NVIDIA Executive Director Huang Renxun
Do not consider Uber's self-driving accident as an influence
Uber, who had been killed in a car accident in the near future, announced his cooperation with NVIDIA during the CES 2018 this year. He had begun to cooperate with NVIDIA. Therefore, many people began to question whether or not NVIDIA had provided an abnormality for the supercar computer before the police had announced the complete investigation. .
However, Huang Renxun said in a follow-up response that the specific cause of Uber's accident was still unknown, but the vehicle used in the Uber incident was Volvo XC90, and responded to Volvo’s response to Uber’s use of its own technology in the accident. Without looking at the standard security technology features of the original vehicle, Uber did not directly use the technology provided by NVIDIA.
Regarding the current suspension of external car-related technical tests, Huang Renxun expressed the hope that he could learn relevant experience from the accident and avoid similar situations in the future. However, it does not mean that NVIDIA will abandon the development of self-driving technology, even if it fails to continue in the real world. Self-driving cars are tested, but they can continue to be trained in simulation mode through Autosim, Drive Constellation computing platform, etc. At the same time, it is emphasized that all car maker related companies still pay considerable attention to the development of autonomous driving technology, and maintain the full-speed development of such technologies as NVIDIA. Think that such technology should not be completely stopped in a single accident.
GPU acceleration has never stopped
In response to the current market demand for GPU acceleration, Huang Renxun said that due to the increasing demand for high-end games, e-sports and content creation, the demand for GPU acceleration has barely stopped, and NVIDIA's NVIDIA RTX technology is expected to be available. Bringing new image vision applications in the past, the images must be rendered for a long time and be completed in an imaginary second under the Volta display architecture. Even real-time rendering of movie-level light and shadow interactions will enable content creators to More efficient, intuitive way to complete more works.
At the same time, the application of artificial intelligence technology continues to grow, and at the same time it will become the mainstream of software applications in the future. By accelerating the computational response speed of GPUs and allowing terminal-to-cloud-equipment operations to be accelerated, artificial intelligence training time will continue to shrink. At the same time, artificial intelligence capabilities can also continue to increase. In addition, in the recent trend of blockchain technology continues to become apparent, but also to promote GPU acceleration demand also has an explosive growth, so we can expect the market demand for GPU has almost never stopped.
Huang Renxun said that due to the continuous expansion of market demand for GPUs, almost every region in the world has NVIDIA GPU production applications, so NVIDIA can actually be regarded as using the world's largest database, but also the supplier of the entire blockchain technology.
Not due to process speed limitations
Some people questioned whether NVIDIA's current adoption of process technology has not yet gone down to 10nm or less, even though other competitors in the market are already preparing to enter the 7nm process development. However, Huang Renxun emphasizes that process technology is indeed very important. It can put more transistors in the same area, and at the same time, it can also make the same power produce higher computing performance. However, there are still more development models in chip design. For example, from the Maxwell architecture to the Pascal architecture, it is not just a process improvement. Higher computing power, which also includes the impact of changes in the architecture itself.
The NVSwitch design announced this time allows applications that have been cascaded to two GPUs via NVLink technology to scale to more GPUs simultaneously. This has led to the total use of 16 new Tesla V100 and 12 NVSwitch sets. The GPU "DGX-2" was born to take advantage of the greater demand for GPU-accelerated computing in the market, and also to solve the problem that GPU development in the past was limited due to process technology.
In addition, working with ARM on Project Trillium platform design will enable NVIDIA's original open architecture training framework for Issac training platform NVDLA to be further added to ARM chip design, when Qualcomm, Samsung, Huawei, Marvell and other vendors use ARM. When designing new processors, you can use NVDLA to learn applications that cascade with NVIDIA terminals or cloud-based collaboration-based GPU acceleration effects. This will allow more IoT devices to accelerate learning through NVIDIA technology and allow more artificial intelligence technologies. Can accelerate the application of IoT devices.
Looking for opportunities from the risks
Whether or not he is worried about the current investment in development because of greater risks, such as the market's influence on the insecurity of artificial intelligence technology, Huang Renxun thinks that any business in the market will be accompanied by risks. For example, NVIDIA makes quite a few display adapters, but it may Faced with users not playing games at all, or assisting in driving more autonomous vehicles, users do not want to get out of the car. However, because the market has these demands, they urge manufacturers to propose more new technologies and continue to learn from experience and feedback. Let technology grow, and NVIDIA continues to invest in such challenges.