to 64-bit expansion, Intel and AMD have generated significant differences in the fundamental aspect of the instruction system. We know that from 80386, the Decision of the Instruction System has been in the intel, and AMD can only produce so-called "compatible" processor, that is, the standard is Intel, and AMD is only relying on intersection with Intel, relying on himself The diligence efforts have a place in the processor market. In the corporate market, traditional RISC vendors and UNIX world, Intel wants to play in and get the same success on personal computers, it is just what it dreamed of. Because the name of the AMD64-bit processor is too big, everyone seems to feel that AMD is 64-bit calculated leaders. In fact, Intel enters the 64-bit processor than AMD, but its strategy has different. Intel does not think that the personal and mobile field requires a 64-bit architecture, and at the same time, I feel that the X86 is extended to 64-bit to the 64-bit manufacturer's RISC system to be weak. Moreover, Intel wants to have a single show in the field of 64 enterprises, and I don't want to involve these tails of AMD, form a situation in the 32 market, so INTEL separately released the 64-bit market. The IA-64 architecture, and related 64-bit instruction specifications are the famous EPIC (Explicitly Parallel Instruction Computing, Explicit Parallel Instructions), and published an enterprise-class 64-bit processor named ITANTIUM. Ante is indeed a high-performance processor, but its IA-64 is not compatible with the X86-32 instruction set, that is, the personal computer application cannot run on the Angar, so that the public is actually drawn by Intel. Down 64-bit.
[64-bit road of AMD] If there is such a great opportunity, AMD wants to jump out of its research and development processor architecture and to fight against Intel, it is very courageous. But the opportunity still appeared, AMD decided to develop its own 64-bit processor architecture and named x86-64. From this name, we can see that AMD's 64-bit instruction system is expanded from X86, and AMD makes new 64-bit processor compatible with the past 32-bit instruction set, that is, AMD's new processor Not only a 64-bit processor, but also 32-bit. In the AMD64 structure, a new so-called "long mode" appears. The role is to set the CPU to run in a 64-bit manner. The naming of the long mode may be "real mode" and "32-bit mode" familiar with everyone. Long mode refers not to have 64-bit performance, but also the register can also be extended to 64 bits and add new registers. The long mode is activated by a control bit, which is called LMA (Long Mode Active). When the LMA is not activated, the processor runs in a standard X86 mode, which can run 16-bit or 32-bit OS and applications, and when the LMA is activated, the 64-bit extension method can be executed, at this time New 64-bit CPU. The long mode is also divided into two sub-models, namely "64-bit mode" and "compatible mode". These two sub-patterns are controlled by D bit and L bit in the CS (Code Segment) register. "Compatible mode" means running 16-bit or 32-bit programs in 64-bit mode, similar to virtual 86 mode in the 386 processor, the old X86 mode (32-bit or 16) is called "legal mode (Legal) ". When entering 64-bit long mode (LMA = 1, cs.l = 1 and cs.d = 0), the standard size of the operand is still 32, and the addressing is 64. The performance of the long mode can be summarized as: a virtual 64-bit addressing, register extends to 64-bit, additional 8 registers (R8-R15), additional 8 SIMD registers (XMM8-XMM15), 64-bit instruction pointers, FLAT addressing mode. Among them, additional new SIMD registers reach 16 of the total number of multimedia registers to compensate for the weak links in the X86 structure. In order to make the increased register more logical, AMD takes a way to have a 16-bit and 32-bit register. This approach increases the clock speed to make CPU performance better. Compared to the 32-bit processor, the core of the 64-bit processor is improved. The difference between the two has the following: First class cache maintains the original 128KB, where 64KB is the instruction cache, 64KB is a data cache. Depending on the Architecture of the AMD64 processor, the addressing capability of the secondary cache allows the capacity of the secondary cache between 1MB to 8MB. But although Athlon can support 8MB of secondary cache, but in fact AMD has never done this. The server market is an AMD64 processor's primary target, so it is necessary to have a secondary cache greater than 1MB. In addition, AMD will use a three-level cache. The pipeline length of the AMD64 processor adds two levels than Athlon, which allows it to run on a higher frequency. The AMD64 processor has improved on the branch prediction unit. The AMD64 processor supports a larger translational backup buffer (TLB). The following is a simple summary.
