Over the past ten years, we have been noticing the trend of application performance demands starting to outpace Moore’s law in a variety of fields. The solution has been to rely on specialized processors like GPUs, FPGAs and other specialized ASICs. With these alternative compute architectures, hardware was becoming more complex and software was becoming more abstract.
Over the past two decades, first we saw x86 getting into the datacenter and after it became the dominant compute platform for applications, virtualization of x86 happened to make it more efficient. Then, there were different levels of storage systems in a datacenter which created the need for storage to be virtualized. After that, it was networking which had to be virtualized. Two years ago, my co-founders and I made a bet that the future datacenter is going to be a heterogeneous in-terms of compute with different types of co-processors like GPUs, FPGAs, and other ASICs connected by a fast backbone. We wanted to build the virtualization and the abstraction layer for the futuristic co-processor based datacenter, and so Bitfusion was born.
Just like how the virtualization advancements around x86, storage and networking made those technologies accessible to the masses, we wanted to make at-scale heterogeneous compute available to any organization with Bitfusion. We believed that this would not only help companies to do the same things faster and better, but will also fuel innovation by enabling new capabilities—pursuing new markets with new products and services.
Two years ago, heterogeneous hardware like GPUs, FPGAs and other ASICs were just on the horizon, but now this is real and enterprise applications are taking advantage of these hardware. Even public cloud now offer GPUs, FPGAs and specialized ASICs for on demand consumption. We built the Bitfusion core virtualization technology to help increase the efficiency of managing and utilizing co-processors like GPUs and FPGAs, and applied it to a variety of industries, from pharmaceuticals to artificial intelligence.
Now, fast forward two years later. With deep learning and AI as a major focus area for our company, along with the announcement of our successful Series A financing, we are happy to announce the beta launch of Bitfusion Flex, the industry’s first software for managing elastic AI infrastructure to make development and deployment of deep learning applications faster and more economical. Bitfusion Flex is built on top of our core virtualization technology for GPU management, and will help increase efficiency of end-to-end application lifecycle for deep learning and AI. The goal of Bitfusion Flex is to enable data science groups to do deep learning without any devops and access the full potential of co-processors like GPUs and FPGAs for the same. Our vision is to be the operating system for the next generation heterogeneous computing, and today’s launch of Bitfusion Flex is yet another step in that direction with AI and deep learning as a major usecase.