Skip to main content
  • Home
  • Library
  • Read
    • Popular
    • Charts
    • Browse
    • Categories
    • Wikis
    • Glossary
  • Request
  • Activity
  • Leaderboard
  • Help
Our Apps
  • DaoSearch
  • Narratorr
Social
  • Discord
  • Patreon
The AI Wave Tycoon's Rise
Chapter 423

Computing Power Scheduling

2,578 words 9 min read 1 view 1mo ago Source ​
Skip to the end of the chapter
Prev
Next

Not long after Su Niannian left, Han Luyi carefully reviewed the information she had sent him and then made an internal call to Liu Yu.

"Could you ask Jiang Songran he's here right now?"

A few minutes later, Liu Yu replied, "Boss Han, Jiang Songran always in his office. Should I have him come over?"

"No need, we'll go find him."

Han Luyi finished speaking, he got up and left.

He's currently concerned with the real-time status of Computing Power, and he didn't ask Jiang Songran to prepare in advance. Going to Jiang Songran's place would make it easier to retrieve the data, while having him come over to report would waste time.

Han Luyi arrived at Jiang Songran's office, Jiang Songran already waiting outside the door.

"Boss Han."

"Come on, let's go inside and talk." Han Luyi said, leading the way through the door. Without even sitting down, he immediately asked, "You've been busy with Computing Power since you returned from the Beijing. How's it going?"

Jiang Songran guessed that Han Luyi would ask this question, since it was both a company strategy and his recent work focus, so he made some preparations in advance.

Jiang Songran to the computer and brought up the real-time monitoring data panel: "Boss Han, look, the inference and training clusters are currently running."

After Han Luyi promptly asked the question that concerned him most.

What are the approximate efficiencies of the J590 and Dragon Soar 2000?

Upon hearing this question, Jiang Songran looked a little embarrassed.

Boss Han very accurate eye for the problem; he asked about exactly the worst-looking data right now.

"Based on the benchmark test results, the J590 can reach 80% of its theoretical Computing Power, while the Dragon Soar 2000 only reaches 63%," Jiang Songran said.

Is it really that low?

Han Luyi started to think.

Looking back now, his initial idea of using an intermediate layer to aggregate Domestic Computing Power Alliance was correct. He has also taken a solid first step in the inference direction, and several domestic manufacturers have now agreed to connect to this Inference Framework.

However, the problem now is that this middleware layer is natively adapted to the underlying Hardware of the original Ximing L100. Although the engineering team has tried its best to make it Hardware- insensitive during the abstraction process, some underlying implementations are simply not available. When other manufacturers' designs are integrated with this middleware layer, there will be a performance loss.

Of course, this doesn't mean the middleware approach is wrong. After all, if each vendor were to implement it themselves, they wouldn't even be able to get it working. Using a middleware layer is a time-saving and labor-saving approach, but it comes at the cost of performance. You can't have your cake and eat it too.

Han Luyi suddenly asked, "Was it you who led the team to Dragon Gate Semiconductor for the adaptation?"

"They brought people over and worked with us on the compatibility test in the lab on the fourth floor."

Han Luyi raised an eyebrow. He and Boss Zhang of Dragon Gate Semiconductor only discussed exchanging priority purchase rights for adaptation cooperation. The rest was discussed by the team members themselves. He didn't expect the work to be done in the Origin Building.

With such a large operation, it's impossible to cover everything perfectly.

How long will it take to further improve this adaptation efficiency?

Jiang Songran shook his head. "We only completed the initial adaptation urgently so that these cards could be put into use as soon as possible. It's not cost-effective to continue working with Lungmen to improve efficiency now. It would be better to let the team connect with other manufacturers."

That makes sense, but this would create a significant gap in Computing Power between clusters of different types of Computing Card.

"Can we use mixed calls?" Han Luyi asked.

"This is a direction our team is researching, but there are still some technical challenges we haven't overcome," Jiang Songran replied.

Han Luyi pulled out a chair and sat down. "Tell me more."

Jiang Songran a moment, then asked, "More?"

Han Luyi said, "Just imagine I'm a 5-year-old child and try your best to explain this to me clearly."

Jiang Songran thought to himself, "Boss Han really came up with this on a whim."

Why are we suddenly using the Feynman Technique?

If you can't explain a piece of professional knowledge to a 5-year-old child, it means you don't fully understand it.

"I'll give it a try," Jiang Songran said. "Now, the three different types of inference cards are like three trucks of different capacities, and the different inference tasks are like cargo of different sizes to be transported. Some tasks are long and complex and can only run on the L100, while others are simpler and more direct, consuming less Computing Power, and can run on the J590 or even the Dragon Soar 2000."

"The problem we're facing now is that before a task comes in, or even as part of a Long-Range Tasks, we don't know whether it's a simple task or a difficult task during execution."