[Review Classic] The primary goal of AMD 64-bit processor architecture design is to provide a new generation of performance. The most obvious example of this design idea is that the Opteron processor changes the assembly of the previous generation processor. The front-end command acquisition and decoding logic identifier is reduced and the decoder is provided from the decoder to the execution pipe scheduling program. In order to adapt to this change, the design redefines the pipeline classification to maintain a high frequency measurable. The result is more than two pipelines than the seventh micro architecture. The final product has 12 integer computing line and 17 floating point operations. assembly line. While prolonging the pipeline, the Opteron processor will initially produce 0.13 microns of SOI (Silicon On Insulator, Insulating Silicon) process in consideration of the scalability of the processor architecture. With the use of less than 0.10 microns, the frequency will also achieve corresponding improvement. The key factor in the ultimate determination of Opteron performance is that the OPteron microarchitecture has a higher IPC (number per clock execution instruction) value than the previous microarchitecture. At the same time, AMD's 64-bit processor also integrates a memory controller. This is traditionally completed by the North Bridge, which means that the traditional upper North bridge is the "transfer station" between the processor and memory, and the North Bridge drives memory through the memory bus. Reading and writing, then the processor bus is also forwarding the data to the CPU by the processor bus (FSB, Front Side Bus). This mechanism will inevitably bring efficiency and low bandwidth limit while enhancing processor compatibility and reduced design difficulty. In terms of processor micro-architecture performance, how to reduce access delays while providing a sufficient memory bandwidth of the processor, becoming one of the largest bottlenecks of performance improvement. The work drives memory in the new processor in AMD is directly taken by the processor, that is, this processor no longer has the concept of conventional pre-test bus, of course, there is no bandwidth limit. The Opteron microarching integrates a dual-channel DDR controller that has a 128-bit interface capable of supporting up to 8 DDR DIMMs (4 per channel). The initial design of the memory controller uses non-buffered or registered DIMMs to support PC1600, PC2100, and PC2700 DDR memory. Take PC2700 as an example, the effective bandwidth of the processor can be as high as 5.3 Gb / s under this condition. This direct connection can significantly reduce the memory delay of the processor, and further decrease with the processor frequency boost memory delay. In addition, hardware and software are prefetched with higher bandwidth utilization, thereby achieving the goal of further reducing the effective memory delay of the processor. The memory controller integrated within the Opteron processor has a more amazing performance in the multiprocessor system. Using a parallel connection method, each has its own effective memory bandwidth, with the increment of the number of processors, making system performance have good scalability. In a multi-processing system of a 4-way processor, the system can support up to 32 DIMMs, which can provide up to 21.3Gb / s system effective bandwidth when using the PC2700 memory, is really amazing! ! At the same time, there is also a Ding Hypertransport (Ultra Transmission) technology, Hypertranport technology provides high-performance data transfer schemes for embedded applications. With HyperTransport technology, the computer internal chip (communication between networks and communication devices) can reach 40 times the existing technical standard. And Hypertransport's special design can also be applied to 0.13 micron chip technology. The purpose of Hypertransport is not to replace other I / O technologies. It only provides a high standard-based internal connection criteria to meet the memory requirements of the memory and the original I / O original, and can be used to connect the traditional Low speed I / O equipment and the latest high speed I / O medium.
