Author | Song Jia Ting Editor | Luo Lijuan
Until today, the industry still has different views on whether NVIDIA is a chip company or an AI company.
But Jen-Hsun Huang, NVIDIA's founder and CEO, has said on multiple occasions that "NVIDIA is an AI company."
The change started a few years ago with the global rise of AI, and this company seized the opportunity. With the accumulation of technology in graphics processors (GPUs), NVIDIA quickly transformed from a graphics chip company to an AI platform builder, and it was a huge success.
In a few years, NVIDIA's stock price has more than doubled 10 times, and its market capitalization once exceeded $100 billion, making it the world's hottest AI company.
NVIDIA's presence is ubiquitous in popular AI fields from gaming and autonomous driving to robotics.
In the healthcare industry, some organizations predict that by 2021 AI healthcare will be valued at $6.6 billion. Although the technical threshold of the field is very high and difficult to land, but any player who calls himself an AI company is not willing to miss this big cake.
Attacking the hard bones of technology, NVIDIA also finally ushered in the product landing period. At the 2019 EmTech China "Global Emerging Technology Summit", Kimberly Powell, vice president of NVIDIA in charge of healthcare, shared NVIDIA's development path in artificial intelligence.
"[NVIDIA] uses drones to hone AI technologies, and then expands those technologies into other industries, including healthcare." Kimberly Powell said Clara, the AI-powered medical imaging supercomputing platform developed by NVIDIA to speed up the processing of applications by traditional, older devices.
The core of this platform is described as Clara AGX, based on NVIDIA's Xavier AI computing module, the Turing GPU computing architecture, and the ability to scale from entry-level devices to the most demanding 3D instruments. In Kimberly Powell's opinion, the Clara platform is able to address the huge amount of data that medical devices process at several gigabytes per second. She revealed that Clara has been made available to early partners for free use, with a beta launch to select targets planned for Q2 2019.
This is just one of NVIDIA's forays into AI healthcare.
It is understood that as of November 2018, more than 50 medical institutions have invested in the NVIDIA DGX series of deep learning optimized servers and workstations, and more than 75 organizations have cooperated with them to use AI technology in the medical field, including medical centers, medical imaging companies, research institutions, and start-ups are its partners.
The following is a transcript of an interview with NVIDIA's Vice President Kimberly Powell, who was interviewed by WeatherTech and other media outlets, and has been collated:
Media: The Clara platform was launched last year, and since then, what has been the situation in terms of landings and acceptance? Kimberly Powell : Clara was launched in November 2018. We're also in the exploratory phase, not fully opening it up all at once, but opening up online registration to interested partners first. From the end of November last year to now, between 350 and 400 companies have registered, almost all of the bigger and more famous corporate hospitals and startups in the world have registered. But it's still very new, and there's no question of popularity or acceptance yet, and the current version of Clara is the first one that we've just released. Media: How is the Clara platform being used differently in China and other markets? Kimberly Powell : Customers in the US have a slightly higher level of IT sophistication, so they can implement Clara in the cloud because the US has the data anonymization technology in place to implement Clara in the cloud. The same suite of software can run locally in the hospital as well as in the cloud. For the Chinese market, the support of a hybrid operating environment is very advantageous, because maybe in remote provinces or rural areas of China, where the network conditions are not good enough to get such a cloud service, then they can choose to execute locally; but for those big cities, they can choose to run it in the cloud. Media: Who is Clara's target user base? Kimberly Powell : Clara is targeting three main types of enterprise customers, firstly medical device companies, secondly AI software development companies, and thirdly probably hospitals that have hundreds of apps. NVIDIA provides at least several hundred different SDKs (software development kits) for developers in various fields, and Clara is just one of those hundreds, a toolkit for developers. Media: How does Clara operate? Kimberly Powell : Clara's developer community is more technical and less commercial. For example, Conjecture Technologies uses the inference engine in Clara to execute multiple AI algorithms in parallel. Without this inference engine, an AI model would have to have a specialized GPU to execute. So for companies, Clara enables faster and more efficient execution of their apps in hospitals, running their AI apps with minimal hardware resources. Media: How much does it cost, roughly, to build such a platform in a hospital? Kimberly Powell : Clara is not sold to hospitals as a standalone software suite, but