Our CPU Hu Weiwu I participated in the CPU development project of the calculation site, originated from an accidental opportunity in October 2000. In mid-October, the leaders sent me to "my alma mater" University of Science and Technology to conduct enrollment propaganda. This is the first time I graduated in 1991 and returned to my alma mater. I returned The laboratory I have worked for ten years ago, I have done a processor compatible with 8086 instructions with another classmate. Graduation Project. This is a circuit that uses more than 400 74LS series chips, which can run 8086 instruction systems in addition to all instructions other than decimal and division instructions. Because there is no version of the plate, it is handwind. I went back this time, I saw that I was still lying quietly. In the face of the laboratory, a piece of chip, capacitance, electric resistance, and electric iron, a tentacle, a tentacle, and a tentacle, and the impulse. , Because of the days before the night, there is no day, no night, the day, the trigger, the decoder, the selector is a deep temptation, and I can also say if I happen. A lot of integrated circuit chips currently have used . I thought of our CPU design project, so I called my brother Tang Zhi Min, he was calculating the chamber of the system chamber of the system, and is currently responsible for calculating the preparation of the CPU design project. the work. I am joking, I don't put the general operating system Boot, and I will see it in two years. So after returning to the calculation, I started to consider the Question of the CPU's instructions Uniform and pipelines. August 19, 2001, the 10th anniversary of the former Soviet Union, the godson CPU we designed successfully took Linux system boot. When the prompt of "login:" appears on the screen, a cheering room of the 309 room 309 is calculated. to September, a complete computer system with our own CPU has already floated the surface, the system runs full Linux operating system, the kernel version is 2.4, Make other computers that run the Linux operating system, including things that run the GCC compiler, X-Window window system, web server, and SPEC CPU2000 baselines. We are most proud or the system structure design of the CPU. It can be said that the world's most advanced CPU system architecture is available, and there is a lot of godson, and there are also innovations. Although we are currently based on FPGA design, the main frequency is not high, but when our CPU runs to 12.5MHz, its performance is no longer more than 50MHz frequency Intel 486 (of course , our motherboard is better than 486 motherboards), exactly, floating point performance is more than 486, and the fixed-point performance is 486 poor. Value It is, when we use a test program called "Paranoia" test to test whether the CPU's floating-point component is in the character, when you use the IEEE 754 standard, The IV processor detects that floating point is not strict in accordance with IEEE 754 standard and our CPU is fully compliant standard. In addition, in Godson, it is also necessary to safety design for network attacks, which can effectively prevent attacks using buffers overflow technology.
It should be pointed out that we only completed a logical design of a CPU, which is currently using FPGA to check this logic design.! ----> card, just a phased result, not yet Take a tick. Said that Li Shou said, "Two-third of the work is still later . If we are complacent about the current work, it is very beautiful. But even if this is achieved, it is not easy to come. Among the taste, sour and bitter, it is difficult to understand the outsiders. Review We develop the godson processor's passing We do the direct promotion of the Director Li Guojie from the director of the CPU design. Teacher Li is a more admirable in the academician I have touched, because he can stand in how to develop the perspective of the information industry throughout the country to consider the problem, not a partial Angle. Now I slowly know that he is not easy to drive our CPU design. Perhaps due to the fact that the repeated impression of the previous year, it has left a bad impression, making many people feel that there is no ability to make CPUs, and Teacher Li finally "! ----> only Can first do the CPU design as a calculation office. I am in calculating the reading of the student for ten years, and the first experience will start to do a topic without the application. CPU design technology is the core technology, but the market wall is high, even if it is now tick up, it is very good CPU. If no one will use it, it will go into the previous "Identification will be a memorial"! -> 会 "'s strange circle. But I believe that Teacher Li's experience and influence in the information industry, so determined to do it. My brother Tang Zhimin is the director of the system structure room, and he grasps the overall situation of the entire work, leading the entire CPU design. His massive and modest and can unite a group of very dry young people, making everyone from each other.! ----> guess. Now there is a strange phenomenon in the scientific industry, just after a young man makes some results, I like independence portal, resulting in scientific research power, not a big event. Everyone is in touch with some hundreds of thousands of items, and it is a hundred million project to form a lot of power. In our project group, there is a batch of itself! ----> harm, you can unite together, work together to do something, an important reason is Tang Zhi Min is a leader that can be able to hold people. With a simple example, in the project, a CPU design, I am a project leader, but includes the project application, a monthly progress status and expenditure status, And the material, etc., I have never been written in a word, all Tang Zhi Min, make me more than 90% of the time can be used in programming and logic design. This is just a small thing, but I am calculating the ten years, I have seen a lot of things to help the leadership, but I have never seen the leadership. Tang Zhimin has unique insights in global grasp and overall planning, At least what I can't.