Like a method of integrating the memory controller into the processor to enhance the memory bandwidth, the Hypertransport interconnect controller is also integrated into the Opteron processor, providing a strong extension between the processor and the I / O subsystem. Data connection bandwidth. The data channel inside the Opteron processor is a bidirectional 16-bit communication, which can reach the work efficiency of 1600mt / s (one million transmitted per second), which provides a bandwidth of 6.4 Gb / s. In the Socket 754/940, the Hypertransport frequency supported by the AMD 64-bit series processor is only 800MHz, and on the latest Socket 939 series processor, the supported Hypertransport frequency has reached 1000MHz, which is why high-end Socket 939 processor The reason why Socket 940 processor performance than the same specification has a certain degree of improvement. In this new architecture, the frequency of the Hypertransport bus is usually several times the frequency of the traditional memory bus, so the bandwidth that can be provided is very high. In this way, the Athlon 64 processor has gone from the lack of natural bandwidth, which can help further improve the overall system. The Hypertransport frequency is available from the lowest 600MHz to the highest 1000MHz, which is available up to 8GB / s bandwidth. For large working loads, TLB subsystems with enhanced branch prediction, Opteron micro-architecture TLB and previous generation AMD processor micro-architecture, except for larger TLB entrance size, and less In addition to the delay, there is a multi-process filter without software intervention to share TLB. The enhancement of the branch prediction function of the Opteron processor enhances performance, which is especially obvious under larger workload. It is mainly achieved by increasing the number of bids in the global historical counter to 16K (this is equivalent to 4 times the seventh-generation micro-architecture). In terms of enhanced 32-bit X86-32 instruction sets, the Opteron processor micro-architecture supports all 32-bit X86 standard architectural enhancements, including Intel MMX and AMD's professional 3DNOW! Technology (integrated 3DNOW! Technology and SSE ). In addition, the AMD Opteron processor also introduces support for SSE2 instructions.
[64-bit processor launch] However, the development process of AMD's new generation processor is not smooth, and the processor's repeated extension launch not only makes the chipset vendor into the stabilization of the two difficulties, but also triggered various species of the media - Is it AMD? Do you really want to fall? As Athlonx is gradually taking late, the improved Barton core is quite in front of Pentium4, and the huge loss of AMD has exposed this problem in recent quarters. AMD you also know that the new generation of processors is getting better, the better, only this is likely to reverse the current dilemma, but as a product containing a variety of innovative technologies, the problem in the development process will naturally More. Hammer repeated reasons are not in 32-bit / 64-bit instruction compatibility and performance issues, but long-term hindering the manufacturing process of AMD processor frequency improvement - mainly the application of insulating silicon (SOI) technology, so that a long section Time AMD can only take out 800MHz samples as a display and media test. At the beginning of 2003, the technology cooperation with IBM solved the urgency of AMD in SOI, and the frequency of Opteron and Athlon 64 processors were also improved to a relatively reasonable level. On the other hand, since the new processor integrates the memory controller, the memory standard has been upgraded during the research and development process, and the dual-channel memory technology has become mainstream, and the AMD has to temporarily modify the memory control of the new processor. The architecture, which also added difficulty to Athlon 64 on time. But no matter how, AMD is moving, it released the Opteron processor (called K8 when research and development, the core code Hammer hammer) quickly was accepted by the market, within one year, AMD launched three series 64-bit Processor: Opteron Series for Server and Workstations, this series of processors integrated dual-channel DDR controllers, only supporting memory with ECC checks; Athlon FX Series (Dual Channel DDR Controller, Support for Euclings and Top Personal Computer) Memory with or without ECC checks) and Athlon 64 series facing high performance, low-priced desktop (single channel DDR controller, support for memory without ECC check), and is launching Moblie Athlon 64 for notebook markets The series, such as the robbing of the wolf, 64-bit, calculated to enter the generous return of the individual computer field.
