Chapter 8: Never Fought Such a Wealthy Battle

But he had no other choice now.

He obediently typed two words to his senior fellow students: "Understood."

He opened his computer, downloaded Zotero, spent half an hour setting up cloud storage and installing several plugins, and finally imported all dozen or so papers his professor had sent him.

It was Monday, and with four days left until Friday, he had plenty of time.

He certainly didn’t need to read all of them; for the presentation, he would likely just pick one. The group meeting was on Friday, so preparing on Thursday would suffice. He planned to use the intervening days to refine his own research.

Edge deployment of AI models—this was not a topic he had chosen randomly, so he naturally had to take it seriously.

However, the screen he was using was his laptop's built-in display. It was far too small, making reading papers quite uncomfortable.

He checked his phone and saw his balance: a little over two thousand. This was money he had saved from his undergraduate living expenses.

After a moment's thought, he decided to buy a monitor.

A model costing a little over a thousand would be sufficient; he could always upgrade to a better one once he had more money.

After placing the order, Zhou Yun pulled a sheet of white paper from his desk and began to jot down notes.

This was one of his habits; when thinking through problems, he preferred paper and pen over a computer.

Time always flew by when he was deep in thought.

"Knock, knock. Son, dinner's ready."

"Coming."

After dinner, he returned to his research, continuing until just before the evening meal.

"Ugh!"

Having sat for the entire day, Zhou Yun leaned back in his chair and stretched.

Today, he had primarily focused on further refining the technical roadmap for "AgileEdge" and polishing the details of the overall framework. This was as far as he could take it at this stage.

Even if he were a genius, he couldn't simulate the experimental results in his mind. This was a common issue in the entire AI industry; due to the black-box nature of these models, results were almost entirely unpredictable until one actually ran the experiments.

Unless one was replicating someone else's project, and even then, it often required multiple adjustments to reproduce their results. Sometimes it was impossible to reproduce them at all—after all, if the original paper didn't state the parameters clearly, heaven knows how they were tuned.

Now, relying solely on himself, the project had stalled. According to Zhou Yun's estimates, even though AgileEdge didn't involve image data—meaning it didn't require the exorbitant VRAM needed in the visual field—it would still require at least 100GB of VRAM. This figure was based on his research experience from his past life.

Currently, the best consumer-grade graphics card, the 4090, only had 24GB of VRAM. AMD cards were automatically excluded because, without CUDA, they were nearly useless for AI training.

The prices of professional-grade cards were well beyond Zhou Yun's reach. A common choice like the A6000, with 48GB of VRAM, cost thirty to forty thousand, while the more powerful H100 could cost one or two hundred thousand per unit.

In other words, to run Zhou Yun's model, it would cost at least a hundred thousand just to get started.

This was exactly why he needed a professor with resources. If he had a minor supervisor, the entire group might only have one 4060, which would be barely enough for personal use, let alone running actual experiments.

They say the squeaky wheel gets the grease; if you don't speak up, a professor won't proactively offer resources even if they have them.

Zhou Yun sent a message to Deng Yonghua.

"Professor, I have some new ideas regarding the AI edge deployment project I reported to you earlier. However, my own computer isn't capable of running the experiments, and I need a server. Does our group have one?"

He waited a few minutes after sending the message, but there was no reply. The professor was likely either in class or a meeting.

There was no helping it; that’s just how it was with professors. They expect an instant reply when they message you, but if you message them, waiting for hours is common. He was used to it.

He set his phone aside and logged into his GitHub account.

He created a new repository, set it to public, and named it "Learning Log."

Ever since he started using GitHub, Zhou Yun treated it as a universal notebook. Whether it was code, reading notes, or study notes, he recorded everything there.

Although he had spent over two years doing AI research in his past life, he knew that what he had touched upon was merely the surface.

The primary reason for this was that his mathematical foundation was not solid enough, which caused his research to remain at a superficial level.

Mathematics is the cornerstone of all disciplines, especially for a practical field like computer science.

For example, the convolution operation, which has been most commonly used in the field of computer vision for the past few years, was actually proposed as early as the 18th century, yet it wasn't applied to neural networks until the 20th century—a gap of a hundred years.

There are many algorithms that look incredibly impressive, but behind them are simply applications of ordinary mathematical formulas.

This clearly shows the importance of mathematics to computer science.

Zhou Yun's purpose in studying mathematics was not to prove conjectures—that was the domain of the gods. Even though he was significantly smarter now, he wouldn't be so arrogant as to dabble in those things.

Compared to pure number theory research, he preferred applied fields like computer science.

He made a simple list; there was quite a lot he needed to catch up on.

Linear algebra was certainly the most important. He hadn't studied it very deeply while preparing for the graduate entrance exam, but it was now the most widely used component in various neural networks.

He also needed to study calculus, probability theory, optimization theory, discrete mathematics, and Fourier analysis, among others—all, of course, to a deeper level than when he was preparing for the entrance exam.

He checked the time; it was already 5:30.

His professor had just replied to his message.

"The group doesn't have one at the moment. If you need it now, I can borrow an account for you, or you can rent one and I will reimburse you."

Perhaps fearing that Zhou Yun might overthink it, Deng Yonghua added an explanation.

"Our group has previously focused on cybersecurity, and artificial intelligence is something I only decided to get involved with this year. We haven't bought a server yet. Things have been busy at the school lately, so the server might not be finalized until the summer break."

"Understood, thank you, Professor. May I ask what the server configuration is? If it's not sufficient, I'll go ahead and rent one."

"Let me ask."

After waiting two minutes, Deng Yonghua replied.

"An eight-card H100 system. That should be enough, right?"

"That's plenty, thank you, Professor."

"Alright, I'll send you the account and password later."

"Understood."

After replying, Zhou Yun clicked his tongue. He knew his professor was capable, but he hadn't expected him to be *this* capable.

Eight H100s. He had never had such abundant resources before, and he wondered how his professor had managed to get them.

Even so, he didn't intend to continue his research now. He decisively shut down his computer; his workload for the day was enough, and there was no need to push himself unnecessarily.

He would deal with the rest tomorrow.

Now, it was time to eat, and then do something he enjoyed.