Rebirth 08: Rise from copycat phones

Chapter 433 Anyone can develop artificial intelligence

With Zhiyun Group releasing a number of chip products in late October and early November, especially the relatively important APO4600 graphics card, and also developing the GTAI2 open source model.

This can be said to have further promoted the development of the artificial intelligence industry chain on a global scale.

Imagine an ordinary high-tech enterprise, which has little capital but has the GTAI2 open source model in its left hand and the APO4500/4600 graphics card in its right hand.

what can it do
It can immediately deploy the so-called self-developed artificial intelligence model based on GTAI2, and then apply it to its own series of software services.

All they have to do is to buy APO series graphics cards from Zhiyun Group and then train based on their own data.

High-performance graphics cards are not only used to train artificial intelligence, but are also necessary hardware for running artificial intelligence.

To provide artificial intelligence services to a large number of customers, the computing power required is enormous!

Once the DS in the original time and space was released, it attracted a large number of users around the world to download and use it, but many users encountered a problem, that is, it would encounter lag, network busy and other situations.

This is because there are too many users, and DS’s computing resources are limited, and it is unable to supply the computing power requests of such a large number of users at the same time.

Every question asked by the user requires a certain amount of computing power resources... This is a very critical issue in the artificial intelligence industry, that is, computing power resources!

Only with sufficient computing resources can we provide artificial intelligence services to users...

Unfortunately, the GTAI2 model open sourced by Zhiyun Group only supports graphics cards from Zhiyun Group... and does not support graphics cards from other companies.

This means that if other companies want to use the GTAI2 model, they must simultaneously use the graphics cards of Zhiyun Group.

For small enterprises deploying GTAI2 for small-scale internal applications, according to the configuration plan announced by Zhiyun Group, it is recommended to use an 8*2 APO4600 graphics card server.

A single server is deployed with eight APO4600 graphics cards, which is also the recommended single-server interconnection solution for the APO4600 graphics card, and two such servers are used at the same time.

Then you can use the full-blooded version of the GTAI2 open source model to obtain complete logical reasoning capabilities... However, this level of computing power can only maintain basic operations and provide computing power requests to a small number of users.

If the number of users is slightly larger and the questions asked are more complicated, it will take longer for the server to give users feedback on their answers.

The reference number of parallel users for this configuration recommended by Zhiyun Group is twenty people... Only twenty people can use it at the same time, and if there are more than twenty people, it will become very laggy.

However, even a data center GPU server consisting of 8*2 APO4600 graphics cards is very expensive. Just these graphics cards cost about million... and this does not include other expenses.

In addition to a graphics card, a GPU server also requires a CPU, memory, and flash memory, and without exception, they all require the current top-level configuration.

Zhiyun Group recommends using the WZ260CPU, which is a flagship server CPU with multiple GPUs in parallel and based on the X86 instruction set, launched to the market by Weizhi Technology Co., Ltd., a subsidiary of Zhiyun Group. This CPU has powerful performance but is also very expensive.

It is also recommended to use the top-level ZC68 enterprise-level DDR4 high-speed memory released this year by Zhiyun Storage Technology, which is also very expensive.

Flash memory is cheaper, but it is also very good enterprise-grade flash memory.

This is the configuration recommended by Zhiyun Group when it open-sourced GTA2...Except for the GPU which is special and must use the graphics card from Zhiyun Group, the other CPUs, memory, and flash memory can all be replaced with other products of the same level from other manufacturers based on the X86 instruction set.

Zhiyun Cloud Computing Technology Co., Ltd., a subsidiary of Zhiyun Group, also provides finished servers with the above standard configuration to customers in need. The price of a single server is about two million yuan.

The above recommended configurations refer to small enterprises or scientific research institutions with needs.

In addition, micro-enterprises or individual users with high-performance needs can also try to deploy a low-power version, or even use fewer APO graphics cards or older graphics cards, but no matter what, it will cost hundreds of thousands to do it.

Then, ordinary individuals and technology enthusiasts can actually have a basic experience... because the GTAI2 open source model also released a retarded version based on the X series graphics cards. Of course, this experience is just to satisfy the interests of some technology enthusiasts. It can be deployed, but it is useless after deployment.