[Confusion of Interface] AMD's 64-bit processor has three series, or divided into three grades in accordance with different market needs. Interestingly, AMD gave up on the 64-bit processor's needle on Socket462 from a good life style, big player, Intel, is good at the forces, in just a year, Three needles for 64-bit processors have been introduced. When I was just released, for high-end Opteron is a specification of Socket940, there are more than 940 pins, more than Socket 462 more than Socket 462, more than Socket 4262, the general processor's pin is arranged around, and there is a space in the middle, but we are in Socket940 I saw on the CPU, because the needle is too much, the hand of the processor has been densely numb. For low-end Athlon 64, the 754 pin foot of Socket754 was originally used, and later there was a specifications of Socket939, and this specification processor, not only Athlon 64, but also Athlon FX, which is objectively consumer The processor for choosing AMD has caused difficulties. Socket 940's number of pins and Socket 939 although there is only one difference, it has a relatively different arrangements in the pin position, which means that the Socket 940 interface is not added by the Socket 939 interface in a certain position. Come, these two interfaces have no way to be compatible with each other. From the blueprint developed from AMD, we can probably predict that the Socket940 interface will be left to the Opteron series (dual channel ECC memory), the existing 940 interface Athlon FX will slowly disappear, the Athlon FX series released later (dual-channel ordinary memory) The plenary is a 939 interface, and the processor of the 754 interface is Athlon 64 (single-channel ordinary memory) until the first 754 interface will be eliminated by AMD, and Athlon 64 may also completed historical mission. There are also some exceptions, and existing Athlon FX processors have some 940 interfaces. At the same time, high-frequency Athlon 64 processors also have 939 interfaces. [Different dual cores] Operton is designed to serve the server and workstation market, so multiprocessor features become an essential elements. Operton is divided into three different series of 100, 200 and 800. The 100 series cannot be handled in parallel. The 200 series supports all the way and two parallel processing. The 800 series supports all the way, the two, four and eight-way parallelism, is the Operton series High-end models. In addition to parallel processing, AMD has a volk in the high-end Operton series, which is dual core capability, which is prepared for HyperThreading Hyper thread technology dealing with Intel. Intel currently added to the ultra-threaded technology is also a multi-core architecture concept. It is mainly to solve the problem of cache hits, but this technology does not have the same efficiency as the previous chip engineer is expected. Hyper-throwing technical principles are capable of establishing a multi-instruction execution unit in a single core processor, if one of the execution units have a buffer failure, other units may guarantee normal operation. But in actual use, it is found that two phenomena of two execution unit data conflicts often occurs, so there is no great improvement efficiency. AMD's dual-core design does not appear above, because the two processors are each equipped with cache, the focus of competition is mainly focused on data exchange with the main memory. The AMD is expected to show its dual-core processor sample at the end of 2004.
[Long March] Although the Athlon FX cannot support the double core, it is divided into different models in accordance with the speed of operation, the first to introduce Athlon FX-51, using the SOI 130nm process, the frequency is 2.2GHz, then Athlon FX-53, ATHLON FX-55, etc. 3600 , 3700 , 3800 ............ Wait. Corresponding to three different interfaces, AMD uses different naming methods in each series, I really don't know what they are doing. In the new 939-pin Athlon FX / Athlon 64 processor and the 940-pin Operton processor to be released in the future, AMD adds an important hardware-level virus protection. This feature is Intel, AMD and Microsoft in order to cope. Early virus software is particularly developed. We know that the buffer overflows the primary means of virus attacks, but the hate is that this means is widely known, but it is better to solve the problem. Forcing Microsoft can only get a patch, with this technology. All this will be fully improved. AMD is called "Execution Protection" (Intel's called method), which is a locking mechanism integrated within the CPU. When the operating system activates this function, the data of the CPU buffer will read only, and Can't be executed, which can effectively prevent the virus buffer overflow attack. The current CPU does not do this, can only be ravaged by the virus. After activating this feature, the intentions spending data security for data security will be greatly reduced, and the virus library update will no longer be so frequent, and the updated more frequent Windows patches, large parts are no longer necessary. The habit of operating the operating system rely on "eat patch" is also changed. Windows XP's Service Pack 2 supports and implements viral protection in the AMD 64 processor. How big is the specific role of this hardware anti-virus function, our evaluation room is planning a test. As for the research and development, it is a fairly complex issue. Overall, the AMD 64-bit processor is called K8 in the R & D plan (the last generation of K7), where Operton first generation core is called Sledgehammer (1MB level cache / core area 193 square millimeters), and the second generation core is more confusing The 100 series single circuit (not supported parallel processing) core is called Venus, 200 series core is called Troy, and the 800 series core is called Athens. By 2005, AMD will launch a third-generation Operton, a single Danmark core, two Italy core and eight roads Egypt (Egypt) core, pay attention, the third generation core does not have parallel, It is a double kernel. It's really a mess.
