Dell is more than a hardware company. Although the company is known for their laptops and monitors, the implementation of AI in the infrastructure of the market is necessary. Tom van Peer, presales director at Dell in the Netherlands, has a background in computer engineering and has seen AI develop in many ways, for instance in computer vision.
What is AI, according to you?
“A few years ago, I would’ve said ‘elevated statistics’, but now I know it’s way more than that. I think it isn’t easy to put a final definition on AI. I always draw a little scheme with on top Artificial Intelligence, below that Machine Learning and below that Deep Learning.”
“Engineers tend to put ‘our’ intelligence into AI, but in the end, the result which was needed is reached through better hardware rather than better software for an AI. An example is Deep Blue, the chess computer that defeated former chess champion, Gary Kasparov. This was an approach in which better hardware was used to elevate the AI.”
How have you seen AI develop in recent decades?
“My first job was at Phillips Research in Eindhoven in 1986. I was working with a group that focused on automated industrialization. We were developing a robot arm with a camera and sensors so that it could pick up (moving) eggs and glasses without breaking them. I was busy developing the real-time computer vision for the camera in this robot arm. Unfortunately, these cameras weren’t as good as they are now, and I had a 16 megahertz single-core CPU on my computer, which is not the best hardware you can work with.”
“The algorithms, which we call deep learning and machine learning, were already existing in the 1980s. The problem was that the hardware couldn’t process these algorithms yet. My job was to transfer the algorithms into dedicated hardware to process them faster. Shortly after, the second AI winter came around, after which the fun began with really powerful hardware like GPUs.”
Dell is known as a hardware supplier. What does Dell have to do with AI?
“We provide infrastructure in the way of data centres and storage. We have an unstructured data solution from the media department that was designed to store and stream video content. However, this appeared to be ideal storage under an AI training system as it is extremely good at serving unstructured data in many parallel streams. Another example, we have engineered a server in which you can mix 10 GPUs instead of 8 (fixed type) GPUs which results in a more space-efficient training system.”
“DELL builds hardware: however, everybody could build hardware. But we collaborate a lot with ISVs (Independent Software Vendors) to certify their solutions on our hardware and publish reference architectures for use by customers to use our hardware in the most optimal way. This prevents problems where servers, networking and storage are not well balanced, which could result in your expensive GPUs only running at half their potential power for instance. “
“We also developed our own software. For example, we created a program called ‘Distributed Analytics’ in which data is not being stored or trained at a central system. But instead, the program sends the analytics to the data and just gathers and combines the results centrally. An example of this is analyzing aeroplane maintenance data, which is in the order of terabytes per flight and is distributed globally. It’s not efficient or even feasible to transfer and store all this data on a central system, so we send the analytics to the data and gather and combine results centrally.”
What role does your Computer Engineering background play in your work at Dell?
“I learned how to develop hardware and write a bit of Python, but my biggest asset is understanding the limitations. When people only learn how to develop software, they seldom get taught what the limitations of the hardware are. They build a model that may not run well once deployed to hardware. I love to help these engineers in the way that they know how to handle a challenge like this.”
How can the AI community help you further?
“We’re looking for software partners that are working with hardware. Especially in the manufacturing corner, for example, sensors and computer vision cameras in a factory monitor all the behaviour and activities happening in the factory. We see that there is a vast market developing in this field.”
“This is already happening in the big beer breweries for instance. They want that the whole process of brewing beer is the same all over the world to deliver the perfect quality all the time. With sensors and AI, you can monitor everything and ensure beer quality worldwide.”
