"What else affects the efficiency of CISC?" Li Yixuan looked at everyone with a smile: "The second reason that affects the efficiency of CISC is the law of 28......"

Speaking of this, Li Yixuan paused, "The law of 28, also known as the 80/20 law, also known as Balett's law, was discovered by the Italian economist Baledo in the late 19th and early 20th centuries. He believes that in any group of things, the most important is only a small part of it, about 20%, and the remaining 80%, although it is the majority, is secondary, hence the law of 28.

For example, 80% of the merchant's sales come from 20% of the goods, 80% of the business income is generated by 20% of the customers, in the sales company, 20% of the salesmen bring back 80% of the new business, 20% of the people have 80% of the sum of social wealth, 80% of the people have 20% of the sum of social wealth, that is, 80% of the income comes from 20% of the high-end customers.

For example, only 20% of our daily lives are spent doing something meaningful that can improve our quality of life, and the other 80% is a waste of time and energy.

This law still holds true in CPUs. A CPU is made up of logic control circuitry and execution source code. According to the 80/20 rule, we divide the source code into "cold code" and "hot code" - the former accounts for 80% of the total number of CISC instructions, and the latter only accounts for 20%. Correspondingly, cold code execution units occupy the vast majority of hardware resources, while highly active hot code execution units occupy much less hardware resources.

This is another reason why CISC is operating inefficiently.

The CISC we use today was not designed with the 80/20 law in mind, resulting in 80% of the tasks in the computer using only about 20% of the instructions, and the remaining 20% of the tasks have the opportunity to use the other 80% of the instructions. If we optimize the instruction system accordingly, we can fundamentally and quickly improve the execution efficiency of the processor.

The so-called RISC processor POWER (short for Performance Optimized with Enhanced RISC) introduced by IBM in 1975 is designed based on the 80/20 principle.

To be honest, POWER is not a RISC in the true sense, POWER is still CISC, but it is because IBM used the 80/20 principle to improve the running efficiency of the CPU by more than 20 times when designing POWER. ”

The students below were a sensation, they didn't expect that just changing the design idea would improve the CPU so much.

Yu Youcheng was also very surprised, he didn't expect Li Yixuan to know so much about IBM's POWER processor, you must know that he also slowly learned about it many years after the launch of POWER, which still used Intel's resources.

POWER is the standard CPU of American supercomputers, and there is no sale on the market at all, so how does this young man in front of him know about it? Someone told him that Yu Youcheng thought about it and felt that this was the most reasonable explanation.

Li Yixuan, who was only listening to the stage, continued: "Although POWER is not a real RISC processor, it points out the future development direction of CISC for mankind, and at the same time points out the direction of RISC, so the real RISC was born. ”

Li Yixuan held up a chip with an area of only 100 square millimeters and the size of a thumbnail and said: "Everyone, please see, this is the world's first real RISC processor designed by our Oranges, and it is currently used as the total control CPU of the hard disk. The application range of RISC is far wider than that of X86, from various supercomputers, top-level workstations, high-end servers, to all kinds of embedded devices, home game consoles, consumer electronics, and industrial control computers.

Of course, CISC is also widely used, in addition to the well-known personal computing, it can be used in multi-functional network servers, multi-function workstations, general-purpose CNC machine tools, graphics workstations, etc. in telecom and mobile companies. The advantages of RISC are about specialization, while CISC is about compatibility and versatility. Although RISC and CISC have crossed over and competed in some areas, they will both develop in their respective fields. Now that you've finished talking about popular science, do you have anything you want to ask?"

Everyone looked at each other, the amount of information received today was too large, and they needed time to digest it, and they couldn't ask questions for a while.

"May I ask Xuan Shao, do you think that maybe in the future, RISC and CISC will move towards unification?" It was Yu Youcheng who asked him, and he also benefited a lot today, unlike those rookies, Yu Youcheng quickly digested the views that Li Yixuan said. As the world's top semiconductor chip expert, he keenly found that Li Yixuan put forward those views that maybe one day RISC and CISC will be unified, so he raised the question.

