Reborn in 2011, 985 missing from two copies

Chapter 667 Graphics cards are put into mass production. Kill him while he is ill!

In the past, Nvidia’s cost-effective card GTX 970 suffered a serious setback and was boycotted due to the video memory gate incident, which led to AMD’s huge sales!

This time, there is no escape from such a fate!

Even Wang Yi planned to test the performance after Xingyi Technology's RTX 960 was successfully taped out.

If the performance is strong, it can be directly renamed RTX 970 to compete with Nvidia's failed GTX 970!

Take advantage of his illness and kill him!
The GTX 970 can’t beat it without any crash, but after it crashes, there is no pressure.

Taking advantage of the failed GTX 970 to take the lead is undoubtedly the best chance for the Xingyi graphics card to become the king!
Not to be missed.

Wang Yi quickly made arrangements and asked the graphics card department to continue their efforts.

The Nvidia video memory scandal is a once-in-a-lifetime opportunity. If we miss it, it would be unforgivable!

"Also, the optimization and compatibility of mainstream games are also issues." Wang Yi thought of this and asked, "You all have deployments, right?"

The CPU compatibility is actually okay, but the graphics card compatibility is really a problem.

Williams nodded and said, "Chairman, we have always attached great importance to compatibility issues. We have already established a compatibility optimization department long ago. There is not only a CPU optimization department, but also a graphics card optimization department. Of course, our graphics cards have not yet been taped out, and the main task of the graphics card optimization department is to optimize the core graphics."

"After the Xingyi CPU was taped out, the graphics card optimization department and the CPU optimization department started the optimization work around the clock. Both teams worked in shifts to optimize. We performed all-round optimization on mainstream applications and games, and solved every bug we found."

"That's why our CPU compatibility is very good, and the integrated graphics compatibility is also very good."

"As for the compatibility of graphics cards, it is a little more difficult. However, in the past few months of optimizing the integrated graphics, our team has accumulated a lot of experience. We are familiar with the characteristics and performance of mainstream games, and have established corresponding databases based on the same or similar games."

"After the graphics cards are taped out, we can quickly advance the optimization of the graphics cards based on the information in these databases and the experience of core graphics optimization."

Wang Yi nodded: "Very good, this arrangement is very appropriate. However, the graphics card will be taped out in July, optimization will begin, and it will be launched in September. Two months is not enough, right?"

"Yes, Chairman, two months is indeed a tight time."

"Then hire more people and use the number of people to make up for the time gap!" Wang Yi said decisively: "Especially experienced talents, just poach them, no matter the cost."

"Okay, Chairman." Williams smiled, a familiar operation.

What Mr. Wang likes to do most is to invest money in research and development and to recruit talents!
Under this kind of money offensive, talent is not a problem; under this kind of money + talent offensive, time is not a problem either.

A team of 60 people is responsible for optimizing a graphics card driver. One person is responsible for many games and software. It may take half a year to optimize it.

But if a team of 300 people is responsible for graphics card optimization, with multiple people responsible for one game, it would take about a month.

Graphics card optimization is designed for different games, with troops deployed in batches. It is completely possible to use a rapid-fire mode and push multiple lines simultaneously.

Currently, Nvidia's graphics card driver team has a total of only a few thousand people, and there are only three or four hundred people specifically responsible for compatibility optimization.

NVIDIA releases five or six different graphics cards a year, and the optimization team assigned to each graphics card has only fifty or sixty people.

Xingyi Semiconductor has recruited a team of 300 people, who are currently only responsible for the optimization of one graphics card. Naturally, their efficiency is several times higher than that of Nvidia, where 50 to 60 people are responsible for one graphics card.

In addition, Nvidia’s latest graphics cards, GTX 970 and 980, were only released in September, and their optimization was mediocre in the first few months.

If they had optimized well, the memory gate incident would not have occurred.

In this way, with the full optimization efforts of hundreds of people, Xingyi Technology's RTX 960 may not necessarily fall behind, and may even have better compatibility!
Nvidia headquarters.

Huang is in a good mood recently: "Wang Yi is so good at playing that he actually went to make computer processors. Hahahaha, that's great, great! Intel will have a headache next!"

The subordinates agreed very much: "Yes, if Wang Yi changes direction and focuses on computer CPUs to abuse Intel, we will be relieved. It is a good opportunity for Intel to experience the fear of being hit by Xingyi Technology and being dominated by Xingyi Technology, hahahaha!"