Han Luyi shook his head inwardly; Jiang Songran gone back to his old ways after talking for a while.

The metaphor of trucks and cargo was used in just one sentence.

Jiang Songran continued to explain, "Because we don't know whether the task is big or small, the scheduling logic is difficult to write; sometimes we can only try it out by trial and error."

Hearing this, Han Luyi asked, "So what direction are you heading now?"

"We're considering whether we can break the task down into smaller pieces, so we don't have to worry about whether it can fit on any card."

Han Luyi sat there and glanced at him.

"Wouldn't this approach limit the complexity of the tasks that can be performed?"

This is tantamount to crippling oneself; overly complex tasks are abandoned, leaving L100 only capable of running the same tasks as Dragon Soar 2000.

Jiang Songran said, "There is indeed such a flaw, so we are still exploring it."

Jiang Songran and Han Luyi have backgrounds in systems engineering, and they know that just because something is realized in a laboratory doesn't mean it can be replicated in reality.

The gap between the laboratory and large-scale application lies in engineering capabilities.

Han Luyi and said, "The current technical challenges all lie in Task Orchestration. Do these scheduling rules have to be written in advance?"

Jiang Songran startled by these words.

Han Luyi continued, "When we were working on power projects before, we had Glutinous Rice Ball handle the dispatch center. Could we also have Glutinous Rice Ball handle the dispatch for this task allocation?"

Jiang Songran carefully considered Han Luyi's words.

Dynamic scheduling is indeed another direction they are exploring, but directly involving Glutinous Rice Ball may not be mature enough, because one of the current problems with a is its slow speed and Concurrency. However, as the number of calls to the Tangyuan Model increases, this scheduling decision may be made tens of thousands or even hundreds of thousands of times per second.

"Boss Han, the scheduling logic is quite complex. If we just let the model handle it, the concurrency might not be high enough," Jiang Songran replied.

While looking at the real-time data on the screen, Han Luyi said, "I'm not an expert on Computing Power infrastructure. If I'm wrong, you can correct me. The problem now is that the complexity of the task is hard to judge in advance, right? What if we don't give the scheduling power to Glutinous Rice Ball, but just let it judge the task complexity?"

Yes!

Jiang Songran previously gotten stuck on a minor issue, thinking that a scheduling layer with hard-coded rules had good performance but wasn't intelligent enough; and that completely handing over the scheduling layer to Glutinous Rice Ball would result in poor performance.

But what if we don't let Glutinous Rice Ball handle the scheduling logic directly?

What if we use the task difficulty assessment as the Starting Point of the call chain?

Jiang Songran's mind opened up, and new ideas suddenly emerged.

"Boss Han, we hadn't considered this direction before, but now that you've brought it up, I think it's worth a try," Jiang Songran said. "It's indeed unrealistic to directly have Glutinous Rice Ball handle the scheduling, but we can have Glutinous Rice Ball predict the complexity of the overall task flow in the first task, and then pass this judgment to the calling layer at every step in the call chain."

Han Luyi; having a direction was good enough.

"In that case, tasks are categorized, and Computing Power is also categorized, so the efficiency of mobilizing existing Computing Power can be further improved--" Jiang Songran stopped halfway through his sentence.

"How much will it improve?" Han Luyi asked.

Jiang Songran gave a wry smile, "I don't have the data right now. We'll report back to you after we finish testing."

"Okay, I'll leave the Computing Power to you. Report back to me when you make any progress," Han Luyi said.

Then he got up and went outside.

I checked my watch; it was time to leave work.

Haicheng Siheng Law Firm.

Some of the Lawyer and Lawyer sitting in the office area have already started packing up to go home.

Everyone has had a good time these past two months.

Because the director is working less overtime, everyone else doesn't have to sit in the office and work overtime anymore.

Of course, this doesn't mean there will be less work, it just means I can bring my computer home to work.

After all, the client is waiting there, and the court date is already set.

But don't underestimate this benefit; it's also called flexible hours, or remote work.

As for why Director Gu has suddenly stopped working overtime recently, there are rumors in the institute that she's in a relationship.

The rumor was quite detailed and convincing, but no one knew where it came from. Some people in the institute believed it, while others didn't.

Even those who don't believe it can understand—Director Gu may have long legs and be beautiful, but she's so serious every day; her boyfriend would be terrified.

Some countered, "You see how serious the director is usually? That's how she is at work. She's definitely not like that when she's at home with her boyfriend."

The person who said this was intern Lawyer Xiao Wang Lawyer. The reason she was so confident was because she had sounded out her mentor.

The strange expression on Lawyer Shen's face when Xiao Wang asked about it clearly indicated that Director Gu's love life was in trouble.