Yu Youcheng's words surprised Li Yixuan, but if he thought about it carefully, he didn't think it was an accident. Because Yu Youcheng was the first to propose the unification of RISC and CISC in history, many semiconductor giants in later generations, including Intel, AMD, IBM, DEC, etc., have worked hard in this direction, especially Intel is the most persistent, so that the CPU launched by Intel later is less and less like a standard CISC, but more like a RISC with CISC characteristics.

Therefore, it is not surprising that Yu Youcheng was able to put forward this concept, but due to the influence of Li Yixuan, this idea was advanced.

Unfortunately, although this idea is great, it is a pity that due to the different length of the source code of the two, this idea was not realized in the end, and Yu Youcheng later gave up the idea he proposed.

Later, a scientist in Intel's laboratory in Israel proposed the idea of core modular CPU design under Yu Youcheng's thoughts.

What is kernel modularity is based on the principle of the law of eighty-two, whether it is RISC or CISC, both use the same kernel, and then put this kernel in different systems, so as to achieve the unity of RISC and CISC.

For example, if you agree that a core is placed in X86, then it is a CISC processor, if it is placed in the ARM architecture, it is a RISC processor, and if the core is placed in a different architecture, it is a processor with a different structure. Although it does not really unify RISC and CISC, it greatly reduces the difficulty of CPU and design and fabrication, and also enriches the product line of chip companies.

Kernel modularization has also become the mainstream guiding ideology of IC design companies in later generations, and at the same time, kernel modularization has also brought about the birth of another scientific research topic, that is, microarchitecture technology.

For example, we have Intel's famous Core CPU as an example, which puts RISC as the core in the X86 architecture, and then effectively connects RISC and X86 through microarchitecture technology to become the core CPU we are familiar with.

The principle of operation is that when the complex instruction set enters the CPU, it can be decomposed into countless short reduced instructions for the microarchitecture, and then handed over to the kernel for processing, which greatly improves the execution speed of the instructions, and at the same time can solve the problems of compatibility with previous CISC applications.

After speaking, Li Yixuan drew a concept map of the Core Standard Edition on the blackboard, and marked the functional modules on it one by one.

Yu Youcheng was shocked in his heart, he didn't expect that the CPU could be designed like this. A careful look at the Core CPU architecture diagram drawn on the blackboard shows that the traditional CISC processor can achieve several times the performance increase under the premise of the same transistor size. There is no doubt that this design is undoubtedly a relief for the CPU companies that are currently deeply trapped in CISC, and the CISC processor has seen a cross-generational progress.

The shock in the soul is indescribable, Yu Youcheng can't describe it, he can only give Li Yixuan a thumbs up, amazing!

However, he soon found that Li Yixuan's chip design ideas are very wild, and the design ideas of various colleges are there, but they do not belong to a certain school. For example, he pointed to the area above and said, "There is a multi-core idea that is very similar to that of the Massachusetts Institute of Technology in the United States, and the CISC you designed has too many RISC features." ”

Li Yixuan did not deny this. CPU multi-core is nothing new, scientists at the Massachusetts Institute of Technology in the United States came up with this idea in the sixties of the twentieth century, with the intention of improving the execution efficiency of CPUs.

However, Li Yixuan's design is different, MIT's design idea is to have two or more equivalent CPU cores, and obtain performance gains through parallel computing, which can be regarded as a horizontal dimension of equivalent design. Li Yixuan's dual-core or multi-core idea has obvious future DEC company's design ideas, which is regarded as a longitudinal dual-core concept, which defines 20% of commonly used instructions as "hot code", and the remaining 80% of instructions are not used so frequently and are defined as "cold code".

The corresponding CPU is also logically divided into two parts: one is the hot spot, which is only for the program that calls the hot code, and the other is the cold spot, which is responsible for performing 20% of the sub-common tasks.

Since the thermonuclear part performs 80% of the tasks, designers can design it to be more powerful and occupy more transistor resources.

