16
“In summary, to achieve true independence and autonomy, we must have our own instruction set.
Only you here can provide full support from software to auxiliary hardware, which is why I hope to establish a research plan.”
“You plan to develop the core?” Wang Xu listened to the explanation from the thirty-something professor in front of him with some surprise, quite astonished by their courage.
This project application meeting was arranged by Secretary Jiang. Now that Wang Xu has opened up channels for project applications, it provides a relatively independent possibility for project applications outside of the university and government, currently only open internally.
The applicant, Liu Weiping, came from a digitally numbered integrated circuit design center in Imperial University. He had already been deeply involved in the previous Northbridge and Southbridge projects, making him an insider. It was clear their goals were not limited to Northbridge and Southbridge.
Wang Xu participated in this project application meeting as the director of Wang Xu's laboratory. He is now fully responsible for the review of various new projects.
Current projects are no longer limited to those proposed by Wang Xu and then assigned to various departments. The opened application channels are intended to encourage professors to think more and embark on the path of independent scientific research.
This was extremely rare in the 1990s. In a period of overall backwardness, enabling some of China's top scientists to break free from convention and think ahead about things beyond catching up was exactly what Wang Xu hoped for.
Scientific research at this time, and for nearly a decade to come, was conducted in a way that was written as catching up, but read as imitation.
They would see what China couldn't produce, then study foreign products, and any progress made was considered a victory.
Another approach was introduction, which involved spending money to purchase foreign technology and then organizing domestic experts to digest and absorb it.
Most of these research directions were engineering-related, rarely involving principles and underlying Science, but they could yield some practical benefits, so both the state and enterprises were willing to provide some funding.
However, a small number of purely theoretical studies faced almost universal funding shortages, making their daily existence difficult.
The channel provided by Wang Xu was not purely theoretical either, but it was at least close to the underlying layers, basically things that wouldn't show benefits in the short term. However, the approval and review process here would be relatively simple and direct: if Wang Xu liked it, it could be funded.
“Yes, we want to give it a try. We see that your Supercomputing ultimately performs quite well, already far exceeding the level of foreign 486 workstations, which indicates that the approach of your project itself is excellent.
But obviously, you lack a core. The reason you didn't use the 486 was because coprocessors were hard to buy, so you settled for a compromise. But we can totally develop independent coprocessors, specifically for floating-point operations.
You once mentioned in Supercomputing design that a large number of identical small cores could be integrated into a single unit for parallel computing to replace the existing multi-board card model. I believe this is definitely a new type of chip, and its applications are absolutely wide-ranging.
After the last project was completed, we formed a small team and expanded on your ideas. The content I just explained is our latest achievement, and I hope Professor Wang will seriously consider it.
It's better to have your own than to rely on others. Foreign countries don't even have this kind of chip yet. We can start with a few cores and then expand the number of cores generation by generation. This is definitely the best starting point for researching a new instruction set.
Reduced instructions, parallelization, no internal cache, unified external cache, multi-pipeline, high pipeline, multi-channel.
This is the novel model you proposed. We can completely miniaturize it into a single unit, and it would still support the multi-board card model. This would mean M multiplied by N cores operating simultaneously. Just thinking about it feels wonderful; it would certainly be an absolute weapon for parallel computing Supercomputing.
Our individual chips may be weaker, but we can stack them in this way to further extend your basic concept of Supercomputing, offloading a large amount of computational work to these processing units.
The host part can then significantly reduce its computational requirements, only needing to complete task distribution and human-computer interaction. Isn't this the true future Supercomputing you mentioned in your plan?
Look, I heard you are planning to establish a second-generation workstation project, but if it's called Supercomputing, isn't it a misnomer to just produce a workstation? Why not really make it a Supercomputing?” Liu Weiping spoke like a storm, powerful and resonant. He was the same when he explained earlier, appearing particularly forceful.
This person has quick thoughts and a fast speaking pace. Someone without a certain level of skill might not even keep up with his speed. Wang Xu didn't even have time to respond before he rattled off a ton of information, from basic technology to future development—it was incredibly comprehensive.
Well, Wang Xu actually had some impression of this person; he had dealt with him in his previous life. He later became one of the big shots in core development, a high-level executive at Huada Jiutian. However, their main product was EDA, which means they focused on software.