Regarding our future CPU, my claim is to do high performance workstations, but do what to do, how to compete with others, but Tang Zhi Min is very much, he said that we now have the software room Being an e-government, it is incorporated herein, at least in terms of safety, foreign products cannot compete. I felt this at the time, I really became "forward". Zhang Zhi Min is responsible for engineering management in our CPU design, he is a passenger researcher in Li. According to our division of labor, in the team of our design CPU, Tang Zhimin is always responsible. I am responsible for designing, Teacher Zhang is responsible for engineering management. Teacher Zhang is a very strong person, very loyalty, engineering experience is very rich. I admire the teacher's two points, First, he looks very lightly, he gave up the work of the original salary to do this here; 2! - -> It is very real, very special. I have a point of view. It is not easy for a person to do a good thing for a lifetime, so I hate some of the side here for a while, and I want to have good people in both sides. But Teacher Zhang is not such a person, He is very real. After we basically complete the logic design of Godson, you need an environment that simulates the motherboard so that Linux is started by software simulation in logic design. I think this is very difficult, but Teacher Zhang has written it to one or two stars. "---->. I am very good with Teacher Zhang, it is the best partner. His project experience is very rich, I theory Superior; do things, he is more stable, I am more embarramented. At the meeting of Monday morning, I lay down the work of this week. After the week, I always mobilized it, and Teacher Zhang always put some difficulty and splashing cold water. It is really a relaxation. It is not a hot water at all, and everyone is not used to it. Although the CPU design we have prepared has been previously (prepared in early 2000), many rounds are discussed. But we don't have the design of 2000 in November 2000. Just starting to be Tang Zhi Min and I have received seven, eight graduate students to do system structure gauge, mainly to determine instruction systems and pipeline structures. At the beginning of the design, Lee Director will set the goal of "high performance, general, one step" at all meetings. Later, Tang Zhimin and I have identified compatibility and using the RISC structure These two specific goals. At that time, although the IA-64 structure was fried, we decided to adopt the RISC structure, and now it seems , this is right. The instruction system I started suggesting that Alpha is compatible with Alpha, but Tang Zhimi is set to be compatible with MIPS. To be honest, from this time from November 2000 to April 2001, I am still not paying enough to CPU, because I originally do a cluster and a total of The system is also very interesting. During this time, I still spend some time in the same time in the work group, I wrote some papers, and I have opened a parallel processing international conference at the end of April. Beginning from the United States, starting with the United States, starting from the United States.
Due to time, when determining the pipeline structure, there is no way to do more the texts, mainly by the feeling of feelings and the accumulation of the past, encountering the weighing and gaining, and discussed with Tang Zhi Min Let he will win. However, now go back to see, because there is no reference to others, there will be less imprisonment. Now, when you start watching others, we have found that our operational queue-based dynamic instruction line is also is very advanced, there are many innovations, I feel that it is stronger than the MIPS R10000. Our work is truly full to spread after May 1st. On the one hand, the system structure has been fundamentally, and the simulator written in C is almost almost, and simple instructions and short programs can be run. On the other hand, in June, I suddenly received the "Notice of the 45th Anniversary of the 45th Anniversary of October this year to show the results of my CPU design. Since June, we will open the night in succession. During this time, the team has expanded rapidly, and there are already three or more people. Although just a logical design, I can find a complete universal CPU within three months, I feel surprised. Now, there are three reasons for success. First, the technical route is correct, and the other is a high-quality team, three is a good environment and strong Technical reserves. There are two lessons. First, the engineering experience is insufficient, and the initial management is not good! ---->; I think our technical route is very successful in three aspects. One is the positioning of the director on high performance universal CPU. Not Those think that the maximum market in the processor is in embedded, and China's processor design should be positioned in embedded, not Necessary or no universal processor design. It is true that the demand for embedded processors is very large, but the big market is not necessarily "! ----> Make money, the small market sometimes makes big money, because the latter is the core technology, others can't do it. Intel is a typical , the number of Intel processors only accounts for 1.5% of the global market. Second, China is not a small country, the core technology is not Master. I feel that something like embedded processors can do, because others' embedded processors are very low, and they will not k. Some foreign companies even actively put forward, as long as we use their production line, they can provide embedded IP cores for free. As for someone thinks we can't do a high-performance universal processor, my answer is to make a dawning machine with the "our own processor. Although we are still very backward on the high performance universal processor, it is much better than the current conditions for the current conditions, more than 8920, which is much better than that year, and the workload is much smaller. "---->. Anything, the key is that some people are quietly doing seriously. We are compatible with the second success on the technical route. Review the history of China's computer development, it should be said that our The ancestors are also very "broad", until the early 1980s, our processor design technology is not much difference than others. Falling Today this is a large extent to eat incompatible losses.