"That's right. How can Intel be spared from what NVIDIA has gone through? They have to go through it too. I just didn't expect that Xingyi Technology is so powerful. They just started making computer processors, but they beat I3, I5 and AMD, especially the integrated graphics card, which is so powerful that it doesn't even look like an integrated graphics card. But it's okay. It's Intel that should have a headache, hahaha."

Old Huang gloated and felt refreshed.

Previously, Nvidia entered the mobile chip market in order to expand its business.

In 2012, NVIDIA won the favor of Xingyi Technology with the world's first quad-core mobile chip, Tegra 3. The sales of Tegra 3 exploded, and NVIDIA made a lot of money. At one point, it felt that it had made it big and wanted to make a big splash in the field of mobile phone chips.

But the good times did not last long. Soon, Xingyi Technology launched its self-developed Kunpeng chip, which not only replaced the market of Nvidia chips, but also snatched away many of Nvidia's chip customers!

The result is predictable. Last year's Integration 4 didn't sell very many units and the company suffered huge losses!

It can be said that whatever money that was earned by Integration 3 two years ago, was lost by Integration 4 last year!

After a comprehensive calculation, not only did I not make any money, I also suffered heavy losses.

In addition, the baseband problem was really unsolvable, so Huang simply gave up on mobile phone chips and turned to making car chips, tablet chips, and even TV chips that did not require baseband.

This year, they launched the tegra K1!

I thought this track would be easier to get into, but unexpectedly, Xingyi Semiconductor immediately launched Kunpeng 902C, Kunpeng 706C, Kunpeng 508C, chips for tablets, TVs, and car machines...

Not only does it have a powerful configuration, it also has lower power consumption, better optimization, better compatibility, and a 64-bit architecture. It covers all ranges of high, medium and low-end, allowing customers to choose according to their needs. It is simply not comparable to a single NVIDIA chip.

The result is predictable, Nvidia's tegra K1 also failed.

It went into mass production in the first half of this year, but by July, not a single one had been sold!
This makes Lao Huang very annoyed!

"How's things going with Audi? Do you want to use our K1?" Old Huang couldn't help but ask, feeling melancholy when he thought about it.

The subordinate frowned and said, "Chairman, Audi originally planned to use our K1 chip as the car machine chip, but when they heard that Xingyi Technology was about to launch the Kunpeng 910 SOC and Kunpeng 910C, they changed their minds."

"Fuck!" Old Huang couldn't help but curse. "The Kunpeng 910 SOC is designed for mobile phones, and the Kunpeng 910C doesn't have a baseband, so it's just right for car computers. So, is Audi planning to give up our tegra K1 and choose Xingyi Technology's Kunpeng 910C?"

"It's possible, but there is another possibility. They plan to directly use the Kunpeng 910 SOC with baseband!"

“???” Old Huang was confused: “Does a car chip need a baseband?”

"Chairman, it is a trend that luxury cars are equipped with communication functions as standard. For example, in the event of a traffic accident, emergency SOS service, fault repair, etc., baseband communication is required."

"Mercedes-Benz, for example, has all its fuel-powered vehicles equipped with these services. If something happens and you can't find your phone, it's OK. You can just press SOS to ask for help from the Mercedes-Benz service center. The service center will immediately locate the vehicle and call the police for rescue."

"As for trams, future car navigation and so on, they will also require baseband online navigation."

"With our tegra K1, there is no baseband, so we can only share it through mobile data and Bluetooth. But with the Kunpeng 910 SOC, we can directly get a data card for online navigation and entertainment..."

"Fuck! So, future car computers cannot do without baseband? Then our tegra K1 is also useless?"

"That's the theory."

"No wonder Audi, which had said it would place an order, is now hesitating! Fuck! Fuck!"

Old Huang cursed: "We still can't get around the baseband!"

"Doesn't that mean that our tegra K1 can only be used as a tablet?"

"That's right, Chairman. But the problem is that most tablet manufacturers have directly purchased Kunpeng 902C, Kunpeng 706C, Kunpeng 508C..."

The gloating that Lao Huang had felt before was gone, leaving only the twitching corners of his mouth: “!!!”

"Forget it, then let's learn from Xingyi Technology, develop NVIDIA tablets, and make our own tablets!"

