Chapter 277 - 270: Far Ahead_1

The initial batch of Kirin959 processors was just over two thousand, and thus the unit cost was high. However, it was primarily for testing and not a big deal as the cost would only increase slightly when mass-produced.

Compared with the previous flagship Mate9, the all-tri-ternary Mate10 had a slightly higher energy consumption, but this was acceptable.

The system based on the optimized XOS3.2 skin was named HarmonOS (HarmonOS) and was already usable.

The lab had already assembled four prototype devices. There wasn’t much to say about the appearance. Yu Chendong picked one up and smoothly entered the desktop.

XOS was refined but not bloated, more fluid than the current Android7, and of course, this was also due to the powerful processing ability bestowed by Kirin959. Now, SinoCore internally considered that 28nm tri-ternary integrated circuits should be competitive with 14nm mainstream processes, while the 950 used TSMC’s 16nm process.

Although most of Kirin959 was a direct translation of the 950 chip design, following guidance from Yellow River Semiconductor, HiSilicon designed a small AI module, the first attempt at utilizing tri-ternary attributes for AI computations in hardware.

They named it NPU (neuron processing unit) and planned to use it to support smart voice assistants and camera algorithm tuning.

Upon comprehensive evaluation, the lab believed that Kirin959 had surpassed the Qualcomm Snapdragon 820 by about 6% in performance and was now likely the best.

On the previous show-off delivery ceremony, the Qiongqi dialogue AI on the Great Wall G1X laptop, equipped with X32115, greatly shocked the domestic digital manufacturers; subsequently, companies like Xiaomi, OPPO, and LeEco all declared they were developing their dialogue AIs.

Huawei was no exception. They had previously considered voice assistants and acknowledged that to upgrade to artificial intelligence was no small difficulty. The lab estimated that it would take at least a year to feed enough data for the smartphone to match "Qiongqi" and specially tune the NPU, with a release planned for 2018.

The Mate10, expected to go on sale in January, would initially lack this feature, but it should be available after subsequent upgrades.

The current issue was apps.

In 2016, the ecosystem of commonly used mobile software had not expanded as rapidly.

Yellow River Semiconductor had already joined forces with domestic manufacturers to seek direct translation for community apps like QQ, WeChat, Weibo, Baidu, TikTok, etc., rather than developing them from scratch, since these apps had low performance requirements. First, they’d make do and later develop optimization specifically.

The rest was games, with King of Glory already online for nearly a year being too bulky for direct translation. Tencent had taken over the re-coding work and planned to complete it in 4 to 6 months.

Lin Ju knew that after 2017, with the ultra-high penetration rate of 4G, the number of software applications would explode. Thus, Wang Minjiang was fervently promoting teaching of the X language and basic tri-ternary circuit design in universities to build up talent reserves ahead of time.

Centered around Yellow River Semiconductor, the group of manufacturers spanning both software and hardware had established a new Tri-ternary Semiconductor Union under the active promotion of the science and technology commission, ready to support downstream software vendors.

Yu Chendong jotted down some future work priorities in the memo of the test device. The operating logic of HarmonOS was pretty much the same as regular Android; he hardly felt any difference.

The basics were sufficient, but in the earliest four months later mainstream game developers would complete adaptation to the XOS environment and needed to fully unleash the potential of the SOC, hence optimization patches would probably be numerous.

"I will run I will climb I will soar..."

Yu’s personal phone rang; it was the science and technology commission.

"Hey, leader... What? 500 units? No problem, no problem, office software will certainly be there, all documents are supported, Kingsoft Company has also joined in. Okay, I’ll get right on it."

The lab’s research director was on good terms with Old Yu and stepped forward to ask, "What’s up, someone wants to purchase 500 phones?"

Yu Chendong: "The science and technology commission suddenly needs a batch of Mate10s for secure use. It looks like ZTE and Unigroup aren’t as fast as we are, so they have come to us. They need them ready and delivered by mid-December."

"Tsk, tri-ternary is indeed secure enough. Even our own internet access and phone calls need a special converter chip to work. Outsiders won’t have any means to break in."

Yu Chendong: "It’s not necessarily true, X language and the teaching materials for basic ternary integrated circuit design have already started to be forcefully promoted in universities, which makes them basically semi-public, so don’t underestimate foreigners."

"So what? We are already at the forefront, even if we are merely competent with ternary but not fluent, we are still far ahead of our peers, both in hardware and software."

Yu Chendong: "You can’t include Yellow River Semiconductor among those peers. They are rumored to be launching three series of consumer desktop processors and two server chips in one go. They are the inventors of ternary computers; no one in the world understands them better.

Alas, we also need to work on servers and big data, for the future belongs to AI, and without these, it won’t do. Mobile phones are just one of the many businesses we have."

Having said that, he lifted the Mate10 prototype in his hand; its substantial weight felt reassuring.

"However, computers are a different story. When I hold it up at the press conference at the end of December, I will be able to say that in the field of mobile phones, this is the real deal—far ahead of the competition!"

...

California, Santa Clara.

This is the headquarters of Intel, and at the moment, in Intel’s product testing lab, a group of executives and over a dozen Chinese engineers were gathered, studying three books on the table:

"X Language Programming Fundamentals (Editor-in-chief: Wang Minjiang)"

"Integrated Circuits: Ternary"

"The Third State and Artificial Intelligence"

After the semiconductor summit, during the technical challenges coordinated between industry enterprises and academic institutions by the Science and Engineering Committee, they didn’t forget to nurture the foundation. They produced three basic textbooks as quickly as possible and distributed them to select universities, starting to teach them in computer science and microelectronics departments.

The textbook writing wasn’t complicated, and the content was quite simple because, although they had already started teaching them in the new semester, the teaching faculty also had to go nearby to participate in technical battles with enterprises and then simultaneously teach the students.

Therefore, the textbook content was only for beginners, with more textbooks in the process of being written.

However, since the textbooks were written by professionals (primarily from Yellow River Semiconductor), these basic teaching materials for ternary software and hardware were going to be rare resources in the national education system that could keep pace with the industry’s cutting edge, preventing the situation where textbooks are severely outdated compared to practical applications.

The foreign behemoths that were humbled at the summit weren’t idle either; they got hold of these three textbooks at the first opportunity:

The China regional president directly bribed the cleaning staff at the university, stealing the three books during a break between classes, then immediately sent them on an overnight flight to America.

Although the X language supported both Chinese and English, the grammar was still significantly different, and it was all written by local professionals, making it challenging for foreigners to understand.

Pat Gelsinger organized the Chinese engineers at headquarters, and after more than a month of arduous study, they were finally able to test the performance of ternary chips their way.

They transformed a 32116 chip, removed from a Xin’an car, through various complex methods into a computer, then used their own rudimentary system to complete a series of performance tests.

Gordon Moore: "How is it? What level has the 32116 reached?"

Chinese engineer: "We’ve cracked a small part of Xin’an car’s external environment to get as close as possible to their performance release. To make an analogy, the 32116 could roughly match the computing power of an E5 2689 V4 but leans more towards the operating method of a graphics processor, and also possesses rendering and deep learning capabilities for AI.

Especially the latter has enormous potential. The symmetric ternary’s third state gives AI more flexible thinking capabilities, and lower calculation delays. All in all, it combines the functions of mainstream professional processors and computing cards, with a particular emphasis on the AI part."