Athlon FX Series and Opteron have unclear blood relationships, first launched 940-pin and recently launched 939-pin Athlon FX is the Sledgehammer core (1MB secondary cache / core area 193 square millimeters), the next generation seems to development route Will be brought to Operton, launch San Diego, which is released in the first half of 2005, and the fourth generation will become a double kernel TOLEDO, 2005 released. As for the Athlon 64 series, the first generation core is called Clawhammer (1MB level 2 cache / core area 193 square millimeters), now the second-generation core Newcastle (512KB secondary cache / core area 144 square millimeters) has also been released, As for the next generation, the Winchester core will be used in accordance with AMD, which is a single core and cannot be handled in parallel. [Red flowers also have green leaves] Light has a CPU is meaningless, and AMD in the supporting chipset also has a lot of work. The chipset is extremely important for the AMD's Opteron and Athlon64, because for today's PC architecture, the key to decide the overall performance is not only the CPU, the chipset is also not ignored. Objectively, the design difficulty of chipset is under the processor, but there are many problems that need to be considered, but the AMD is obviously not too much energy in chipset design. Therefore, AMD needs VIA, SIS, ALI, and NVIDIA, which enables it to concentrate, and do a good job in the development of AMD64-bit processors. Whenever AMD introduces a processor, it will be equipped with the corresponding chipset, the AMD 640 chipset in the K6 period, ATHLON K7 period AMD 750/760 chipset, ATHLON MP period AMD 760MP / MPX chipset, These all followed the sweat work for the AMD processor station. Interestingly, once VIA, SIS, ALI, and NVIDIA launched a more mature chipset, AMD always works back, more playing the role of technical guidance, while winning adequate development time for third-party chipset manufacturers. The K8 chipset is no exception, and the AMD 8000 chipset is the first to launch, act as Opteron's open hand, and followed the VIA, SIS, ALI, and NVIDIA have followed up. VIA released the K8T400 while the AMD released Athlon 64 processor, and later released K8T800 and integrated chipset K8M800. In May 2004, in order to better support the processor of AMD just released Socket 939 interface, VIA launched The upgrade version of the K8T800 chipset K8T800 PRO chipset. K8T800 Pro made two important improvements relative to K8T800. The first is that the AGP and PCI clocks are independent relative to the Hypertransport bus, which means that the AGP bus frequency and PCI bus frequencies can be freely adjusted. It is an important improvement for overclocking players, it can make players Possible mining CPU's overclocking potential without affecting overclocking due to other devices of the system. The second important improvement of K8T800 PRO is to support the 1GHz Hypertransport bus. This means that the North Bridge chip is faster than the CPU dialogue. The K8T800 supports two-way 16-bit, 800MHz Hypertransport bus connection, so its theoretical peak transmission bandwidth is 6.4 Gb / s. K8T800 Pro supports two-way 16-bit HT bus connection, which means that the peak bandwidth of the CPU and the North Bridge is 8Gb / s at a 1 GHz bus connection.
Relative to K8T800, K8T800 PRO has the following improvements: First, the bus interface uses a more flexible asynchronous way, while the K8T800 only supports synchronous mode; the second is that the link between K8T800 PRO Nort-South bridge has adopted a new generation V- LINK technology, the maximum rate can reach 1066MB / s, one-fold higher than before, and another K8T800 Pro supports the latest Socket939 interface. It is also compatible with the Socket 754 interface and the Socket 940 interface. Of course, the final product supports which interface of which interface is determined by the CPU socket provided on the motherboard, although these motherboards may use the K8T800 PRO chipset. In general, K8T800 Pro will match VT8237 South Bridge chip, this chip provides 6-channel sound cards of AC'97 Version 2.2, two ATA133 IDE interfaces, two SATA150 interface, and provides RAID function for this interface, six BUS MASTER PCI interface, eight USB 2.0 / 1.1 interfaces, through the Super I / O chip of LPC IT8705F, flies, infrared, two serial ports, parallel ports, games / MIDI, and hardware monitoring interfaces. In addition, add the PS / 2 keyboard and mouse interface. K8T800PRO does not support PCI-E, so Veterus will also launch the K8T890 chip, in addition to supporting the 1GHz Hypertransport bus, while also supporting the AGP8X and PCI-Express 16x, allowing users who have not upgraded to the PCI-Express 16x display card can also continue to use Current AGP8X display card. In addition, integrated K8M890 will also be launched simultaneously.