It was only now that he learned that Liu Weiping also started with integrated circuit core R&D, and he was an alumnus, a Computer Science Department PhD who graduated a few years ago.
The greatest advantage of top universities is this almost ubiquitous network of connections. Before you know it, you discover an alumnus hidden somewhere. This is also the most important reason why Wang Xu has always stayed at Tsinghua University.
It was precisely this resource from Tsinghua that allowed him to almost seamlessly pass through the novice stage, growing into a major figure in a very short time, and now even possessing the authority to approve projects.
“Your introduction is fine. There are some details in the approach that need discussion, but overall, the direction is correct, and I approve. But are you only going to work on the coprocessor? What about the main core? How will the instruction sets be coordinated if they are all different?”
Wang Xu's question was sharp. If this problem couldn't be solved, researching a coprocessor specifically for floating-point operations would be meaningless.
“This is also the main reason we came to you. Our suggestion is the same as last time: let's work on it together.”
“It's still like the Northbridge and Southbridge project, bringing together multiple institutions. The last collaboration was very pleasant, but the premise was that you provided quite detailed low-level designs.
We hope that this time you can also take the lead. If the low-level designs remain consistent, both the central processor and the coprocessor can be designed together.
Moreover, the equipment and software teams that collaborated last time should also be brought in. The new instruction set still requires operating system support, which is another huge project, and you still need to come forward to integrate it.
This way, the second-generation Supercomputing can become a true Supercomputing. The first-generation Supercomputing's floating-point performance has already exceeded 100 million operations per second, fully reaching the level of the Galaxy Supercomputing from ten years ago.
But this speed is not enough; internationally, it has already reached one billion operations per second. However, if we develop our next-generation Supercomputing in this way, 10 billion operations per second should not be a problem.” Liu Weiping was truly talkative. Wang Xu rarely met anyone who talked more than him; today he had witnessed it.
However, what he said made sense. The previous project was basically completed, and it was indeed time to start the development of the next-generation product. This kind of thing naturally requires continuous iteration.
“Considering it this way is acceptable, but this project will be massive. I need to consider it comprehensively. Wait for my news; at the latest, within a week, I'll give you a reply.” Wang Xu didn't intend to continue the conversation. This project was fine.
Wang Xu had actually already decided to start the R&D of the second-generation “Supercomputing,” even if Liu Weiping hadn't made this trip.
However, it was better for them to propose the project. This lays the groundwork for future long-term cooperation. Fully mobilizing the enthusiasm of all participating units doesn't just rely on money; sentiment is also very important.
They are not from the same unit. In the future, there will be more cooperation, and mutual relationships still need to be built well. They propose, and this side adopts; these are all human favors, completely different from unilaterally assigning tasks.
Tsinghua Tongfang workstation is already applying for national acceptance. This is a big project; finally, there must be a plausible summary, a series of machines displayed, and various test indicators listed.
Although it cannot have various evaluations like online platforms in later generations, comparisons with workstations like SGI and Sparc on the market still need to be done.
Wang Xu wasn't worried about this aspect; he had already used this set of equipment. Although there were still some bugs, overall, it was genuinely very advanced, except for the 19-inch screen which looked a bit too small, and the 1280x1024 resolution wasn't very high.
But it was more than enough for audio synthesis. Wang Xu wasn't greedy.
To put it bluntly, this time was about team building. Many projects were even developed at the underlying level, and it was already unexpected to achieve so many results.
The future is the time for various application developments. He plans to list out the projects and then slowly consider their priorities.
Even though it hasn't been accepted yet, the commercialization of the workstation is already being handled by Tsinghua Tongfang, a Computer Science company, and will be licensed to several other companies in the future.
Founder Group from Imperial University and Inspur from Shandong province have already signed up, and it's expected they will get licenses very soon. The state will not allow such an important thing to be monopolized by one company.
Founder Group started with laser phototypesetting, making it a leader in the printing industry. Recently, it also started making computers, but sales weren't great. Tsinghua Tongfang came much later; in this life, it was founded from scratch.
Even the name itself implies a rivalry with Founder Group.
While Inspur also makes personal computers, it's not their main business. They primarily target the server market.
They just launched their mini-server products this year. In terms of computing performance, they are not as good as his own.