The cold core part of the task is relatively simple, and there is no need to spend the same amount of effort on it. Theoretically, designers can spend 80% of their transistor resources on thermocores to perform tasks efficiently, while the remaining 20% of transistor resources can be used for cold cores that only complete 20% of tasks.

The core of the dual-core concept in the vertical dimension is that the status of hot core and cold core is not equal, and they cannot operate independently, which can only be said to be a two-part discrete logic in a CPU core. What it does is to improve the utilization of hardware resources of the CPU and achieve high performance with high execution efficiency, which is obviously more revolutionary than the "dual core" advocated by the current industry.

Intel's approach is to fuse the cold core and the hot core into an independent core, so that the cold core and the hot core can share a part of the components and logic gate control circuits, but they are still two independent parts of the cold and hot pipes, and they will be analyzed in advance when executing the instructions to determine whether they belong to "hot code" or "cold code".

If it is an active "hot code", it will be sent to the "hot pipe" logic for efficient processing, and if it is a "cold code", it will be sent to the "cold pipe" logic for processing, and the results will be summarized and output.

As for what Yu Youcheng said later, his CISC has obvious RISC design ideas, which is not a matter at all in Li Yixuan's opinion.

In the future, whether it is RISC or CISC, both sides are learning from each other in terms of design, and at the same time integrating the ideas of various schools, so the CPU cannot be simply divided into CISC or RISC in the future, they are all learning from each other, and they cannot be called pure RISC and CISC.

Taking X86, ARM and MIPS, the three mainstream architectures in the future as an example, in addition to MIPS still really maintaining a pure RISC design, Intel started from Core, and its X86 is developing more and more in the direction of RISC, while ARM, although the overall architecture still retains RISC, but adopts more CISC functions.

Therefore, there is no certain rule for the design of the CPU, and the biggest feature of the core modularity is that you can do it as you think the design is appropriate. Intel relies on this, from desktop personal CPUs to desktop workstations and desktop servers. From large servers and workstations to laptops and tablets, to mobile phone CPUs, Intel can be seen.

Although it is dominated by ARM in the field of mobile phone and tablet computer CPU, Intel's mobile phone CPU is not weak strictly speaking, and a large part of Lenovo's mobile phones and tablets are used by Intel's mobile phone CPU.

As for Qualcomm in the United States, MediaTek in Taiwan, and HiSilicon Kirin in Huawei, they all belong to the ARM architecture system. Not to mention the high-end server and supercomputer markets, where more and more companies are already adopting Intel's Xeon processors.

Relying on the modular design idea of the core, Intel opened the market of high-end general-purpose servers and high-end supercomputer CPUs with the help of blade server design ideas, and in turn defeated the Power PC processor of IBM, the overlord of the supercomputer CPU market for a long time. Even Sun Microsystems' (sun) SPARC processors were abandoned and eventually withdrawn from the high-performance computer market after being acquired by Oracle.

Some people may ask, Intel's Xeon series is really more powerful than IBM's Power PC and SUN's SPARC?

The answer is no, in the high-performance server and supercomputer CPU market, Intel's Xeon E5 and E7 and E9 series can only be regarded as high-end, but it is also said that it is not the top CPU, IBM's Power PC and SUN's SPARC can really be regarded as the top CPU, and even DEC's Alpha is a notch stronger than Xeon.

However, the Xeon series processors have defeated all competitors in the server and supercomputer market by relying on the four magic weapons of universal use, good compatibility, low price, and low power consumption.

Taking IBM's "Deep Blue" series of supercomputers as an example, in order to achieve the same performance, using Intel's processors, the production cost of the entire supercomputer is only one-third of IBM's.

The only price that users pay is to build a slightly larger computer room, but in exchange for the common software and various system components. There is no need to spend unnecessarily high prices for a special component to be customized, which leads to a reduction in the procurement, use and maintenance costs of the system.

Isn't that powerful? It's not a dimensional thinking at all.