Secret of domestic CPU and domestic operating system: what is it?

Recently, Loongson held a grand press conference in Beijing, and claimed that there was only a 30% gap with Intel's strongest technology. Because from a certain point of view, Godson is a self-designed product. As a China person, you should be excited. But the Loongson conference made us pay more attention to the real situation of domestic CPU and supporting domestic operating system, so we found a better article in Zhihu:

Let's not talk about manufacturing, let's talk about design. There are many CPUs under the banner of domestic production, which can be divided into the following categories:

1. Loongson is designed independently from bottom to top. Although the instruction set is imitated, it is barely independent. There is not much difference in instruction sets under modern CPU architecture.

2. With an open source hardware solution, the representative types are some CPUs under the SPARC architecture, such as those used in shenzhou spaceship, which can be used by modifying other people's open source things.

3. directly buying other people's hardware solutions refers to buying a whole set of things. The representative type is alpha architecture, that is, Shenwei series, which is used by the light of Taihu Lake.

4. Take the authorization of others and then produce CPU. The representative type is Huawei Hisilicon, which was authorized by ARM and then redesigned. Although ARM will provide a public version, it is still a bit difficult to use it directly.

5. There are some research institutes that crack, copy and polish other people's CPU.

Which is the best choice for domestic CPU?

Godson is the most remarkable. Basically, all the water designed by CPU has been tried. Although there is a desktop version that can run Linux, the overall technology is a bit backward and the market prospect is not optimistic. MIPS itself is dying.

Godson will certainly live, at least as a technical reserve of national strategy, and some people will let it live, but it is hard to say whether it will live well or not. Playing with hardware is too expensive. Watching Intel earn more and burn more money.

SPARC and Alpha are actually similar in practice, one is open source, and the other is overall purchase. The problem is that whether it is open source or bought, it is basically impossible to get good things. Although Shenwei won the first place in supercomputing, Shenwei itself has many problems, such as weak memory design, which is suitable for parallel computing and not suitable for civil use.

When the light of Taihu Lake won the first place in supercomputing, many people thought that domestic CPU had become a climate. In fact, this is not the same thing. In the civil field, the performance of single core is very high. Shenwei single core may not be as good as Godson (the specific data has not been studied and may be biased). The advantage of Shenwei lies in its strong floating-point performance, but why do ordinary users need such strong floating-point performance?

SPARC is mainly used in aerospace field. Europeans have been using it, and we have followed suit (it was wrongly written before, Americans used PPC, and Europeans used SPARC). Generally speaking, there is still a certain gap in our own SPARC. If we just use it, it will be enough.

Again, neither of them is suitable for civilian use.

The fourth category is authorized design Soc, which contains many things. Broadly speaking, it seems that the PPC instruction set is also authorized in China. After being authorized by MIPS, Loongson can actually fall into this category, but after all, in the early days, Loongson was really completely designed by itself, only "compatible" with MIPS instructions, and the conditions were much more difficult when Loongson released it.

In a narrow sense, it is actually not a sign, because there is no autonomy at all. In fact, it is VIA, but if VIA is sold to China one day, it may not be a bad thing, at least it can legally produce x86 CPU (comments say that the latest megabox has redesigned IP, and if so, it will be almost the same as HiSilicon).

Different authorizations have different degrees of autonomy. Some can be deeply customized, and some can only be simply modified, because there are too many such manufacturers, and I don't fully understand them, so it is not good to evaluate them one by one. But personally, I think this road is the best from the perspective of commercialization. There is no threshold for compatibility, and there is no need to burn a lot of money (but not a little) in the early stage, so it is relatively easy to commercialize.

The last category is to crack and copy other people's films. There is basically no market at all, either in the name of scientific research and national defense, or simply cheating money. Of course, there may be two or three situations. Also, we should pay special attention to distinguish that Loongson is made by ourselves, and Hanxin swindles money by polishing. The two are not the same thing.

All of the above are about CPU design.

About manufacturing industry

In terms of production, domestic chip technology is relatively backward, and mask aligner is stuck, so it is impossible to produce high-end chips. Chip types here include but are not limited to CPU, memory, flash memory, etc. The low-end ones are ok, and the high-end ones are basically hopeless. It costs a lot of money to design CPU, and it costs more to build CPU. In this respect, there is no shortcut except buying others, but it is a little better to limit China's acquisition (Magnesium). Godson used to be commissioned by others.