[Red flowers also have green leaves, two] NVIDIA as an important partner of AMD, active in AMD 64-bit chipsets, mainly because ATI and Intel are very close. The NFORCE3 series chip is a 64-bit platform recommended by AMD. NVIDIA has introduced NFORCE3 Pro 150 and NForce3 Pro 250 chipset based on the Opteron platform (NForce3 150 is the first single-chip design 64-bit die-chip chip. This product is mainly due to the maximum performance of North Bridge chip performance. Factors - The memory controller has built in the Athlon 64 Series CPU, the concept of the North Bridge has also diminished. In addition, NForce3 integrates the AGP 8X controller, USB2.0 controller, provides two FAST ATA133 controllers and provided Support for RAID 0 mode disk arrays and support for PCI 2.3 specification.), The CRUSH K8 series chipset, including NVIDIA CRUSH K8S, NVIDIA CRUSH K8S, NVIDIA CRUSH K8 PRO, NVIDIA CRUSH K8 3GIO, NVIDIA CRUSH K8G3, etc. Where CRUSH 3GIO supports the PCI Express X16 graphics interface. In addition, it also supports four PCI Express X1 bus expansion slots, actually a PCI Express controller. CRUSH K8G3 is based on Crush 3GIO, which integrates the FX5600 level graphic core, which provides more ideal graphics performance. CRUSH 3GIO and CRUSH K8G3 will match the Crush K8-04 South Bridge, connect the South-North Bridge through the Hypertransport bus. CRUSH K8-04 South Bridge supports 4 Serial ATAs and 4 ATA 133 interfaces, support NV RAID capabilities, directly integrating Gigabit Ethernet and wireless network technology, and supports network firewalls. In addition, CRUSH K8-04 supports 24bit / 96kHz 7.1 channel Azalia sound efficiency, providing 8 USB 2.0 interfaces. ATI is active in Intel and AMD two platforms, based on the AMD64-bit platform-based chipset planning with RS480 integrated with Radeon 9600 graphics, supports PCI Express, and RX480, RX480 and RS480 without integrated graphics cores. You can match the SB400 or SB450 Nanqiao. The new product was sampled in the first quarter of 2004, and the second or third quarter began mass production. SIS has also launched SiS755 facing the mainstream high-performance platform in the AMD 64-bit platform, SiS755FX, high-end high-performance platform, and SIS760 integrated with graphics chips. Both the three chipsets are combined with the SiS964 South Bridge, 1GB of Multio's South-North Bridge bus, supporting AGP8X, supporting ATA133 and SATA150, supporting USB2.0, supporting HT bus frequencies in 800MHz, support single-channel memory (960 will also support dual channels RAM). Since the most important memory controller is integrated into the CPU inside, the SIS North Bridge reduces performance, it only needs to control the AGP and Hyper Transport and Mutiol. SIS has made a lot of preparations in this regard, on the one hand, through the widow technology (MUTIOL), the bus bandwidth between the North and South Bridge is enhanced by HyperStreaming Engine technology to increase data transmission efficiency. SIS will also introduce the SIS756 and SIS761 two chipsets in 2004, which supports Athlon 64 processor, which is increased from 800MHz to 1000MHz, and the one-way transmission rate is up to 2000MB / s. The SIS761 is integrated with DX9 graphics kernel.
The SIS756 is a next-generation product of the SiS755 series, which is equivalent to K8T890, supports the PCI Express X16 graphics interface. SIS756 bridge chip is SiS965 and SIS966, with ASHE technology to achieve higher disk performance, and in other functions, SIS756 chipsets are also weak: Serial ATA RAID, PCI Express X1 expansion slot, Gigabit Network, USB 2.0 and IEEE 1394 and other functions. The SIS761 is an integrated version of the SiS756, and the SIS760 upgrade version, the integrated graphics core can support the DirectX9 API as the SIS 662 of the Prescott platform, as for the 3D performance to be verified. About SIS also says more, SIS is traditionally a low-quality route, but now it seems that its AMD chipset is not a savage lamp, and its performance can basically include VIA with any strong opponent. NVIDIA, ATI Phase anti-balance, and feel a bit of a little over the momentum of the blue. Now, I think some mainstream motherboard manufacturers should put more attention to the best time in SIS, according to existing performance performance, plus SIS consistent low price, I think it is entirely possible to play a market through SIS.