Ask a question, and GTAI2 will have to think for several minutes or even longer before giving you the answer... The reason why the retarded version is open to individuals for experience is purely for advertising and to create a public opinion propaganda effect!
For example, if you ask those technology bloggers to spend hundreds of thousands or even millions of dollars to deploy a full-blooded version, that is unrealistic. But if you ask them to use hardware worth tens of thousands of dollars to deploy a retarded version of GTAI2, then there will be no problem.

Deploy it first, then experience it, publish an article to brag about it to attract traffic, and then delete the model directly... GTAI2 takes up too much computing power. After deployment, experience it a little, satisfy some curiosity, and then delete it. Otherwise, this broken computer will not be able to do anything else.

But even so, Zhiyun Group has achieved its publicity purpose.

----

In fact, for individual users, micro-enterprises, or even some small business users who do not have any special confidentiality requirements and only need to use the service for their own use or to provide services to a small number of customers, there is actually no need to deploy the GTAI2 model, even if it is open source.

Why... Because there is a Yun AI next to it that provides a commercial interface!
What’s more, Yun AI is easier to use and the price of the interface is not expensive, which is much more cost-effective than deploying GTAI2 yourself.

However, for some large enterprises or enterprises with confidentiality requirements, it is necessary to deploy GTAI2 on their own.

Basically, except for Yihai Technology, all major high-tech companies will deploy GTAI2 on their own, even though everyone knows that GTAI2 is far inferior to Yun AI.

Because no one wants to tie their core business to Zhiyun's Yun AI.

Regardless of whether they are several large domestic Internet companies or some well-known foreign Internet companies, they will use their own artificial intelligence models whenever possible... The path of self-research is too difficult, so using the GTA2 open source model can barely make do.

They can’t all follow Yihai Technology’s example and just connect their entire core services to Yun AI... They also want to do that, but the problem is that their boss is not Xu Shenxue.

Yihai Technology is Xu Shenxue's company, so there is no pressure at all to use Yun AI.

But other Internet companies are not in the same position… they don’t have the courage.

This is no different from putting your own neck in front of Zhiyun's machete... From now on, life or death depends on the decision of Zhiyun Group.

No one would rely on the life and death of their company to the whims of others.

If one day Zhiyun Group suddenly has a brainwave and directly disconnects the Yun AI interface of their core service from the Internet, they will be doomed in a matter of minutes.

Therefore, even though it is expensive to deploy GTAI2 locally and the performance of GTAI2 is far inferior to Yun Ai, you still have to deploy it yourself!
These large Internet companies or similar large-scale high-tech companies need enormous computing power to deploy GTAI2, and they are also the core customer group of Zhiyun Group's APO series graphics cards.

A data center consisting of tens of thousands of graphics cards is the minimum standard for them. For slightly larger Internet companies, such as some well-known large high-tech companies, the demand for APO graphics cards is in the hundreds of thousands, otherwise the computing power will not be enough.

This is a huge hardware expense. The APO4600 graphics card is priced at yuan each. graphics cards would cost billion yuan, and graphics cards would cost billion yuan.

If a large corporate customer purchases 100,000 graphics cards, ten similar large corporate customers can provide an order for one million APO series graphics cards. These one million APO series graphics cards can bring 150 billion in sales to Zhiyun Group.

This is also the core purpose of Zhiyun Group's open source GTAI2: I will directly give you the GTAI2 open source model, and you can buy a graphics card from me!
Xu Shenxue doesn’t know whether artificial intelligence, especially the generative AI that these Internet companies are playing with, can make money, but he is definitely making a lot of money selling graphics cards... The gross profit margin of the APO series graphics cards has always been very exaggerated, and by the time of the APO4600 graphics card, its gross profit margin has reached %!

The APO series graphics cards are the most profitable products of Zhiyun Group, no doubt about it!

The fact that the APO graphics card can win the title of the most profitable product in Zhiyun Group, which has always been known for its high gross profit margins, shows how crazy it is.

This is mainly because the APO series graphics cards are out of stock all year round, and the supply exceeds the demand, resulting in high prices.

The computing chips under Zhiyun Group are all produced by Zhiyun Microelectronics. During the production process, the processing of the chip itself is not a problem, especially since the current 12-nanometer process has a large production capacity, which is more than enough for the use of computing chips.

What limits the production capacity of computing chips is mainly packaging... to be more precise, CoWoS packaging/3D packaging.

Before October this year, Zhiyun Microelectronics' 3D packaging capacity was only pieces, most of which were supplied to the AI ​​series graphics cards, ZY terminal computing chips, and EYQ general terminal computing chips, which are the three core computing chips used by Zhiyun Group.