Because most of the costs in the computer are not spending on the software, but flowers on software. It is very troublesome to design a command system, which can only figure it out; it is still very troublesome with other mainstream instruction systems. Sometimes it is necessary to modify the data path for an instruction, but it can be a lot of things. Yong! ----> Yi. In our group, there is a very well-dried operating system group, only four people, to get everything in 3 months related to software, including BIOS, Linux 2.4 kernel, GCC compiler, x -Window, debugging tools, Web Server, etc., it is to benefit from our compatible design. Of course, these four people are absolute Linux masters. We even be compatible with the processor pin level of the person. As long as you pull someone else, you can insert it. Therefore, the motherboard is now . Of course, we are now doing the motherboard design, because we want to do infinite band in the future, do not master the motherboard technology. Our third success on the technical route is the design method of steady stabilization. That is, the design of the system structure, to the C simulator to Verilog emulation, to FPGA verification, to standard unit punch, and then to a full-specific ticklay method. After determining the system set , we designed a Cycle-By-Cycle C language simulator, which describes the system structure of the godson processor It is possible to simulate the change in each bit of the processor. The design simulator has three obvious benefits. First, verify the correctness of the design, we have running Linux 2.2 on the C simulator, linux2.4, gcc and other one The series system software discovered a lot of errors in the design process. The second is to use the C language to describe the system The structure is more stringent, no erliness, more clear than the document written by text. After the C language simulator is verified correctly, the C language simulator is converted into a module that is less than a month of the Verilog language. And due to the design of the design in the design, the error is not appeared in the Verilog description phase. The third benefit is that the C language simulator provides a development platform for software development to develop software such as BIOS and other software. After converting the C language simulator into a Verilog design, we have successfully run the Linux operating system on the "running platform" of! ----> Verilog running platform. Then synthesize and form a firing document of the FPGA. After the above repeated verification, it is basically successful when the joint is adjusted. If we are not steady, but take a leapfrog-based Direct logic design, I am afraid that the speed will not be reached, because the more the upper layer is designed, the more easily debugged, and " The more the bottom floor design, the more difficult it is, and we even then even the Verilog language will not be used. After the joint adjustment is successful , we have made a larger optimization for the pipeline, and each time it is verified on the C language simulator. On our laboratory wall, there are two slogans, one sentence "life can have a few pumps", and the other is "truthful, truth"! ---->, truth, innovation " .
The first sentence slogan is the slogan used by the dawn series of high-performance computers, I borrowed it . Indeed, in the three months we work hard, the members of our topic group have paid hard to imagine, especially in the stage of debugging. Many people have the experience of the debugging process, but few people have debugged the operating system, and in one of the possible processors that may be erroneous. Our challenge is here, when there is a "application, operating system, and processor itself, it is doubtful objects, requires many coordination and analysis of"! ---->. We used to run the Linux operating system in the C simulator, Verilog analog environment, and FPGA verification system, each of which is a few nights for a few days. Especially the last one is the most tragic, because even if you find a small error, modify a new FPGA boiled file, you need for at least 8 hours. Only one day 24 hours a day, can guarantee a chance to modify several design a day. I am not sleeping in the topic group, but there is also a week.! ----> There is no experience in sleep. In mid-August, I learned that I had to come to where I had to come to the inspection of the next day before work, so I decided to take the lead in advance of the original joint adjustment time. I hope to succeed. . But I am afraid that I don't have a successful usage. - I only find a few backbone on Monday night. We only have two opportunities, because modifying a design is 8 hours. At 4 o'clock on Tuesday, write the FPGA file to FPGA, without any moving. The problem is now using a logic analyzer. It turned out to be defective with the motherboard handshake when starting, causing the motherboard to have not withdraw the reset signal. Hurry and form a new FPGA firing document at 12 noon, there is still no movement after writing. In the afternoon, the leader came to , opened an afternoon, then debugged in the evening, found two solder joints on the CPU card, remove the liquid on the motherboard crystal display As shown in the words "godson", we cheered. We decided to play continuously, to Wednesday Successfully run the BIOS system on the transformed motherboard, which is equivalent to a simple operating system , another cheer (At this time, other members of the topic group know that we are running procedures on our own CPU). At 12 o'clock, everyone is required to go home after going back to sleep. At night, thunder, wind and rain, to create a celebration. I am excited to have "difficult to fall asleep, because I will always compress my pressure slightly easely in the three months. After tomorrow, start the Linux operating system after work, but every time you start entering the user state to start a variety of applications. Doubt is a problem with TLB , because accessing user space begins using TLB. I have been fighting TLB for three days and two nights, and there are a lot of "----> questions in the middle, I always got disappointment every time, I'm always disappointed until I have a good value before Saturday.! ->, have been revised after dinner, 2 o'clock in the morning of August 19th, forming the FPGA file and write FPGA, 2:42, the screen has appeared "login" on the screen, login Go in place to play, just like Linux on other machines.