She now hopes that her mysterious boyfriend, Director Gu, can "train" him—no, make him a little gentler, so that things will be easier in the future.

Meanwhile, Director Gu, who had received Xiao Wang's "best wishes," was sitting in his office.

Shen Yuwei over to discuss the defense strategy for the trial, glanced at Gu Siyue was working at her desk, and suddenly realized something.

"Why do you look a little unhappy?" Shen Yuwei asked.

In fact Gu Siyue didn't seem unhappy; she was just acting normally.

However, after being colleagues and friends for several years, Shen Yuwei had developed unique observation skills regarding Gu Siyue.

When she's being particularly efficient and focused on her work, that's when she's in a bad mood.

This is a classic example of turning grief into work efficiency.

Shen Yuwei looked at the mountain of folders piled up in front of Gu Siyue; this was at least three days' worth of her own work.

Workaholics are terrifying.

Gu Siyue had just finished finishing the last document in her hand when she heard Shen Yuwei's words, but she did not respond.

Shen Yuwei became interested. "What happened? Did you have a fight with your CEO Boss Han?"

Gu Siyue gave her a disapproving look.

"I'm not unhappy, just a little troubled," Gu Siyue said.

Shen Yuwei put the folder in her hand aside and sat down in the manner of a caring older sister.

Work can wait, but gossip must be heard now.

"What's bothering you? Tell me," Shen Yuwei said. "I'm a relationship expert."

A relationship expert who specializes in helping people get divorced.

Gu Siyue was silent for a moment, then said, "Luyi has been looking at houses recently."

Shen Yuwei's eyes widened slightly upon hearing this.

Her occupational hazard kicked in, and the first thing she thought of was the legal disputes involved.

"Is a house bought before marriage considered premarital property? Was it paid in full or mortgaged? If it's mortgaged, the increase in value of the property due to joint mortgage payments is eligible for division," Shen Yuwei said.

Previously, I handled mostly post-divorce disputes, so suddenly having the opportunity to plan pre-marital assets is quite exciting.

Gu Siyue frowned; this wasn't her concern at all.

Shen Yuwei she had guessed wrong and asked again, "Okay, which house is he looking at? Huangpu? Jing'an?"

Gu Siyue got up and walked to the window, looking outside.

Shen Yuwei didn't know what she meant, but she followed her and stood beside her, looking outside.

The window faces the Huangpu River, but the building is not tall, so you can only see a little bit of the river view peeking through the gaps between the buildings.

Shen Yuwei followed Gu Siyue's gaze and suddenly realized something.

She swallowed hard.

"You mean, your Boss Han is looking at houses there?" Shen Yuwei held up one finger.

Gu Siyue nodded, her brows furrowing.

"Lawyer Gu, I was wrong. I said earlier that the gift you gave me was too expensive, but I spoke too soon."

A few sets of high-end menswear and a few watches—that's not much money at all.

Compared to a house, it's really nothing.

"So, what's troubling you is how to get him to add your name to the deed?" Shen Yuwei on her hand. "It's indeed difficult to ask, but don't worry, it's very common nowadays. Unless there's a clear written agreement, the court generally won't recognize it as a 'gift intended for marriage,' so it doesn't need to be returned."

This is a little trick from a seasoned divorce Lawyer.

Because of the customary backing, cash dowries are generally considered "gifts given with the intention of marriage" even without a written agreement. If the marriage does not ultimately take place, they can be partially or even fully returned.

However, houses are different. Unless there is a very formal and explicit accompanying agreement, they are generally considered to be voluntary gifts.

Therefore, her advice to male clients has always been that it's better to pay a cash dowry than to add their names to a property purchased before marriage. Otherwise, from the moment the property is registered, it's considered an unconditional gift, and it's almost impossible to get it back.

Of course, now that the role has changed to Gu Siyue, Shen Yuwei's advice has naturally reversed.

After all, one's perspective is determined by one's position; she and Gu Siyue are best friends, but she and Han Luyi are not.

Gu Siyue glanced at Shen Yuwei and sighed.

Her displeasure was no longer subtle; she exuded a sense of sadness.

It's as if the usually decisive and efficient iron lady has suddenly turned into a melancholy young girl.

This was the first time Shen Yuwei had seen Gu Siyue like this. Before, no matter how difficult the case or how much work pressure she faced, she seemed to be invincible.

Shen Yuwei wondered if her joke had backfired, so she quickly shut up.

At this moment Gu Siyue slowly spoke:

But I—don't want to get married.

End of Chapter
Prev
NovelSummaries
Next

Explore the wiki

Han Luyi
Su Niannian
Gu Siyue

Comments

No comments yet.
AboutStatsAPIChangelogStatusContactPrivacyTermsDMCA
© 2026 RandomTranslator·v3.26·RSS