The CPU is 'stuck', and Chinese companies are changing tracks one after another: Can domestic GPUs be successful?

China has an unusual number of GPU startups as the country seeks to gain dominance in artificial intelligence as well as semiconductor sovereignty, according to a new report from technology market researcher Jon Peddie Research.

As the demand for artificial intelligence (AI), high-performance computing (HPC) and graphics processing increases at an unprecedented rate, the number of global GPU manufacturers has also grown in recent years. When it comes to discrete graphics cards for PCs, AMD and Nvidia maintain the lead, while Intel is trying to catch up.

In the 1980s and 1990s, dozens of companies around the world were developing graphics cards and independent graphics processors, but in order to obtain the highest performance in 3D games, cutthroat competition, in which the vast majority of companies were eliminated.

By about 2010, only AMD and Nvidia were able to offer competitive standalone GPUs for gaming and computing, while the other companies focused on integrated GPUs or GPU IP.

Starting around 2015, the number of PC GPU developers in China began to increase rapidly, thanks to China's push for technological self-sufficiency and the emergence of AI and high-performance computing as high-tech The emergence of major trends.

According to data from Jon Peddie Research, a total of 18 companies are currently developing and producing GPUs. Two companies mainly develop SoC-bound GPUs for smartphones and laptops, six develop GPU IP, and 11 GPU developers focus on GPUs for PCs and data centers, including AMD, Intel and Nvidia.

In fact, if other Chinese companies, such as Biren Technology and Tianshu Zhixin, are added to the list, the number of GPU companies will be even greater. However, Biren Technology and Tianshu Zhixin currently only focus on AI and high-performance computing, so JPR does not consider them to be GPU developers in the traditional sense.

China’s road to technological salvation: GPU development

As the world’s second largest economy, China will inevitably compete with the United States and other developed countries Countries compete on almost every front. China is doing everything it can to attract engineers from around the world.

In fact, in China, hundreds of new IC design companies are established every year. They develop products ranging from tiny sensors to complex communications chips in an effort to become self-sufficient with Western suppliers.

But to truly jump on the wave of artificial intelligence and high-performance computing, you need CPUs, GPUs, and special-purpose computing acceleration technologies.

When it comes to CPUs, China is facing an increasingly tight technological blockade from the United States in terms of manufacturing equipment and technology, and it is impossible to catch up with the global advanced level soon. However, on another track, it can be said that developing and producing a decent GPU is more fruitful than trying to build a competitive CPU.

“For Chinese companies, artificial intelligence training is the main driving force for independent research and development of GPUs. On the one hand, it is because Nvidia’s GPUs are too expensive, and on the other hand, it is also out of self-sufficiency. Desire," said Jon Peddie, director of JPR.

GPU is essentially a parallel device. There are a large number of computing units inside it for redundancy. This makes the GPU easier to start and run. The cost per transistor is relatively low. Overall Yield is also good. Additionally, the parallel nature of GPUs makes them easier to deploy in a scalable manner.

Compared with CPUs, GPUs have less stringent requirements for process technology in design and manufacturing, even if China's most advanced chip manufacturer SMIC is not as advanced as TSMC With the production technology, it is still possible to achieve sufficiently impressive performance by utilizing GPU performance expansion.

In fact, even if China's GPU developers lose the opportunity to use TSMC's advanced nodes (N7 and below), at least some of them can still make it simpler at SMIC GPU design and meet the needs of AI, HPC and some gaming/entertainment markets.

Moreover, from a national perspective, GPUs with AI and HPC capabilities may also be said to be more important than CPUs, because AI and HPC can enable completely new applications, such as autonomous driving Applications such as automotive and smart cities.

Although the U.S. government has vigorously restricted the export of supercomputer-based CPUs and GPUs to China, the relatively lower threshold for design and manufacturing of GPUs compared to CPUs makes such restrictions far less effective. The CPU comes obviously.

GPU micro-architecture is relatively easy, but hardware design is expensive

However, it should be noted that although there are currently many GPU developers, But only two can truly create competitive independent GPUs for PCs. This may be because it is relatively easy to develop a GPU architecture, but it is really difficult to implement it correctly and design appropriate drivers.

CPU and GPU microarchitecture are basically "the intersection of science and art." These architectures are complex sets of algorithms, and the teams developing them can be quite small, but can take up to several years.

It can be understood that microarchitecture is done on napkins and whiteboards. As for the cost, if it is just the architect himself, the team size can be only one person, maybe three or four people. But any kind of architecture, building, rocket ship, network or processor is a complex chess game.

For example, trying to predict what manufacturing processes and standards will be five years from now, cost-effectiveness tradeoffs, what features to add, and what features to abandon or ignore are all very tricky and time-consuming tasks.

Architects spend a lot of time making assumptions in their heads, such as if the cache is made 25% larger, if there are 6000 FPUs, should a PCIe 5.0 I/O bus be made? ? Can this be completed on time? And so on.

Since microarchitecture development can take years and requires talented designers, in a world where time to market is of the essence , many companies simply license off-the-shelf microarchitectures or proven GPU IP from companies such as Arm or Imagination Technologies.

For example, China's Innosilicon licenses GPU microarchitecture IP from the British company Imagination for use in its Fantasy GPU.

There is also a Chinese GPU developer that uses Imagination’s PowerVR architecture. Meanwhile, another GPU manufacturer, Zhaoxin, uses a GPU microarchitecture obtained by Via Technologies.

The cost of developing a microarchitecture can vary, but is relatively low compared to the cost of physical implementation on modern high-end GPUs.

For years, Apple and Intel, two companies with significant engineering talent, have relied on Img for GPU design. MediaTek and other small SoC vendors rely on Arm. Qualcomm used ATI/AMD for a long time, and Samsung switched to AMD after trying to design its own graphics engine for several years.

Recently, two new Chinese GPU companies hired former AMD and NVIDIA architects, and two others use IMG. Time to enter the market and learn the skills as an architect, what to worry about, and how to find solutions, is a very time-consuming process.

"If you can find a company that already has a design plan and has been designing it for a long time, you can save a lot of time and money, and in the market, time is Everything."

"There will be too many problems in this process. Not every GPU designed by AMD or Nvidia can be a winner. However, a good architectural design can After several generations of adjustments, it has gradually improved." said the person in charge of Jon Peddie Research, a research organization.

For new production processes, the cost of hardware implementation and software development is too high. According to estimates by the International Business Times, the design cost of equipment manufactured using 5nm-level technology exceeds US$540 million. If the process is 3nm, the design cost will increase three times.

“If you take layout and floor plan, simulation, verification and drivers all into account, the cost and time of GPU development explodes,” Peddie explained.

"Hardware design and layout is very straightforward: if you get a line wrong, it can take months of troubleshooting."

Currently, the world There are only a few companies in the world that can develop modern gaming or computing GPU (46 billion-80 billion transistor scale) chips at the level of AMD and Nvidia.

However, the BR104 and BR100 released by China's Biren Technology not long ago have reached almost similar levels. (It is speculated that the BR104 contains approximately 38.5 billion transistors).

Currently, 8 of the world’s 11 PC/data center GPU design suppliers are from China, which speaks for itself.

Maybe in the near future, we won’t see competitive indie gaming GPUs except from American companies. Whether China can launch a competitive competitor remains to be seen.

The above is the detailed content of The CPU is 'stuck', and Chinese companies are changing tracks one after another: Can domestic GPUs be successful?. For more information, please follow other related articles on the PHP Chinese website!