I used VI to edit this historical moment. After the deployment exits, I will send it ftp to another machine to Tang Zhi Min and the you've! ----> guide, and Excited, I immediately called Tang Zhi Min to tell him this good news. At that time, although we have become extremely exhausted, 6 people present were excited to have no sleep, chatted to dawn. I remember that night is also the wind and rain. At 6 o'clock in the morning, go home Sleep, I don't know how others, I slept for more than 20 hours, I added this week. After that, we struggled two worships with the aforementioned "biased" floating point test procedures, but it was not so hard. I sometimes feel that I am more embarrassed than Zhou, but the members of our topic are indeed very playing. Have a few times, I opened the lab in the morning, I opened the laboratory in the morning, and I found that some people helped the mouse to fall asleep on the chair. I am easy to receive by "! ---->" I can't help tears like this scene. But I still wake them up to ask them last night and let go. I have a student. I rarely sleep in the morning in the past two months, and I often call him on the 8th and 9 o'clock. I will call him after the class, because I am eager to understand yesterday. Evening progress. I remember once we were in the middle of the night. ------------------------------------------------------------------------------------------------------ He said that he did not want to die, and looked at his life. I have a point of view, we are now behind, others are not stupid than us, if everyone is only one day and five days, 8 small When you work, I am afraid it is difficult to catch people, Only like it is desperately desperate as "two bombs,", at least a group of people. Only in this way, we can not be bullied, our children will have hope to re-act "to commit me a big man, although far now.! ---->". I often say that a pot of water can be bloomed, but the water is more vivid with cardiopathy. Our CPU business is a lot of money, we are driving her with heart blood. We gave us a very traditional characterful little name called " dog left", hope the name is a little easy to raise big, transliterate into English is godson. As for another slogan of our wall "truthful, truthful, truthful, and innovation" is an annotation and playback and playback of the trained "truthful, innovation ". Because I feel that the Chinese science and technology community is too impetuous, there are few people who do things. Because of the I hope that people in our group do work hard and make less. The first truth is about the purpose of doing learning, To learn, do not learn, not for the name and profit. To really do 'people don't know what to do. This point is easy to make it easy to do it. As we work in calculation, it is probably not very much, as long as the days will pass, but it is hard to see, this is hard to see, this, I don't do it I don't do it. At, but heart is directed. The second The attitude of doing learning is to practice the work, it is to work hard, just don't write an article for "! ---->]. In scientific research, it is a little innovation, it requires a lot of practical work. This is an objective law , no one hid. We often encounter some people who don't want to do specific and cumbersome work in scientific research.