Although others don't sell mask aligner (Editor's Note: According to relevant sources in ASML, this restriction has been lifted now), the main reason is that the industrial accumulation time is not long enough. From the perspective of science and technology tree, mask aligner relies on high-precision machine tools and high-precision optical equipment (lenses), but these things are not made in China. Speaking of Nikon, everyone knows that it is a SLR. In fact, Nikon also makes mask aligner, but there is no such enterprise in China, so mask aligner still has a long way to go.

"Bad Accounts" of Domestic Operating System

Speaking of domestic operating systems, this topic is too big, because there are too many domestic operating systems in a broad sense. Roughly divided into these categories:

1. Completely independent design;

2. Exchange with open source;

3. Legally change the closed source;

4. Illegal.

The above categories can be subdivided into two parts: civil market and special field, of which 2 and 3 are divided into logo for skin and deep customization.

Software is not as easy to define as hardware. After all, the instruction set can be seen at a glance. It is difficult to know whether the software is borrowed, copied or original.

The following are detailed one by one:

Completely independent design, how do you say this? There is a lot to say. An undergraduate or graduate student majoring in computer science can make a simple kernel of 16, 32, 64 bits, based on SMP/SMT and even virtualization. I wrote something similar myself. The difficulty is not high at all, but the important thing about the operating system is ecology. In this regard, the domestic self-designed system is basically zero.

Completely self-designed domestic systems are generally in special fields, in short, there are few civilian ones, such as embedded DJYOS (I don't know much about it, just heard of it) and so on.

Open source reform is the most common way, because it can avoid the dilemma of the ecosystem, but it is also easy to be accused, such as Kirin and Hongqi in the past. Basically, they have all been replaced by Linux or BSD series. There are also some deep customization, such as deepin.

Broadly speaking, the Android family also belongs to this category, and several mobile phone manufacturers are engaged in it. In addition, Alibaba Cloud and others also belong to Linux-like systems.

Most of the "domestic" operating systems that ordinary users come into contact with belong to this category. At the same time, this idea is also applicable to the field of supercomputing and embedded systems.

Many people in this category have legally changed the closed source code, but many people don't understand it. Software, like hardware, can get a complete license (source code level), and then you can DIY yourself, but it costs a lot of money. The representative type is VxWorks, which is very common in embedded systems. When I was working as a switch in the last company, I also bought the kernel of OS.

There is another kind that is illegal, but nobody cares about it, because it can't be managed, and it is mainly concentrated in industries such as national defense, aerospace and military industry. As I said before, China likes to imitate the old American plans. Whether it is hardware or software, the CPU of the hardware is copied, and so is the operating system. However, because we dare not copy directly on the surface, we have to use the banner of compatibility with XXX system. I don't know whether this system is completely independent, completely copied or mixed. It is difficult to achieve in the civil field, and I will not comment.

As far as the way out is concerned, 2, 3 and 4 are all similar, and the degree of self-control is basically the same (all have source code). Don't think that what you copy is not necessarily good. In fact, it is not so easy to copy others' own, and vice versa.

In some areas, we have to adopt plagiarism because we have to follow the pace of the United States. This is actually a good way to save money and do things. It takes too much time and money to test the water yourself. Originally, we were behind, and then the gap will only get bigger and bigger.

Some people will say that the source code is not written by China people and it is not safe. What I want to say is: I gave you all the source code and you still feel unsafe, so what is safe? Is it safe and bug-free for China people to write by themselves? This is a joke. Is it easier and more stable to write an operating system from scratch, or is it easier and more stable to study existing source code?

As for the civil system, there is actually no problem taking the road of Linux. In the future, when domestic enterprises have money, they will keep the upper API of Linux and replace the kernel. In addition, the Linux community also has many Chinese contributions. Linux belongs to the whole world.

At the same time, don't think about the importance of domestic systems. From the perspective of science and technology tree, domestic systems depend on domestic hardware, and domestic hardware depends on domestic industrial equipment (machine tools, mask aligner). Explain the dependence first, and then see which is the most important. One day, the United States blocked China's technology, and the biggest problem was the chip, not the operating system.

Moreover, building a perfect ecosystem requires a lot of money, which may be several times that of the Three Gorges Project and the Beijing-Shanghai high-speed railway. There are many places in China that are short of money. Don't think you are a programmer. All domestic resources should be invested in the software industry. Code can't be eaten as food, code can't be drunk as water, and code can't have children.