Back to the imperial capital, Wang Yinan's laboratory.

Li Wenjing's cousin, named Zhang Zifeng, has been working in the hurricane company before, from the front desk to the HR manager, and then to the administrative director, I heard that my cousin came out to start a business, but instead of quitting the hurricane job, saying that I was not happy to do it there, and I will mess with my cousin you in the future.

Sun Wei also fooled Feng Xiang, the IC design master who was originally going to go to the magic capital with him.

Ronghe Technology has finally begun to take shape, let's take stock of the various gods of Ronghe Technology.

CEO Wang Yinan, a young genius scientist, is proficient in condensed matter physics and software design.

CTO Li Wenjing is a genius mathematician.

CFO Zhang Qi, a middle-aged investment uncle, is full of gambling and discernment.

The theory department is headed by Li Wenjing, and his subordinates are Zhao Keqin from the Institute of Computing of the Academy of Sciences.

The software department is personally in charge of Wang Yinan, and his subordinates are Li Fei and Zhao Tiancheng.

The hardware department is in charge of Sun Wei, and there is currently a Feng Xiang under him.

The administration department and HR also have the front desk, which is in charge of Li Wenjing's cousin Zhang Zifeng.

These are the first 9 employees of Ronghe Technology.

Wang Yinan presided over the first technical seminar of Ronghe Technology, and began a heated discussion on the core task of the company at this stage, improving the computing power of the Gödel system.

Li Wenjing first proposed the plan of the Theory Department, which involves a certain deformation of Gödel's theorem, which can reduce the requirements of computational power if some theoretical limitations can be broken through mathematically.

"How much can it be reduced," Wang Yinan asked.

"A conservative estimate of 20%," Li Wenjing replied.

"What about the most optimistic estimates?"

"It should be up to 50%"

"Then give up this direction, the cost performance is too low, spending so much effort to study the theory, at most, it is equivalent to doubling the computing power." Wang Yinan vetoed Li Wenjing's direction for improvement.

"The recent task of the theory group should be to further improve the theoretical tools of the formalization of the real system, only in this way can the real system be mapped to the virtual space more quickly, and the ability of the Gödel system to solve real problems can be improved faster."

Wang Yinan said.

"Well, this is a very good idea, we spent three days to formalize Sun Wei's paper, and if there is a breakthrough in this area, it can be done in three hours," Li Wenjing agreed with Wang Yinan.

"Let me introduce the software department's plan," Wang Yinan said.

"After my discussion with Li Fei and Zhao Tiancheng, I also used the Gödel system to analyze my own operation links. We found that in the current operation process of the Gödel system, the main bottleneck is still in the search and matching of multi-dimensional space, which accounts for more than 90% of the entire system running time, and even for some complex systems, the operation time accounts for an astonishing 99%. ”

"In response to this situation, we have carefully analyzed the search and matching of multi-dimensional spaces, and theoretically, this is an NP complete problem, that is, it is likely to be unsolvable in polynomial time, and this problem is especially suitable for large-scale parallel processing, and the efficiency of the traditional CPU system is very poor."

"In fact, our current Gödel system is already the second generation, the first generation of the main computing is carried out by the CPU, the efficiency is very poor, we have done an upgrade, using the GPU, that is, the graphics card as the main computing unit, the second generation of the system than the first generation of the system computing power increased dozens of times."

"But if you want to further improve, there are still too few parallel units of the GPU"

"And the Gödel system has some unique, complex algorithms, even if they use GPUs to calculate, they are still very inefficient," Wang said.

"Therefore, we pin our hopes on FPGA, that is, dedicated logic circuits, and Sun Wei will introduce this solution."

Sun Wei walked to the whiteboard and took the pen.

"Based on the theory proposed by the theory group and the specific requirements proposed by the software group, our group believes that under the existing conditions, the use of FPGA to realize the search and matching of special circuits in multi-dimensional space is the best solution."

"At present, there is a lot of international research, mainly Dr. Peng from Mizuki University, and Professor Tyler from Standford, who is my supervisor. However, their research still only involves three-dimensional space, and there is no mature solution for higher-dimensional space, and it is quite difficult. ”

"However," Sun Wei said, "even if it is an immature solution, it is more than five times more efficient than the current solution that uses GPUs."

"Of course, don't forget that we also have the Gödel system itself, based on the previous performance on the H265 algorithm, I expect that the overall efficiency can be increased by 20 times after the continuous improvement of the FPGA design scheme through the Gödel system."

"Twenty times", Wang Yinan's eyes lit up.

According to his and Li Wenjing's research on Li Shiyao's manuscript, the current Gödel system can increase the computing power by two orders of magnitude, and the lattice model at the electronic level can be formalized, so that the possible intervals of new superconducting materials can be accurately predicted. In this way, it can not only avoid the patent barrier of Cheng Qian, but also theoretically and thoroughly explain the behavior of superconductors such as liquid nitrogen temperature region.