[Jedi counterattack] 64-bit operating system, it seems that it is not a problem. There are many operating systems running on the PC. There are several kinds of windows. Linux is even more than that of Solaris, BSD 3 Bros., OS / 2, Darwin, QNX, SCO UNIX, etc. Wait, etc., it is simply a few counts. But the mainstream is only Windows and Linux. 64-bit Linux kernel has been released, SUSE and Redhat, etc. Linux have already launched their own 64-bit Linux products. WINDOWS, many people think that it does not build a tree in 64, here must be Microsoft. Microsoft launched a 64-bit version of Windows2000 Datacenter in 2001, which is an IA-64 architecture launched for Intel Angan servers. It cannot be running on a personal computer. In 2003, Windows Server 2003 has launched four versions. They are the professional version, the web server version, the enterprise version and the data center version; the enterprise version and the data center version have 32 digits and 64-bit versions running on the Angar, and Windows XP also has 64 Antan. Workstation version, it should be said that Microsoft is the pioneer (just as Intel) on 64-bit roads. In fact, Microsoft's Bill Gates is 64-bit calculated loyal supporters. He said that "64-bit has come" can be twice. Of course, Microsoft also saw that the Intel is more dangerous. In September 2003, Microsoft launched a 64-bit test version of Windows XP for AMD. In January 2004, the 64-bit test version of Windows Server 2003, at the time The statement was introduced in 2004 to launch a formal board, and now it seems postponed. Some people say that they are out before the end of the year, is dangerous, whether it is AMD or Microsoft, it does not mention it, saying that it will not come. As for this extension, there is no Intel's relationship. Intel is like this. Its Antan has no resistance, and finally released the 64-bit to strong processor that is compatible with X86-64's core called Nocona, staged a game Two 64-bit processors (to strong VS Anteng) the tragedy of the same door. The new to strong instruction sets and AMD 64 are almost exactly the same, called IA-32E at the 2004 IDF meeting, which is YamHill technology that was previously rumored. Like AMD X86-64, Intel's IA-32E technology is to achieve support for 64-bit addressing capabilities based on the original IA32 architecture. It is actually an enhancement design of the original IA32 processor architecture. It also takes into account the support of the 32bit or even 16bit code developed under the original IA32 architecture. The IA-32E implements REX Prefix by deleting the INC / DEC, providing the Directive Code Space of 64-bit mode, 64-bit Universal Register (GPR), and SSE / SSE-2 / SSE-3 Register (XMM) Add 8 groups. In addition, the new 64-bit flat addressing mode, interrupt priority control mechanism, and RIP-Relative addressing mode are all identical to AMD X86-64. IA-32E provides 64-bit modes and compatibility modes, 64-bit modes support 64-bit operating systems, applications, drivers, smoothing virtual addressing spaces (General-Purpose Registers, GPR), Compatible mode supports 64-bit operating systems and drivers, applications, and general registers support 32-bit environments, providing 4GB addressing space.