"Okay, Chairman, this project is progressing smoothly. NVIDIA Shield Tablet is expected to be released in the third quarter and available in the fourth quarter."

"Well, with the power of the tegra K1 desktop-level GPU, the NVIDIA Shield Tablet can crush the Xingyi Tablet 2nd and 2Pro generations, and it will definitely be a global best-seller!"

"That's for sure. Although the Xingyi Tablet 2 Pro has good performance, it is still far behind the Tegra K1. As for their third-generation tablet, that will be next year. Our NVIDIA Shield Tablet is bound to be a hot seller."

"Okay, make a prototype as soon as possible. If there are no problems, we will start mass production as soon as possible and strive to release it in September and put it on the market in October!"

"Yes!"

"We lost so badly before. This time, I must win a game against Xingyi Technology! Win a game against Wang Yi!"

Old Huang looked into the distance with a burning gaze.

But what Huang didn't know was that just when Nvidia was defeated repeatedly and was preparing to turn its attention to the tablet market and launch a surprise attack on the Xingyi Tablet 2, its own graphics card base was also in danger!

After the successful tape-out of the X7 4800 processor, Xingyi Semiconductor immediately started the tape-out of its self-developed graphics card RTX 960.

After several days of preparation, tape-out officially began.

Without any surprise, the RTX 960 was successfully taped out and a series of tests began immediately.

On the day when the test results came out, Wang Yidu personally came to the Xingyi Semiconductor R&D Center to wait.

The current R&D plant has been separated into an independent plant, which is used specifically for the tape-out of new chips.

Whether it is mobile phone chips or computer processors and graphics cards, they are all taped out here.

Yu Longxing, director of Xingyi Graphics Card Department, came out with the test results:
"Chairman, all the test results are out!"

"The yield rate of RTX 960 has reached 85%!"

"85%!" Wang Yi was in a good mood. "Very good. The tape-out yield is 85%. The official mass production will hopefully reach over 90%. Very good! How is the performance?"

"Chairman, Nvidia's GTX 960 and 970 have not yet been released, and we don't know their specific performance. We can only compare them with the high-end dessert gaming card GTX 770 and the high-end flagship gaming card GTX 780 released last year."

"Since our RTX 960 graphics memory uses the more powerful 4G GDDR5, it is ahead of Nvidia's GTX 770's 2G GDDR5 and GTX 780's 3G GDDR5, so the performance is still very outstanding."

"According to our tests, our RTX 960 performance is 770%+ stronger than the GTX 10 in many games, but it is still a little behind the GTX 780, with a gap of about 15%."

"Of course, it doesn't mean that our RTX 960 is inferior to GTX 780. It's just that our graphics card has not been optimized yet. 780 is the flagship high-end graphics card released by NVIDIA last year, and it has been optimized very well." Wang Yi nodded: "So, if our RTX 960 is optimized, it will be about 770% stronger than GTX 25, and it will be on par with GTX 780?"

"That's right, Chairman. Once the optimization is done and compatibility is in place, there will be no problem in increasing it by 15%. Even 20% is possible," Yu Longxing responded.

"So, our RTX 960 actually performs as well as Nvidia's last year's flagship gaming card, the GTX 780?"

"Yes." Yu Longxing said: "According to Nvidia's knife skills, the GTX 960 that will be launched in January next year will have performance close to the previous generation of high-end dessert gaming card GTX 770. And the GTX 970 released in September will be 780-10% ahead of the previous generation GTX 20 in performance!"

"Our RTX 960 and GTX780 have similar performance, which means it is 960-20% better than GTX 25, but 970%-10% behind GTX 20!"

"As for how far behind, it depends on Huang's GTX 970's knife skills!"

"Very good." Wang Yi was very satisfied with this: "RTX 960 was originally positioned as a mid-range graphics card to compete with GTX 960. Now it is so much ahead of GTX 960 and only slightly behind GTX 970. It is very good, very good."

"This kind of performance can definitely be considered a high-end dessert graphics card!"

Yu Longxing nodded: "Compared with the 780, it is slightly inferior to the 970 in performance. As long as we optimize it properly, it can be used as a dessert graphics card."

Wang Yi agreed very much. He knew that in terms of the performance of RTX 960, it was not just slightly inferior to 970, but was on par with 970!

There is no other reason. In theory, 970 is 960%-10% stronger than RTX 20, but there is a serious memory problem, and the actual performance is probably not as good as RTX 960.