The remaining production capacity is used to supply APO series graphics cards and PX general terminal computing power platform.

This is also the core reason why these two products have been experiencing perennial insufficient production capacity and tight supply.

The supply is tight, so the price is naturally getting higher day by day!
Even big customers have to queue up to get the goods. Of course, if you don’t want to wait, you can go to Zhiyun Computing Technology Co., Ltd. to purchase server products... There are GT series servers developed specifically for the GTAI2 model.

why?
By the way, we sell server CPUs, memory, and flash memory...

Don’t forget that Zhiyun Group’s semiconductor business also includes server CPUs, enterprise-level memory, and flash memory.

Enterprise-level memory and flash memory are both doing well and selling quite well, especially the enterprise-level flash memory under Zhiyun Group, which is quite strong and occupies more than 50% of the global enterprise-level flash memory market share, which is stronger than the enterprise-level flash memory business of Sixin.

Many well-known domestic and foreign high-tech companies, especially Internet companies, are enterprise-level flash memory customers of Zhiyun Storage, a subsidiary of Zhiyun Group.

However, the situation in the server X86 GPU field is not so good. Before this, only domestic related institutions, special enterprises, Zhiyun Group, and sister enterprises purchased and used it, and there were very few external customers... The ecosystem was not very good, and the price was not much cheaper. Many corporate customers would rather continue to use Intel's server CPUs due to usage inertia, usage costs, development costs and other factors.

Another factor is that Zhiyun Group also has a CPU with a self-developed SOP instruction set. Apart from institutional clients, even its sister companies are reluctant to use it... There are too few ecosystems and it is too troublesome to use.

However, we can’t just watch the SOP instruction set CPU die... Consumer-grade CPUs are difficult to make, and institutional demand is too small, so Zhiyun Group simply built supercomputers and cloud computing data centers based on SOP instruction set server CPUs and sold computing power to the outside world.

Server CPUs with the X86 instruction set cannot be sold, and server CPUs with SOP's self-developed instruction set and consumer-grade CPUs cannot be sold either.

These factors forced Zhiyun Group, which is clearly a smart terminal and semiconductor company, to build a large number of cloud computing data centers, and forced its own cloud computing services to become the second tier in the world... Its data centers are all equipped with its own hardware products, from CPU to memory and flash memory.

Zhiyun Group does this not because it is eyeing the money from cloud computing services, but purely to boost its own server CPUs and, by the way, to support enterprise-level memory and flash memory businesses.

This time, Zhiyun Group provided GT series servers that can run the GTAI2 open source model through Zhiyun Computing Technology Co., Ltd. for similar reasons, hoping to boost its own server CPU business by relying on the momentum brought by GTAI2.

The half-dead server CPU business annoyed many senior executives of Zhiyun Group...it was too bad.

What kind of company is their Zhiyun Group?
It is one of the world's top high-tech companies.

The world's largest market value, the world's largest revenue, the world's largest profit, and the world's largest technology!

No matter what product we make, we aim to achieve the world's best standards... If it fails to reach the top level and is just average, it will be a failure for Zhiyun Group!

Especially semiconductor products with advanced technology are either world-class or top-notch.

In the mobile phone SOC business, it is competing with Qualcomm and Apple, and anyone who comes will have to say "I surrender".

For the independent GPU business, this advantage is even greater. Consumer-grade GPUs account for more than 80 percent of the global market share, and server GPUs account for 99 percent of the global market share.

In the field of terminal computing chips, the PX platform and the EYEQ platform are competitors, and their total market share is an exaggerated 100%, because no other manufacturer has such high-performance computing chips for terminals.

In terms of memory business, it occupies more than 30% of the global market share, which is comparable to that of Samsung, and is ranked first in the world together with Samsung.

In terms of flash memory business, it is already ahead of Fourstar and other storage chip manufacturers in terms of technology, and its market share has slightly surpassed Fourstar. Among them, the enterprise-level high-speed and large-capacity flash memory business is unique in the world, occupying more than half of the global market, mainly because Zhiyun Storage has done very well in the field of enterprise-level flash memory controllers, not just relying on the process advantages of flash memory chips.

Even in the same CPU business, although the X86 consumer-grade CPU is still the third largest in the world, its market share has increased by more than ten percentage points.