Compatibility mode will not support the IA-32 architectural virtual 8086 mode (Real Mode), and many DOS device drivers will not function. Taking the Microsoft operating system as an example, Windows Me and the above have removed the real mode program code, so it will not be tired, but MS-DOS, Windows 95/98 will be affected. From the above, IA-32E can be compatible with AMD X86-64. After all, AMD64 has been widely supported by the industry, which makes Intel become an inevitable choice using AMD X86-64 architecture. Since Intel has signed an interactive agreement with AMD in the past and AMD, Intel can use AMD X86-64 without delivering the fund. [Destiny] However, IA-32E and AMD X86-64 still have different situations, more obvious differences: AMD X86-64 has 3DNOW!, Intel is not; Intel has SSE-3, AMD X86-64 has not yet. However, these are small gaps, and both parties need to join these features and will not have any difficulties. Microsoft and Intel claim that the operating system will be compatible with each other, which is also the most welcome result of software vendors. This result is probably the only choice, because anyone can't withstand the price of the program for Intel and AMD, Microsoft said that there are already two 64-bit operating systems (really pitiful !!), no longer the third 64-bit architecture write operating system, even Microsoft, and other software vendors are not mentioned. 64-bit operating system, a special 64-bit driver, this is a poor, in fact, this is also the soft rib of AMD, its influence is not enough, everyone holds the attitude of watching, not much So the driver written for the 64-bit operating system is still not rich enough. Of course, some big factories, such as NVIDIA and ATI have released 64-bit drivers, to solve this problem, Microsoft's help for AMD is Off. I believe that most of the drivers have 64 times when I officially released 64-bit Windows, because Microsoft will take advantage of the competitiveness and compatibility when the new operating system is launched, will spend great effort to improve the scale of the WHQL library. Finally, and finally, it is possible to determine the AMD 64-bit processor fate. It is the 64-bit application. We know that the setbacks of Andung is mainly because there are too few procedures prepared for it. It is said that there are only a few hundred, and in the SUN Solaris The operating system (also 64-bit) running the program has been 10,000, which is millions of money on the personal computer. In fact, for users, 64-bit processors, 64-bit systems, 64-bit drivers, etc. are unasvisible, visible only 64-bit applications that use these operating environment, no applications, everything is no good. For the AMD 64 processor, it has a big advantage that not only 64-bit performance is strong, 32-bit performance is also very strong, just we have said, it is also a 32-bit processor. Therefore, at the present stage, everyone is used in a 32-bit application environment. In fact, there is no effort to play all its potential (this also illustrates its amazing strength). But this is not a long time, no 64-bit application environment, 64-bit processors are like leaving the water, there is no survival and development space.
[See who laughs in the end] So far, the number of software written for 64-bit is still very small. Here, this is to say this, this should be a 64-bit performance evaluation of 64-bit processors, please note that it is 64-bit performance, and the evaluation of AMD 64 is too much, but Tong is 32-bit processor test, using 32-bit system, driver, and test software, the measured is of course its performance performance in the 32-bit environment, that is, the register LMA = 0, the long mode is not activated, all The register works at 32-bit, and all 64-bit dedicated registers are unavailable. If you have seen a white paper, it will find that the processor and Athlon XP are almost exactly the same from the kernel. (Of course, Some lines, Hyper Transpot, memory direct throughput and other functions). So I didn't want to do this from the beginning, but I finally didn't make 64-bit tests, mainly because the 64-bit software is too small, the test is less, the game is almost no - but Not completely, Unreal Tourance 2004 has already released 64-bit demo, which is Linux version, Windows 64-bit games have not yet, many manufacturers say that dooms do not launch 64-bit version. As for running 32-bit test software on the 64-bit system and 32-bit games - because 64-bit Windows is backward compatible, you can run 32-bit software on a 64-bit operating system - there is no test meaning, all Software runs in a 64-bit environment, which will face the problem of decline in efficiency. The reason is very simple. The 64-bit system is not a 32-bit software design, and the 32-bit system is different (at that time the processor runs in 32-bit mode. It is a 32-bit CPU). At this time, the processor is running in a long mode, running with compatible 32 bits, and the efficiency is not high. In short, 32-bit system 32-bit tests are not the goals of this article, 64-bit systems 32-bit tests, efficiency will fall, and not the target, 64-bit system 64-bit test, currently, the time is not mature. AMD provides some data, indicating that the processor speed will rise sharply in the real 64-bit mode, which is 64-bit application prospects, Intel will not launch 64-bit desktop processor before 2005. This time for this time AMD is more important. In the 64-bit battle, AMD has led ahead, and can not achieve the final victory, depending on that the 64-bit application environment can become familiar, once matured, Jiangshan is iron. However, INTEL is only afraid that it will not be happy? Source: Drive Home http://hardware.mydrivers.com/pages/200408140012_88172.htm
