The price of game graphics cards has plummeted, can Moore threads carry the banner of domestic production?
On October 23, the Moore thread opened a round of big price reductions, and the MTT S70 domestic graphics card listed in June this year dropped from the original 2499 yuan to 899 yuan, and the MTT S80 domestic graphics card listed in November last year dropped from 2999 yuan to 1199 yuan.
The price reduction is simply brutal, and there are not many routines, the deposit is 150 yuan, and the balance can be paid on October 31.
Photo: JD
Moore thread did not directly disclose the reason for the price cut, but on October 17, the US Department of Commerce just included 13 Chinese technology companies such as Moore thread on the entity list, which makes this big price reduction seem significant. It can also be seen in many communities that many people are rushing to book due to the price and recent events.
Of course, price is definitely the most important factor.
To be fair, there is still a big gap between the actual performance of Moore's domestic graphics card and the graphics cards of first-line manufacturers such as NVIDIA, whether in terms of gaming or AI acceleration.
But in the end, the two core elements of consumer electronic products are "price" and "product power", consumers' pursuit of "cost performance" has always been, product power can not catch up at present, then can only rely on lower prices to improve cost performance, so as to attract consumers to buy.
However, what the Moore thread now has to face is obviously not only the sales of graphics cards, but also more complex problems, the most direct first question is: Will the Moore thread have a new graphics card?
Three years old, Moore Thread is still a newborn calf
Until the release of its first gaming graphics card, the MTT S80, last year, Moore Thread was a little-known startup.
The most immediate reason is still too young. Moore Thread was founded in 2020 by Zhang Jianzhong, former global vice president and general manager of NVIDIA China, just three years ago, which is still young for an Internet company, not to mention the chip industry with high technical threshold and high input cost.
Zhang Jianzhong, who was still in NVIDIA at the time, pictured / NIO
However, in the early spring of two years later, Moore Thread released the first generation of multi-function GPU chips based on the MUSA architecture "Sudi" in Beijing, and based on this, it launched the desktop graphics card MTT S60 for the consumer market, and the graphics rendering and computing card MTT S2000 for data centers.
MTT S60 and MTT 2000 are more like the appearance and display of Moore threads, and the product itself has not many highlights, mainly for the To B market, resulting in many graphics card users may not know that there is also this Chinese GPU company.
Last fall, Moore Thread released the second-generation multi-function GPU chip "Chunxiao" created by the MUSA architecture, and based on this, the first game graphics card MTT S80 and the multi-function server GPU product MTT S3000 were launched. Among them, the MTT S80 gaming graphics card has 4096 built-in MUSA stream processing cores, which can provide single-precision floating-point computing power of 14.4TFLOPS at a main frequency of 1.8GHz.
Figure/Moore thread
This is also the starting point for many people to get exposed to the Moore thread.
Without him, the discrete gaming graphics card market is too in need of new challengers, whether it is Intel or Moore thread's entry into the market has set off countless expectations. On the other hand, we do need a Chinese GPU company as well.
According to a report by Verified Market Research, the global independent GPU market size in 2022 is about $44.83 billion, eighty percent belongs to NVIDIA, two percent belongs to AMD, and the market share of Chinese GPU companies is infinitely close to 0.
There is also a very real problem here, the absolute backwardness of technology has led to the main direction of domestic GPUs is the industrial application market, not the game market, and the limited production capacity is naturally more inclined to supply enterprise-level customers.
From this point of view, Moore thread is indeed different from other domestic GPU companies, showing rare patience on gaming graphics cards.
Since the release of MTT S80, after 9 important driver updates, it has supported mainstream graphics interfaces such as Vulkan, OpenGL, OpenGL ES, including DirectX 11, and has been officially launched for driver optimization many times, and the number of supported games has exceeded 80.
Moore thread also revealed that MTT S80 has been able to initially run through the UE5 project in the DirectX 11 environment, and the support for tools such as Nanite and VSM based on DirectX 12 will continue to improve.
The continuous update of the official, the support for Chinese GPU companies, and the price of the thousand-yuan machine are all factors that have triggered consumer snapping up again. But for a GPU company, the demand on the consumer side is of course a crucial link, and the other link is the production of the supply side.
The reality is skinny, but there is always hope
On October 17, there was actually an incident, some netizens found that Moore's next-generation gaming graphics card MTT S90 and data center graphics card MTT S4000 appeared at the same time.
Figure: PCIe Device Certification Database
The names of both new cards can now be seen in the PCIe Device Certification Database, with device IDs changed from the previous 02xx to 03xx, model numbers S90 0301 (engineering sample 0300) and S400 0323. There is no information about the specific specifications, but there is speculation that IP from the PowerVR C series will be used.
According to the normal rhythm, Moore Thread should release these two upgrades soon and launch the next-generation gaming graphics card MTT S90 for the consumer market during Double 11, which is probably one of the reasons for the price reduction of Moore Thread.
In theory, at this point in time, the Moore thread must also have prepared the corresponding inventory of new products, which may be able to ensure sales for a period of time. But in the end, it is necessary to go back to square one, no matter which wafer foundry it goes through, the Moore thread needs to continue to manufacture it.
The MTT S80 and MTT 3000 released by Moore Thread at the end of last year are based on the "Chunxiao" multi-function GPU chip, which uses TSMC's 7nm process to integrate 22 billion transistors and built-in 4096 MUSA cores and 128 tensor computing cores.
Figure/Moore thread
However, after October 17, Moore Thread needs to obtain a license from the US Department of Commerce in advance if it wants to pass TSMC, Samsung and other overseas wafer foundries tape-out and mass production, and the severity of the problem can be imagined. Regardless of the subsequent actual progress, Moore thread must rely more on the domestic chip industry chain.
This will not be an easy choice, we all know that there are still many problems in the distance of domestic chip manufacturing, especially in advanced processes need to face a series of challenges such as processes, equipment, and materials. However, for the consumer graphics card market, although it is not necessary to use the most advanced process like mobile phones, 7nm is also a relatively basic requirement.
There is no lack of demand for chips in China, nor is there a shortage of chip design companies, the key is who can provide tape-out and mass production of advanced process chips. If yield and cost are not considered, the use of DUV lithography machine multiple exposures can indeed be a short-term compromise, but the long-term will definitely depend on the technological breakthroughs of fabs and more upstream equipment and material manufacturers.
But in the end, our chip manufacturing can only rely on ourselves, and as the old saying goes, "as long as the mind does not slip, there are always more ways than difficulties," and every inch of breakthrough in the chip production link has the meaning of advancing an inch.
On the other hand, the wave of generative AI and large models is also re-changing the global GPU market, and a large number of large domestic models require massive computing power to provide training and inference, which is also where the opportunity for Chinese GPU companies lies. Liu Qingfeng, chairman of iFLYTEK, pointed out that China's big model can only have an endless future if it is built on the foundation of China's computing power.
This is not necessarily a turning point for the Moore thread and a group of start-up Chinese GPU companies.
This summer, the Moore thread held another press conference, the release of major software and ecology, Zhang Jianzhong said in the media interview after the conference, "At least survive for ten years, which is the first short-term goal of the Moore thread."
Bless him and the Moore thread for being able to make it happen.
The title image is from the Moore thread