through NVIDIA's enterprise partners. Since it's used as an application installed in a hardware system, it's hard to answer how much Clara would cost on its own. NVIDIA's hardware exists as a basic device in almost everything that computes, so the scope of Clara's use is broad, and even the gaming graphics card you buy can support Clara's operation. Clara is not just for certain types of hospitals, and there may be hospitals that don't yet realize the advantages of Clara. They will come to realize that there are three different types of computing that they can do with the Clara platform, regardless of the hardware they purchase for the computer, and that is a huge benefit to them. Media: What improvements are planned for the Clara platform in the future? Kimberly Powell : Clara itself is a suite of software, and the current release is still relatively early. We've got some key improvements to make, such as interoperability with external hardware systems, such as support for communication protocols, and adding more acceleration engines to Clara to help startups accelerate the deployment of their programs. At the same time, what we are doing is learning about knowledge transfer and assistive features. Hospitals in different regions have different conditions themselves and use different equipment. We hope that the knowledge or conclusions analyzed on a particular hospital's equipment can be popularized and promoted locally, rather than just exporting the results. We should be releasing the first such version at the end of January. Media:What does NVIDIA want to gain with the Clara platform? Kimberly Powell : The Clara platform uses three important technologies from NVIDIA, accelerated compute, artificial intelligence, and visualization. In medical imaging, we don't want the three different workloads of compute, visualization, and AI to be on separate hardware; we want one computer to be able to do all three with Clara. For Clara, NVIDIA's idea is software + hardware, in fact, Clara is also NVIDIA's paving the way for future smart devices. We believe that the collection of data to do post-doctoral analysis depends largely on what device you are on and when the data is collected. In fact, we hope to empower medical devices through software innovation, there will be smart devices on the hardware side, while we configure the SDK for software development, which means realizing anytime, anywhere computing in the medical industry. Media:What are NVIDIA's competitive advantages in AI healthcare? Kimberly Powell : NVIDIA is more of an enabling company, and now that a lot of the big companies are grabbing the market for healthcare AI, it's really NVIDIA that's helping them to better execute the AI healthcare applications at the infrastructure layer. Help them realize the purpose of such a market. Most computing devices use NVIDIA's GPUs, and that's where we're positioned. In addition, NVIDIA has a very large community of developers, we have a CUDA SDK download suite that gets 500,000 downloads per month, and the people who download CUDA are startups or researchers in academia, and such a large base will also help the industries that are doing healthcare because that's going to be their customers as well. Media: Who are NVIDIA's partners in the healthcare industry? Kimberly Powell : We have four main types of partners, in every region. The first type of partner is academic, because NVIDIA itself is not a doctor, we don't produce doctors, we don't do medical research, so we're looking for partners in that area. We also have a NVIDIA Artificial Intelligence Lab (NVAIL), which is a globalized program formally for this type of partner. The second type of partner is startups, and we have a program, Inception, where we have a local NVIDIA team in charge of the healthcare industry to help startups in China. By opening this program, we can provide technical support to these startups, and the latest technology can be given to these startups to be the first to use. If they have good solutions and products, we also help them with commercialization. The third type is the commercialization partners in the industry, such as UW Genetics and United Intelligence, where we mainly have in-depth cooperation at the code development level, and also have joint commercial sales support. Media: What are the main aspects of NVIDIA's cooperation with Chinese companies? Kimberly Powell : We released a blog during the GTC CHINA conference on Accelerated Data Science, and we named the project RAPIDS, which actually represents the evolution of NVIDIA's platforms: first starting with the accelerated computing, then into deep learning, and now machine learning, which is what our RAPIDS platform represents. RAPIDS is the NVIDIA platform for the healthcare industry more broadly, not just for medical imaging. We have attracted a number of companies since we released RAPIDS, including Ping An Insurance and UW Genetics. Ping An Insurance has a lot of claims data and medical data from its insurance customers, and UW Genetics has a huge amount of genetic data, and even companies that make digital wearable devices, like Tantric Cube, have welcomed RAPIDS as a platform. (For more exciting financial information, download the Wall Street Journal App)