This phenomenon is very common, I do this, but I will force it. The third truth is about the way to learn, to do it Be courage to practice, do not talk on the paper. It is to "learn from time to time", it is a constant loop reciprocating of understanding and practice. Learn Ask is not just knowledgeable, more experience and feelings. It is difficult to find problems, it is difficult to innovate. And these experiences, others can't see it, there is no big amount of physical strength. ----> As long as you have achieved these three truths, I don't want to innovate. Nowadays, many people think that my country has no hand, mainly because of the process level, therefore, the processor design Do not be a calculation office, but do The work of people with microelectronics design. It is true that reviewing our country's processor design An important reason for comprehensive backward after the 1980s is that there is no manufacturing process, causing the processor design team overall Drop, the person is lost. But now there is a change, there is no big problem in the tickle, mainly the design is not designed. The processor design is not just a logic design. The processor design is a large system, only the entire computer, including the operating system, compiles, including, I / O, etc., can design high performance processors. Just as a "fixed asset and funds only provide the necessary conditions, the key is also management. In a general purpose computer, the cooperation of the operating system is particularly important, and some instructions are specifically designed for operating systems, especially in the system. , Atomic operation, exception processing, etc. I am using our own structure to implement the MIPS instruction system , there is a deep understanding that a command system has been developing in more than a decade (" The MIPS instruction system experiences the development of MIPS I, MIPS II, MIPS III, MIPS IV). Sometimes, by analyzing the analysis of the "operating system", you will find the original intention of the designer to set these instructions, it is really to shoot . Therefore, I have been very glad that we use a compatible practice. If you design a command system, there is no more than a dozen! ----> The practice of the year is not perfect. Looking at China in architecture, operating system, compilation of comprehensive power than calculations (I admire Other units only), The CPU design is very advantageous. Although the toss in the past few years has made the calculation of the yuan gas, but the foundation is still there. At the beginning of the CPU design, my complexity of this thing is not estimated enough (this is a "one" profound lesson), just leads eight, nine graduate students. Later, after the project was gradually launched, it was found that the existing humanity is more than the logic design. It is not enough to start the general operating system. In desperation, I had to make a few operating system masters from the group of the group and the group of embedded operating systems, and things have improved.
For example, we must run the operating system on the C simulator, you need to cut and modify the operating system kernel (if you remove some and motherboard configuration closely related), this It is not done without the in-depth grasp of the operating system. Later, I have been added from other places (I now understand why the commander is very important to the preparation of the preparation when the movie is fighting. Now, our group already has three, 50 people's scale, the soldiers are not more than the fine, these people are all young guns Hand, the quality is very high, and it is very fun. For the things I have done, the calculation is almost "what talents do"! ----> Talents ". Of course, in the future, the project will now be far less than enough. The scientific research environment of the calculation is very good, especially in logistics services. Take a few small examples. The first is that after the work is started The whole class is more than the overtime. We have a small bed on some tables in some tables. When Deng The secretary knows this situation, he immediately finds a small house, put six beds, greatly improved our overtime! ----> rest environment. The secretary also allows people to prepare cookies, instant noodles, etc. every day. The second example is that the work is continuously in-depth, and the group needs to continue to add new microcomputers, but the machine is with you (because you didn't have "! ----> I don't know how many machines want to buy). In this regard, the comrades of the business office always cooperate, and ensure that the group is in time in time. ----> machine. The fastest time, I raised the purchase application to the business office Wang Yugjie, at 11:30 am, 1:30 in the afternoon, was sent to the laboratory. This is hard to imagine in a financial system, and the approval of strict institutions is hard to imagine. The third example is once we urgently need a Matrox G200 graphics card, but because of the product a few years ago, run all over the company. Wanner helpless, only to ask for help from some teachers in the inner teacher, enthusiastic business It immediately released this news on Tianbo, less than an hour, in the help of Wang Zongong, looking for it ! ----> to this graphics card. This reduces at least two weeks of work, because if we use other graphics cards, we have to have yourself Write drivers. Effective logistics protection makes me sometimes don't do things well, sorry these enthusiasts to help us . We have two lessons in the previous period. First, the engineering experience is insufficient, mainly in the way to do research, and the difficulty is expected to be estimated to task, and the research strength is not enough, resulting in low efficiency. Fortunately, the joining of Zhang Zhi Min, the promotion of Zhang Zhi Min, who has experienced the experience of the project, has changed our research, and the goal of "rear wall does not fall". . In addition, fortunately, the technical reserve is strong, who has anyone else, make it possible to improve efficiency in the later stage. However, we are currently in terms of document management. The second one is not enough to pay attention to tools. Now doing processor design, EDA tools are very important. We have a lot loss because we are not valid enough.! ----> EDA tools.
