Military Industry of Great Powers: Build the Strongest Flanker

Large computers can quickly promote the development of high-precision technology, especially in the military industry.

In fact, the initial impetus for the development of mainframe computers was the military.

Zhao Guoqing's plan has been demonstrated. It is not an ordinary von Neumann structure, but uses parallel computing cards to achieve large-scale vector calculations.

The main task of the CPU is to decompose the program into several tasks and assign them to the computing card array for calculation through a certain algorithm.

Wu Zhongkang, chief engineer of the 102 Computing Center, is also responsible for the research and development of 64-bit ARMv7 processors. He is aware of Director Zhao's ambition.

Pile of hardware.

The 45nm process 64M DRAM memory particles give too much room for imagination.

"Director Zhao, 8 memory chips on the front and back can realize a single 512m capacity memory stick, four channels can achieve 1600M Hz data bandwidth, and 16 memory chips can achieve 8G capacity memory. We use ECC error correction technology to ensure data storage safety."

Wu Zhongkang took a long breath after the introduction. There are 16 memories with a total capacity of 8G. Even the previous mainframes did not use so much memory!

The CPU is two quad-core ARMv7 processors with a total of 8 cores. It uses hyper-threading technology and can simulate 16 cores. The design frequency is 1.6Ghz.

What’s even more terrifying is that the two CPUs integrate a total of 256 PCI channels directly connected to the CPU, which means that 256 computing cards can be connected.

Each computing card is designed with 64 stream processors. There are 256 computing cards with a total of 16,384 stream processors. The design frequency of the stream processor is 400Mhz. Each computing card is equipped with 16 DDR second-generation memory particles with a capacity of 64M. The total capacity is A G.

Director Zhao was so ambitious that Director Pang, who was listening to his introduction, was stunned.

He asked: "How fast can the design be calculated?"

Zhao Guoqing said: "It is estimated that the computer can achieve 370 billion operations per second. In fact, some optimization of parallel computing needs to be done in the software!"

Director Pang shook his head and said, "I didn't expect it! We have made such rapid progress in supercomputers!"

This speed has far exceeded the existing supercomputers in technologically powerful countries such as the United States, Japan, and Germany.

Everyone present, including Zhao Guoqing, was very calm. A lot of data he introduced seemed very impressive to others. In fact, in the original time and space, the processing performance of this performance could not even catch up with the 4090 launched by NVIDIA thirty years later. graphics card.

You read that right, the computing performance of a 4090 graphics card can instantly beat the mainframe computers in any country in 1994, and the performance is even dozens of times that of these mainframe computers.

Director Pang asked: "Mr. Zhao, how long will it take to see such a computer?"

Zhao Guoqing said: "The CPU is an ARMv7 64-bit processor, which applies many new technologies, such as 64-bit addressing, multi-core, hyper-threading, and deep pipeline. These technologies still need some demonstration. The PCI bus technology is relatively mature, and the graphics card stream Processor design is also a brand-new technology, and the Linux64 operating system needs to redesign the core for these new hardware!"

In the past, when Zhao Guoqing developed new projects, most of them used mature technologies. Director Pang has studied this. For example, radar used the 1553B data bus, and the first generation of computers used the SGI compatible data bus promoted by IBM, ARMv3, v5, and v6. The processor has also made slight changes based on the original version.

And this time...

In terms of the ratio between the use of new technologies and the use of old technologies, the proportion of new technologies used has exceeded 50%.

Director Pang said with some worry: "With so many new technologies, will the risks be greater?"

The tape-out cost of 45nm chips is too high. The tape-out cost of CPU has reached 40 million US dollars, and the tape-out cost of voodoo v2 stream processing chip has also reached 30 million US dollars.

The main control north and south bridge chips do not need to use the 45nm process, but NAND storage and memory DRAM particles must use the latest process.

Director Pang said this, and Zhao Guoqing was also a little worried.