Especially in the field of mobile CPUs, driven by the Yun Book series of high-end business notebooks under Zhiyun Group, its brand recognition is quite good. Many domestic notebook manufacturers have successively launched notebooks based on Zhiyun mobile CPUs, and the shipment volumes are also good.

The desktop version of X86 CPU has also produced several so-called "God U" models that are popular and even praised by junky people. In the field of assembled computers in China, it still has a very good reputation... with high cost performance.

In general, although the global market share of Zhiyun Group's consumer-grade X86 CPU is a bit low, it sells quite well in China and is barely acceptable.

However, in the field of server X86 CPU... after several years, apart from our own and our sister companies and institutional clients, we have very few external commercial customers, and our overall market share worldwide is extremely poor.

How should I put it? Ordinary consumers are easy to fool, but enterprises are not... No matter how awesome Zhiyun brags, the ecosystem of your server CPU is not good, and it is troublesome to use.

In fact, the overall cost of using Zhiyun's server CPU is even higher than that of using Intel's server CPU!
In this case, marketing becomes very difficult.

So Zhiyun started to engage in cloud computing and sell computing power, and the roundabout way was CPU.

Now, it is taking advantage of the wave of artificial intelligence, engaging in bundled sales and directly selling finished servers... which are equipped with Zhiyun's own WZ260 server CPU.

Of course, having said that, only a company like Zhiyun Group can engage in such bundling sales and forcibly boost the server CPU business.

If it were Intel or AMD, that would be absolutely impossible.

----

If customers do not want to purchase finished servers, they can wait for a few months and queue up to get the goods, and they can also buy the APO4600 graphics card.

Zhiyun Group will not fill the entire server GPU business just to boost the server CPU business. This is impossible.

At most, we will allocate some APO4600 graphics card production capacity to Zhiyun Computing, so that it has some cards to play, and we will adopt a model of willing parties taking the bait!

Those big customers are naturally unwilling to use Zhiyun Group's own finished servers... The big manufacturers have powerful technology and often directly purchase Zhiyun graphics cards, and then match them with CPUs and other various hardware to build their own GPU data centers, or they have fixed cooperation with some server manufacturers, purchase graphics cards and integrate them into servers to build data centers.

They will not directly use Zhiyun Group's servers plus supporting software and other services... It's not their fault, but Zhiyun's servers are indeed difficult to use for them, and the overall cost-effectiveness is too low.

However, some small and medium-sized customers, especially those small and medium-sized technology companies that lack the ability and experience to build, maintain and operate large data centers, are very interested in the full range of services provided by Zhiyun Computing Technology.

Because Zhiyun Computing Technology not only directly provides servers, but also provides various supporting professional software, and also helps customers deploy GTAI2... For customers, they only need to provide a venue and a check.

Simple and hassle-free, no need to wait.

Therefore, since November, many small and medium-sized enterprises have frequently consulted Zhiyun Computing Technology, hoping that Zhiyun Computing Technology Co., Ltd. can help them build local servers for the deployment of GTAI2.

This has also brought a new wave of business growth to Zhiyun Computing Technology Co., Ltd.

At the same time, some other large manufacturers have also begun to deploy GTAI2 based on their existing GPU data centers!

This leads to an interesting situation.

It seemed as if the whole world was working on artificial intelligence overnight!

Not to mention the many high-tech companies in China and the United States, even the desert areas of the Internet, artificial intelligence, and semiconductor fields, such as Europe and India, have actually started the so-called artificial intelligence plans one by one. Many local companies have also openly deployed GTAI2, and then launched so-called self-developed artificial intelligence services or used them themselves.

It feels like anyone can engage in artificial intelligence!

This scene was even more lively than when Zhiyun Group launched the open source GTAI1 in April.

Because in April, although Zhiyun Group also launched the open source GTAI1, the technology of GTAI1 was relatively poor at that time, and no personal trial version was provided, so the response was still a bit poor.

There is a lot of discussion about it, but few people actually follow suit and deploy GTAI1.

But now, it’s different!

This time, Zhiyun Group was very conscientious and came up with GTAI2, which is quite good, at least much better than the so-called large-scale generative AI they developed themselves.

This means that for most companies, if they have this need, they can deploy GTAI2 on their own...

The only problem is whether we can purchase Zhiyun Group’s APO series graphics cards in large quantities and in a timely manner, especially the latest APO4600 graphics cards!

Zhiyun Group released GTAI2 with great conscientiousness, but they were very unscrupulous in selling graphics cards!
(End of this chapter)