There is a wiring cloth that does not pass, as long as you modify the settings, it is a pity that we will not, and you will not be able to find a week, and then discovered. has a 50MHz 486 performance at 12.5 MHz. The Godson's pipeline structure is my own feel more proud. When designing, there is no reference to other processors, it is entirely a feeling. The godson pipeline includes a number of advanced technologies in the current processor design, such as pipeline dynamic scheduling, Tomasulo algorithm, register rename, guess measurement, exact exception processing, 64 Bit floating point arithmetic components, Cache technology, etc., and innovation in some respects. The two problems in the assembly line design are exceptional processing and related solutions. First, we closely couple the exception to the stream The water line together, in the design of the data path, in parallel, in the execution process, the result and Normal results do not distinguish between the exceptional results only in the instruction end phase. In addition, I "! ----" guess the transfer instructions as a special exception, using the elimination of exception processing and maintains the exact logic on site. This not only simplifies the design, but also allows the execution instructions to be executed as much as possible. Our pipeline is a dynamic pipeline, using a DE-COUPLED structure, control logic distribution among each module. In the control of the stream lines, "Let the pipeline flow" is our design concept. In the instruction pipeline, data related and control related to the pipeline waiting, if the following instructions use the result of the previous instruction or whether the instruction is executed by the previous direction Movement instructions successfully No decision. For the wait for data, our principles are trying to postpone until you have to wait. That is, the instructions are not prepared for the source data, and the instruction decoding and launching will not stop waiting for data (simple ) to stop waiting when decoding is not prepared, blocking The continued execution of the following instructions), is established to establish a data dependency, and the functional components are resolved by the listening result bus. Such data results come out, Waiting for its instructions to calculate, do not have to write back to the register and read, and wait for the previous command Waiting will not block the back Continue execution of the instructions. For controllation, we also guess the directive of the transfer branch first guess , waiting for the determination of the transfer target when it is necessary. If the transfer guess is successful, it is immediately forward. If the transfer guess is unsuccessful, the correct site will be restored to the correct site with the unsuccessful command to revoke the mechanism. In the design of the pipeline design, "simple, smooth" is the goal we pursue. The former guarantees the simplicity of the stream level, and the latter guarantees the high performance of the pipeline. . After our joint adjustment is successful, it has been 4 times a large number of pipelines, which is more simple and smooth. I think there will be several times before the formal voiced tablet. In addition, Godson's floating-point computing parts pipeline design also has its own characteristics.
Another feature of the godson architecture is flexible, modular, and can be arbitrarily cropped to meet different applications based on different needs. Since the Godson's pipeline is used in the DE-COPLED structure, the control logic is part in each module. Increase or decrease the function module has no impact on other parts. For some embedded applications , no floating point components, as long as the floating point ALU, floating point multiplication, and floating point register modules are removed, do not need to do anything for other modules! ----> modify. System security design is also an important feature of godson. In addition to the security mechanism required by the MIPS system requirements, the Godson system has implemented a new access restriction mechanism for network attacks. The buffer overflow is a very common, very dangerous "!" The vulnerability of the risk is the approach to most network attacks. In various operating systems, there is wide existence in applications. With buffer overflow attacks, it can lead to failure, system crash, restart and other consequences. More seriously, it can be used to perform unauthorized instructions, and can even obtain system privileges, which in turn makes various illegal operations. Godson effectively prevents the attack of the buffer overflow by allowing the operating system to limit the attack of the buffer overflow. Of course, Godson still has some defects in the current design, mainly for Cache. The current design is only the 4KB of data cache, all direct association (this is related to the FPGA capacity, the current design has been Used 1.5 million gates The 85% -90% area of FPGA). It is not ideal for programs that are high-visiting requirements when performing performance test. If Dang Godson is running at 12.5MHz, the performance of the Idt 64474 CPU (R4000 core) is 3-5 times the performance of GODSON, but is sensitive to godson. The performance of IDT 64474 is 6-8 times the godson. IDT 64474 has 16KB of instructions Cache and 16KB data Cache, all of which are associated with the Cache capacity and organizational mode than Godson. Be sure to improve the cache part before the tickle. About Godson's future development, we have a ambitious plan. The first step is to perform the standard unit's tickle after optimization, the main frequency is about 200-300MHz, and strive for 2002 to be completed. > Make station / thin client PC and e-government. The second step is to design a double launch 64-bit structure (currently singular, fixed point 32 bit, floating point 64-bit), the tickle frequency is about 500MHz, and strive for 2003-2004. Used for high performance servers, also can be used to make high performance computing in the unit. The third step is striving to break through the processor, and the large SMP node based on Crossbar , 16-64 CPUs per SMP node for high performance calculations.