In this way, RTX 960 can completely replace 970 and become this year's annual high-end dessert graphics card.

"Chairman, the performance is clear. If we want to optimize it, we have hundreds of people. It is not a problem to optimize this graphics card in two months. How do we set the price?" Yu Longxing asked, "Should we price it according to the mid-range graphics card or the high-end dessert graphics card?"

The graphics card market these days has a very clear situation.

High-end flagship graphics card, priced at 4000-5000.

High-end dessert graphics card, priced at 2300-3000.

Mid-range mainstream graphics card, priced at 1500-2000.

Mid-range low-power graphics card, priced at 1000-1500.

Entry-level independent graphics card, priced at 700-1000.

Whether it is NVIDIA or AMD, the positioning and price are similar.

Of course, NVIDIA also has the GTX Titan series of ultra-flagship graphics cards for scientific research and AI training, priced at 8000-10000. AMD currently has no comparable products.

For example, for Xingyi AI training, we purchased some GTX Titan graphics cards.

However, there is a high probability that you won’t need to purchase the GTX Titan X next year, as Xingyi Technology’s self-developed AI graphics card will be almost released by then!

According to the market conditions, the performance of RTX 960 is comparable to that of the high-end dessert graphics card GTX 970, and the price should be between 2300-3000.

But as the first independent graphics card of Xingyi Semiconductor, if it is sold at such a price, its paper parameters are 970%-10% lower than that of 20, which is obviously not suitable and will not sell well.

After all, after the failure of 970, it is almost the same as RTX 960, with similar performance and similar price, users still prefer the old giant Nvidia.

But if it is sold as a mid-range mainstream graphics card with a price of 1500-2000, then it is really a bit of a loss!
Soon, Wang Yi made a decision: "For the sake of sales, RTX 960 will be renamed RTX 970, and advertised as comparable to GTX 970, but with a lower price, let's sell it for 2000-2500 yuan."

"It's 500 more expensive than mid-range mainstream cards and 500 cheaper than high-end dessert cards, which is somewhere in between. That's good." Yu Longxing agreed: "And with this pricing, our profit is much higher than NVIDIA, Colorful, Asus, MSI, Gigabyte, etc."

"Yes, Xingyi Semiconductor develops and produces everything from chips to heat dissipation PCBs in-house, so naturally its profits are higher than those of its competitors."

In addition to the GPU chip, the graphics card also has a cooling module and PCB (equivalent to the motherboard).

Nvidia mainly designs GPU chips, and then TSMC manufactures GPU chips on a foundry basis.

Only some public version graphics cards are launched by NVIDIA.

Other graphics cards are from companies such as Colorful, ASUS, MSI, and Gigabyte. They purchase NVIDIA's GPU chips and then make or customize their own cooling modules and PCBs. This resulted in graphics cards from various brands such as ASUS, MSI, and Gigabyte.

The core GPUs are exactly the same, all are purchased NVIDIA GPU chips, all are designed by NVIDIA and manufactured by TSMC. The difference is the PCB and cooling module, which leads to different performance and the selling price differs by several hundred yuan.

But Xingyi Semiconductor is different. It has just started making graphics cards and no other competitors have purchased them yet.

You have to do it all by yourself.

From the research and development and design of GPU chips, to production, to heat dissipation modules and PCB motherboards, etc., Xingyi Semiconductor handles all of them by itself, with full-scale self-research and production.

Even sales and after-sales are the responsibility of Xingyi Technology itself.

The disadvantage is that you have to do everything yourself, and there are not so many third-party brands.

The advantage is that you can do everything by yourself, including optimization, production, upgrading, after-sales service, etc., which is more convenient and the profit is more substantial.

After all, our competitors are divided into three companies from R&D to production and assembly, and the profits must also be divided into three parts.

The Xingyi graphics cards are all manufactured by themselves, and all the profits are theirs.

The meat was rotten in the pot!
Coupled with the strong heat dissipation strength and R&D capabilities accumulated by Xingyi Technology in the mobile phone field, there is no pressure in graphics card heat dissipation. The corresponding technologies and products have been produced long ago.

The PCB motherboard has also been prepared.

Just waiting for the most core GPU chip to be taped out and put into mass production.

Once the GPU chip is successfully mass-produced, you can quickly start assembling the Star RTX 970 graphics card!

In the following days, Xingyi Wafer Factory immediately began debugging and mass production of RTX 970 chips.