"Then what are you waiting for, let's start doing it, Sun Wei, I also worked on FPGA for a while, you are short of manpower, just use me as an animal"

"Boss, we have to wait for money, FPGA is quite expensive"

"Buy, buy, buy, we are rich people, we can buy whatever we need!"

In the following months, in Wang Yinan's eyes that were about to cry, the soft girl coin went with the wind without hesitation.

One piece at a time, the equipment is installed in the laboratory. Sun Wei and Feng Xiang, and of course Wang Yinan, based on an article published by Professor Tyler in IEEE, Li Wenjing and Li Wenjing designed the first generation of FPGA dedicated algorithm circuits after improving Tyler's algorithm a little.

Using the expensive equipment they had purchased, they fired the chips and built a new dedicated cabinet in the basement to test the new FPGA chips.

It is said that the new FPGA chip is first connected to the existing Gödel system, and then the Gödel system's own analysis and diagnosis capabilities are used to find defects in these FPGA circuits and make improvements.

After the improvement, the FPGA chip was re-fired, and then the previous generation FPGA chip was replaced and reconnected into the Gödel system.

Wang Yinan likened this iteration to "a giant who grabs his hair and raises himself little by little"

"Well, we theoretically proved that it is possible to grab the hair and lift yourself up"

As the computing power of Gödel's system increases, the time to find defects for improvement is slowly decreasing.

The time required for each iteration has also been reduced from the earliest week to the last three hours.

Finally, after more than three months of hard work, Gödel's system was already struggling to find the flaws in the final version of the FPGA circuitry.

In the afternoon of this day, the whole team gathered at the workstation on the first floor, and Wang Yinan entered the Gödel system with a formal backup of Sun Wei's H265 paper. How much the Gödel system has improved after evolution, the operating data will tell. At that time, the Gödel system of the previous generation searched for flaws in that article and ran from 8 p.m. to 4 a.m., which is 8 hours.

After initiating the analysis command from the terminal, there was no sound at all.

One minute, two minutes,

Five minutes, ten minutes passed, and there was only the humming of electricity in the room.

Twenty minutes passed,

Just after twenty-five minutes, a beep sounded.

"Yay!"

"We made it!"

Li Wenjing hugged Wang Yinan, "We succeeded," Sun Wei looked at the big men around him, okay, brother won't hug it.

"The running time has been shortened from eight hours, or 480 minutes, to 26 minutes, which is nearly 20 times more efficient, and basically met the design goal," said Wang Yinan.

"I'll see the results of the run", Wang Yinan clicked on the output defect list.

"Well, there is no problem with the results, the first defect is to remove the screening link, but the results are more accurate, and the efficiency is 16.4%"

"The second flaw is still the problem of competing locks, and it is no problem"

"Wait, what is this?" Wang Yinan's fingers on the screen suddenly stopped.

"In addition to the Coding Unit (CUS), Prediction Unit (PU) and Transformation Unit (TUS), a preprocessing unit can be introduced to classify the distribution of images in the temporal and spatial domains, so as to automatically determine the preset macro block size, which is expected to improve the coding efficiency by 24%".

In the H265 specification, there is no pre-processing unit at all.

"Wen Jing, is this the result of your latest research?" asked Wang Yinan.

"No," Li Wenjing was also confused.

Wang Yinan looked at Sun Wei, Sun Wei's head shook like a rattle, "I don't have such a big ability to modify the H265 standard"

"Looks like something terrible has happened...

"Sun Wei, let's add a pretreatment unit according to this plan, Wen Jing, your theory team analyzes whether it is theoretically feasible."

"Go Go Go" Wang Yinan clapped his hands, "all move"

An hour later, the new FPGA chip was fired, and this time they didn't use the simulator to burn the chip. The video stream is fed in.

The results did not surprise Wang Yinan, and the running time was shortened by a quarter with the same quality.

This is by far the world's most efficient H265 chip.

There is only one explanation, that the Gödel system has acquired some kind of deeper intelligence through an order of magnitude increase in computing power, from quantitative to qualitative, and in some cases, can even go beyond the initially set problem domain.

"This matter must be kept strictly confidential," Wang Yinan said.

Everyone understood what that meant, and everyone nodded stupidly.

"The company's security system needs to be strengthened, let Zhang Qi come over tomorrow, let's discuss the security plan"

"Today Sun Wei, you and I won't go back, we are on duty in the company"

Early the next morning, Zhang Qi hurriedly came to the company, Wang Yinan was washing, and he didn't have time to tell Zhang Qi about this major progress.

"We're in big trouble," Ms. Zhang said.