Adapting to survive
Steve Rogerson reports from Xilinx’s XDF developers’ forum in The Hague. The intelligent, connected world has become a reality. This was how Victor Peng, President and CEO of Xilinx, opened the company’s XDF developers’ forum in The Hague in November. For this, he said, adaptability was vital.
“Large scale systems must be resilient and adaptability helps with that,” he said. “In the natural world, the species that are more adaptable survive. The same is true for digital systems. It is all about helping hardware and software developers use these powerful platforms.”
He said the goal was to help developers create innovations without deploying new boards and new silicon. “That adaptability is available today,” he said.
Peng was quick to brag about how the firm’s devices are being used in diverse applications from space to transportation, automotive, data centres and medical. One of the big applications today is in the roll-out of 5G mobile networks with around 50 operators around the world already deploying 5G in 27 countries. In Europe, Xilinx is working with the likes of Ericsson and Nokia on deployment.
“We are really excited at being part of the 5G deployment,” said Liam Madden, Xilinx Executive Vice President. “In Japan, this is being deployed in time for the Olympics.”
He said the amount of data going through the internet is doubling every two years. “A lot of this is to do with streaming data,” he said. “About 15% is due to Netflix. If you look at the problem, you have to come up with solutions. Xilinx’s disruptive technology is in 5G.”
However, automotive is one of the firm’s key markets and the firm used the event to launch its XAZU7 and XA ZU11 domain controllers, qualified for automotive safety and reliability.
“These were developed due to customer demand,” said Peng. “We are already in edge devices but now we are also in the central domain controllers, so we have a common architecture throughout the vehicle. If you have a mix of different architectures, that involves more development time.”
Earlier this year, Xilinx announced its Versal adaptive computing platform, which Peng described as, “the most powerful processor on the planet”. This is sampling today and a Versal development board is available.
“But a platform is not just hardware,” said Peng. “You need software to run on the hardware for it to truly be a platform. Vitis is our unified development platform and it is available right now, today.”
Vitis marks a big step for Xilinx in the way it has decided to join the open source world. The software and its open source libraries are available for immediate download, free of charge.
“We are embracing open source,” said Ramine Roane, Vice President for Software. “We have used open source in the past and have contributed to open source since 2007. This year, we decided to embrace open source with our run-time and libraries. This is new for Xilinx.”
Using Vitis, software developers can accelerate their applications with Xilinx adaptive hardware, without the need for hardware expertise. Rather than imposing a proprietary development environment, the Vitis platform plugs into common software developer tools and uses open source libraries optimised for Xilinx hardware.
The associated Xilinx developer site provides access to examples, tutorials and documentation, as well as a space to connect the Vitis developer community. It is managed by Vitis experts and enthusiasts, providing information on the latest updates, tips and tricks.
“We see the future of computing as moving towards adaptive platforms,” said Roane. “With Xilinx hardware, you can keep up with the speed of innovation without updating silicon. This is what Vitis lets software developers do.”
Madden added: “We are moving up the food chain by having platforms. We are now thinking about software first, whereas our history was all about the hardware.”
And Peng said: “It is becoming so challenging. We are going down the platform road because it suits our customers.” Vitis has been in development for five years. “It was not easy to get there,” said Roane.
In his keynote, Xilinx Executive Vice President Salil Raje picked up the Vitis theme. “I am super excited to be presenting Vitis today,” he said. “This is the platform to get more out of your adaptable devices.”
He spoke about three main industry trends – the move towards heterogeneous computing, cloud-to-edge unification, and the proliferation of artificial intelligence (AI). Heterogeneous computing was happening because of the difficulties in providing the necessary scale.
“Adaptable devices are gaining momentum because they can be optimised for a specific domain or application, or be customised on the fly,” said Raje.
On the second, applications are going from the cloud to the edge to the end point. Some are starting at one end and some at the other, and some are hybrids to get the best of both worlds.
“One application could be a camera in a smart city with all the analysis in the cloud,” said Raje. “To reduce latency, you might put the analytics at the edge. You can put object detection in the camera.”
On AI, he said this was coming into various applications and being used everywhere from smart cities to autonomous driving, as well as in retail, security, healthcare, video analytics and finance.
This is why Xilinx also announced Vitis AI at the show. This is a development platform for AI inference on Xilinx hardware platforms, including edge devices and Alveo accelerator cards. It consists of IP, tools, libraries, models and example designs.
“All these three trends are intertwined,” said Raje. “That is why you need a unified software environment that can address all three challenges, and that is the idea behind Vitis. This is a major step in us becoming a platform company. Vitis can be used by embedded developers, application developers, infrastructure developers, and AI and data scientists. You can build, analyse and run your development on a variety of devices.”
He said Xilinx was investing heavily in pre-certified library functions, some of which were domain specific and some that could be used across all domains. “We are open source on all these functions,” he said.
To demonstrate what this meant in practice, Raje invited onto the keynote stage Manoj Nayee, CEO of Boston Limited, to announce Stream+, an optimised way to enable large scale ultra-high-density live video encoding. The appliance is configured using
Supermicro building blocks and powered by Xilinx Alveo accelerator cards and V-Nova’s P+ software.
“It is my great pleasure to attend the Xilinx XDF event in partnership with Supermicro to demonstrate the Boston Stream+,” said Nayee. “Boston has been working alongside Xilinx and V-Nova to develop this highly dense encoding, enabling use cases such as live 4Kp60 encoding.”
Hyperscale video services such as social and e-sports networks need to encode and serve vast numbers of live streams to their users. The server infrastructure required to satisfy this is often the highest cost for network operators. Stream+ provides encoding, and offers quality and cost benefits to video delivery users.
“Xilinx is entering a new era, and we are still committed to our traditional customers,” said Raje. Picking up on the cloud-to-edge unification trend was Ivo Bolsens, CTO at Xilinx.
“Edge computing has to be real time,” he said. “It has to be fast enough but if it is too fast you are wasting resources. Low latency and low power are in many cases essential.”
Computing at the edge, he said, had to handle a lot of data bring produced by different sensors. In automotive, for example, high-resolution cameras were producing 30Gbit/s of data.
“You can’t upload that video for analysis,” he said. “Edge computing will have to handle a massive amount of data, and that will only get worse.”
To illustrate this, Thomas James, Senior Research Fellow at European science research centre Cern, explained how the technology was increasing the laboratory’s physics capabilities.
“Cern is the place where we ask the fundamental questions such as what are the origins of the universe and what is the universe made of,” said James.
The famous large hadron collider sees 2.4bn collisions every second producing about 500Tbit/s of data. “It is impossible to store all that data,” said James. “So we have to select the most interesting collisions for analysis. FPGAs are necessary for this so only a fraction reaches the CPUs. We are taking advantage of machine learning and deep learning. Xilinx FPGAs give us unprecedented physics capabilities.”
Summing up, Peng said: “I really love advanced technology. It is not just because it is intellectually stimulating, which it is, it is because we can make the world better.”