The batch of samples that were taped out were directly assembled into graphics cards and taken away by the graphics card optimization department for a series of optimization tests.

Time waits for no one, and compatibility optimization began directly after tape-out.

The entire graphics card optimization department has entered a critical period, working in two shifts day and night, and will not stop work even if people are stopped!
Of course, the treatment is also very good.

All working hours beyond 8 hours will be considered overtime and will be paid double the salary.

Other big companies work 996, 12 hours a day, and they don’t get any overtime pay.

But at Xingyi Technology, 996 is 12 hours a day, which is directly counted as 4 hours of overtime, and double pay is eight hours, which is directly counted as two days' salary!
996 work per month, and you get two months’ salary directly!

Under such a system, many employees have no objection to working overtime and are even happy to do so.

After all, 996 is the norm in the Internet industry, and working eight hours a day without overtime is a dream. Xingyi Technology pays two days' salary, so you can get two months' salary in one month, which is really great.

Many people even want to work overtime every day and earn 24 months' salary a year, plus the year-end bonus, which is directly 30 months' salary!
That would be really cool.

But unfortunately, Xingyi Technology does not encourage overtime work, and overtime work must be applied for and approved.

Except for special periods and special tasks, overtime work can only be done after approval.

It is for this reason that overtime work has become a good job at Xingyi Technology!
A great job where you can earn two days’ salary in one day!
And now, this good thing falls on the shoulders of the CPU optimization department and the graphics card optimization department!
Other departments can only envy.

Of course, the Xingyi Superbook S1 pro team, desktop team, Xingyi gaming notebook team, and all-in-one team also started working overtime.

These departments are under time pressure and have heavy tasks, so they can only work overtime 996 and get double pay!

"The production capacity is sufficient, right? Even a little excessive?" Wang Yi asked again.

"Yes, Chairman. The second phase of the M2 wafer factory in Shanghai has been fully put into production. The monthly production capacity has increased by another 5 wafers, and the total monthly production capacity has reached 30 wafers, with a total monthly production capacity of up to 1.8 million chips!"

"Except for a few with 40-nanometer production capacity, the rest are mostly 28-nanometer production capacity. Such abundant production capacity fully meets the current demand and is in excess."

Fortunately, mobile phone chips were sold in large quantities to third-party brands, which increased a lot of chip orders, and chip manufacturing for Huawei also consumed a lot of chip production capacity.

Now with the addition of CPU and graphics card, the production capacity is still abundant and even overflowing!

Xingyi Semiconductor currently produces 1.8 million chips per month, and its production capacity cannot be fully utilized.

Even small chips such as fast charging chips, sensor chips, and smart home chips are no longer outsourced to TSMC and SMIC. Instead, they are all produced in-house.

There is no way, there is too much production capacity.

Even so, some production capacity is still idle.

"How much idle production capacity is there at present?" Wang Yi asked.

"Chairman, the second phase of the M2 wafer factory is still in the process of ramping up its production capacity, so there is not much vacancy. But by next month, when the capacity ramp-up of the M2 wafer factory is completed, 30% of the total capacity will be vacant, which means 5500 million chips per month."

"So many!" Wang Yi frowned.

At present, most major wafer fabs have vacant production capacity, but the vacancy rate is between 10% and 35%.

For example, Samsung Semiconductor had a vacancy rate of 25%-30% in the first half of the year. In the second half of the year, Apple's A8 chips were outsourced to TSMC in large quantities, and Samsung's situation became even worse, with a vacancy rate of 30%-40%. The annual vacancy rate was as high as 30%-35%!

TSMC’s average vacancy rate this year is only 10%-20%!

Last year, Xingyi Semiconductor had insufficient production capacity, and the vacancy rate was around 5% in the first half of this year, both of which are very ideal conditions.

Now, with new wafer fabs coming into production one after another, the vacancy rate is about to reach 30%!

This is a very scary thing.

"We can't have such a high vacancy rate. We need to increase the production of Kunpeng 910, CPU, and graphics cards as soon as possible. We will produce them first and store them in the warehouse. We can't have too many vacancies on the production line."

Wang Yi decisively said: "In addition, we should actively expand third-party major customers, provide sufficient 28-nanometer high-end process capacity, and grab orders. We should also contact AMD, offer a good price, and grab some of AMD's orders!"

(End of this chapter)