NAND storage, DRAM particle structure is simple, and tape-out costs only a few million dollars. But for CPU and GPU, if it goes well, you can see the results after dozens of tape-outs, but if it doesn't, you won't know.

Zhao Guoqing said: "I have prepared the funds and expect to invest US$1 billion!"

"..." Director Pang didn't know what to say. One billion US dollars, which is seven to eight billion yuan, would be difficult for even the country to come up with so much money at once.

What he is more concerned about is, how much does a computer cost?

Zhao Guoqing said: "Same as before, it depends on how many units you need!"

Director Pang tried to say: "What if there are 100 units!"

Zhao Guoqing said: "The research and development cost is one billion US dollars, and the production cost is one billion US dollars. When the time comes, we will launch some microcomputer versions to dilute the research and development costs. It is estimated that one unit will cost US$15 million!"

"..."

One machine costs nearly 100 million yuan, which is a big step up from the 20 million yuan of the previous generation. This computer is really not that fun.

Zhao Guoqing also nodded: "The United States has long proposed that civilians use it to support military industry. Our country also tried this in 1989. Director Pang, computers have their own special characteristics. If the country can liberalize exports, 102 Institute can make every The cost of a computer is less than 200,000..."

Director Pang said: "This is not something I can decide. Besides, supercomputers are strategic resources. Do you remember that in 1986, our country bought an IBM Kalim V2 supercomputer? Yingjiang was afraid that we would use this A computer is used in the military, and three engineers are sent to supervise our use!"

Zhao Guoqing knew about this. The United States did not even allow our engineers to operate the computer. Whenever there were any computing tasks, they were handed over to the United States. At that time, we spent 30 million US dollars to buy such a big baby, and its performance could not keep up with the 102 The Longteng x109 minicomputer developed by the institute the year before last.

Zhao Guoqing sighed: "I wonder when the Longteng x109 computer will be unblocked?"

Director Pang asked: "When do you think the lockdown can be lifted?"

"...Director Pang, if you ask me, of course I think the faster the better, as long as we maintain a two- or three-generation technological advantage over the West!"

Zhao Guoqing believes that this is sustainable development, and uses the 180-nanometer process Longteng x109 to process the development of the latest 45-nanometer process Longteng v209 supercomputer.

The latest technology will be verified on the Longteng v209 computer. After the technology matures, it will be transferred to ARMv6, ARMv5, ARMv3 processors and civilian-level voodoo graphics cards.

This forms a perfect closed loop, which can perfectly solve the risks caused by research and development, and of course also allow the country to purchase cheaper supercomputers.

Director Pang said: "Isn't there any risk if the 180nm process is exposed?"

Zhao Guoqing smiled and said: "Director Pang, I am afraid that foreign laboratories have already completed the research and development of the 180-nanometer process, but they are still far from commercialization. We can also say that we have not completed the commercial process, so the price of laboratory production is extremely expensive. Well!"

"Besides, the performance of the Longteng x109 computer is not outstanding enough. It is not even comparable to their supercomputer, and the help it can give them is limited!"

"Exporting Dragonite x109 can not only demonstrate our ability to develop high-end computing, but also promote our Linux operating system, our own data bus, and our own technical standards!"

"..." Director Pang thought for a moment, nodded, and said, "You can try it, Mr. Zhao, how about you write a report!"

Zhao Guoqing nodded!

"How much does one cost?"

Zhao Guoqing said: "Super large computers are cost-effective and are a symbol of national strength. Apart from this, the Longteng x109 can compete with mid-sized computers. The current price of mid-sized computers in the world is about 10 to 20 million US dollars. The performance of the Longteng x109 is significantly better." For them, I think there is still a market even if it costs US$30 million!"

"Thirty million U.S. dollars?" Director Pang was in a panic. The cost of state procurement was only 20 million yuan. This was a high price due to the dilution of research and development costs.

Zhao Guoqing smiled and said: "We can also learn from Eagle Sauce and indicate in the sales contract that it cannot be used in the military, and send our technical personnel to supervise their use!"

"Haha, isn't it that things have changed?"