On March 5, a reporter’s question caused Huang Renxun to pick up the microphone and knock it over his head at a media communication meeting in December 2019. By the way, the huge echo can prove that he smashed hard.
This slightly cute move and his next words made all the reporters including me laugh out loud:
“It’s over, when I heard this question, I was in vain for two hours without dripping water on the stage, and my throat was smoking.”
The reporter’s question was a bit vague, but it also asked many people’s doubts——
Why didn’t Nvidia release chips this time?
This understanding is actually correct. At the end of 2019, when “Intels” used acquisitions and releases of new products every few days to “blow” their own AI chip strategy to the outside world without any dead ends. At the technical conference, “response” to various people who challenged themselves.
However, at this meeting, except for a dedicated chip for autonomous driving called Orin, all products released by the world’s most valuable artificial intelligence chip giant were “software.” And Orin will not start production until 2022.
This meeting perfectly fits his important point of view on the 2019 Q3 quarter earnings conference call——
“Nvidia has become a software company.”
After this “software company” announced its fourth-quarter financial report for the 2020 fiscal year in mid-February 2020, its market value once reached nearly 200 billion U.S. dollars, a record high. Although it has now fallen back to more than $160 billion, Nvidia has undoubtedly been the legendary “golden stock” on Wall Street in the past two years. With the market’s increasingly strong demand for data centers, the war between Nvidia and old and new chip players is destined to become more bloody.
Unparalleled software advantages
As Nvidia is increasingly using its software capabilities to sell products that look like hardware, we have reason to believe that this chip company has begun to pursue a “software first” strategy.
Nvidia’s “overwhelming disadvantage” in dozens of AI chip companies’ PPTs every year has been “slapped”, and its stock price can continue to rise steadily.
According to Zhong Hua, technical director of the autopilot company Wenyuanzhixing, when he was studying at Carnegie Mellon more than a decade ago, it was very difficult to program the GPU at that time with machine code deep into the core of the graphics card.
“We use assembly language to write code, which is really brain-burning. After NVIDIA launched Cuda, it is equivalent to packaging complex graphics card programming into a simple interface, which benefits the majority of programmers. Now the mainstream deep learning framework is basically They are all based on Cuda for GPU parallel acceleration.”
In 2007, NVIDIA officially launched the GPU unified computing architecture platform Cuda. The milestone of this architecture is that the GPU no longer exists alone in the personal user’s graphics card, but only works on one’s own “one-acre three-dimensions”; instead, the GPU is generalized and the “personal computer” can be parallelized. The “supercomputer” of computing.
An unnamed self-driving technician made the above statement. He said that many domestic artificial intelligence companies have developed on the free ride of Nvidia: “So you see, Nvidia has occupied two parts at once, one is the training end, and the other is the application end. In the short term, there can be no one. Shake their position.”
Support point for the next decade
In February 2020, the U.S. financial website used “Wall Street was shocked” to describe the strong performance of Nvidia’s data center sector in its fiscal year 2020 Q4 earnings report——
Nvidia’s chip sales reached a record high of US$968 million.
Although “gaming” is the source of application and innovation for image processing and neural network computing, it is also the old line that Nvidia must keep.
However, the PC game market, which is becoming saturated and constantly being squeezed by mobile terminals, is far less attractive to Nvidia than emerging markets.
According to the latest data from Peddie Research, a game market research organization, Nvidia has been occupying more than 70% of the market share in the PC discrete graphics market in the past five years. In particular, its Geforce series graphics cards are extremely popular in the gaming and mining industries. .
Cloud is the focus of artificial intelligence development, and it is also the next ambition of Nvidia.
At the GTC at the end of 2019, Nvidia’s high-profile cooperation with Chinese companies such as Alibaba and Baidu further highlighted the important role GPU plays in the training of cloud artificial intelligence algorithms-“The era of search is over, and the era of intelligent recommendation has arrived.
Nvidia’s Tesla T4, V100 and other processors and supporting acceleration software play a major role in resource allocation, quantification and acceleration of recommended algorithm model training based on billions of data.
Although the actions of competitors such as Intel and AMD in this regard cannot be underestimated, in the next 3 to 5 years, their GPU solutions are unlikely to significantly reduce Nvidia’s market in the field of data center AI accelerators.
History has proven that chip companies can never sleep peacefully.