However, the server's network processing capability is first-rate, so they plan to combine the technologies of both to create a server with a completely new architecture.
On Tsinghua Tongfang's side, they primarily focus on the scientific computing field. Currently, their top single device's floating-point operations exceed 100 million per second, which is already the level of current mini-computers, offering excellent performance.
The current target companies and products are United States's SGI workstations, the kind used to create the liquid robot in "Terminator."
Actually, there weren't many real computer-generated special effects shots in that film; most of them still used practical effects.
For example, the scene where the T800 finally sinks into molten steel used milk underneath, with strong orange light to create the molten steel effect. So, Arnold just took a milk bath in the end; it's quite good for the skin, huh?
Only the shot of the liquid robot rising from the ground was truly made with computer special effects.
Therefore, even now, the proportion of computer special effects in movies is still negligible. The main special effects methods, including those in Hollywood, are still predominantly traditional.
Wang Xu's entry now means there won't be too many technical shortcomings. Currently, dedicated 3D processing software products are already in the open-source community, just not fully developed yet.
Moreover, the technical foundation of this software is a pile of scientific computing content, all physical engines like ray tracing, gas simulation, and explosion simulation, which can be shared with the scientific computing software part.
Therefore, Wang Xu has separately extracted the core function packages of the basic computation part, existing as independent open-source projects.
These scientific computing components, when fully developed and combined with a suitable interface, would become the famous MatLab of later generations. However, MatLab is commercial software and not open-source, whereas this set from Tsinghua University is open-source.
Although no one has the energy to create wrapper packages yet, it is believed that in a few years, this will impact MatLab's market.
The ultimate 3D engine, on the other hand, focuses more on interaction and the connection between various physical engines.
This engine also shares many commonalities with 3D CAD software and 3D animation software, so it has been further divided into smaller open-source projects for public development.
These software engineering divisions and combinations were what consumed the most of Wang Xu's energy and were his most important work as the project leader.
Currently, the open-source community's projects are still very limited, but quite a few excellent projects have already emerged, such as the audio processing software Wang Xu had been longing for, which has been successfully developed.
It is sufficient to meet student Wang Xu's needs for graphical drag-and-drop capabilities in various audio syntheses.
At that time, to achieve this function, Wang Xu also incidentally developed a graphical user interface. As for the style, my friend, have you heard of Windows 98?
Anyway, the core essence of graphical interaction is actually the underlying interfaces; the surface interaction pages can be designed and developed with different interface styles according to preference, like the Windows 98 style.
It was created by Wang Xu out of mischief, primarily aiming to 'walk others' paths and leave them with no path to walk.'
The graphical interface itself was developed in C++. Most of the main engineers came from the Chinese Academy of Sciences' Institute of Computing Technology. This part was undertaken by them, a group of top-level programmers. Coupled with the quite mature underlying framework design provided by Wang Xu, the overall operational effect was quite explosive.
Even on the most basic single-CPU workstation, this graphical interface can run very smoothly and stably. Of course, some minor issues are inevitable, and it is currently undergoing testing from all over the world in the open-source community.
Wang Xu had also long ago provided a dedicated language for developing graphical interfaces, named C#, borrowing the name Microsoft used in later generations. The syntax is similar, but the class libraries are completely different.
The C# provided by Wang Xu can integrate almost perfectly with C language. With the help of development tools, it can become a layer wrapping C language, which greatly simplifies the development process of graphical interactive functions.
Moreover, because this C# memory management is based on pre-allocated virtual memory, it almost never causes memory overflow issues, so the final stability of the product is relatively easy to guarantee. This is an interface development tool specifically for ordinary programmers.
Moreover, this set of tools has also been similarly adopted in the development process of the 3D engine, completely separating graphical display from complex backend computations, making it possible to achieve simple development while maintaining high operational efficiency.
This is one of the biggest advantages of having a unified planner for a product: there are simply too many reusable parts. And precisely because of this, the various standard libraries developed by Wang Xu are all battle-tested experts, almost ubiquitous in their use.
—
(See you tomorrow. Chapter 4 has been revised twice and still hasn't passed.
This chapter, inadvertently, included some chip-related content. May the Three Pure Ones protect it, and let it pass the review in one go.
It's harder to revise than writing explicit content, the key is you don't know what exactly got it banned.)
