FPGAs provide one of the most versatile and customizable platforms available with respect to high-performance computing (HPC) applications. The same should be true of cloud-based FPGA deployments, but this has not been the case... until now.
The amount of data being created and consumed, and the computational capabilities required to process this data, are currently experiencing exponential growth. According to Statista, the two zettabytes of worldwide data in 2010 increased to over 64 zettabytes in 2020 and is expected to rise to more than 180 zettabytes by 2025. Meanwhile, at Intel Architecture Day 2021, Raja Koduri, who is senior VP and general manager of the Accelerated Computing Systems and Graphics (AXG) Group at Intel, noted that the demand for computational horsepower, which traditionally used to double approximately every two years, is currently doubling every three to four months.
There are three technologies that are primarily used to provide computation: central processing units (CPUs), graphics processing units (GPUs), and field-programmable gate arrays (FPGAs). Today’s CPUs are incredibly sophisticated, and they are logically adept when it comes to implementing “decision-making” tasks, but they are inefficient for most data processing algorithms. As their name suggests, GPUs were originally developed for graphics processing applications, which they perform using an array of small CPUs running in parallel. This parallel architecture also allows them to perform many data processing algorithms significantly faster than their CPU cousins. FPGAs offer the highest levels of data processing performance, which are fully customizable and can be configured to perform data processing applications in a massively parallel fashion. From an environmental perspective, the dynamic nature of FPGAs allows them to be adapted to changing requirements, allowing improvements not only in software, but also in hardware.
Markets and verticals for FPGA-based acceleration include artificial intelligence (AI), machine learning (ML), and deep learning (DL); data analytics and database processing; video analytics and transcoding; biotechnology (e.g., genomics and protein folding); finance and high-frequency trading (HFT); cryptography, research, electronic design automation (EDA), and... the list goes on.
Many companies have enjoyed the benefits of FPGA acceleration when creating custom implementations on their own premises (“on-prem”). However, they have experienced less success when attempting to realize these implementations in the cloud (i.e., in data centers).
There are several reasons for this, including that traditional cloud vendors focus on providing CPUs and GPUs, with FPGAs being something of an afterthought. In some cases, these vendors create their own custom FPGA cards, making it cumbersome for users to deploy their applications and difficult to migrate these applications to other clouds without spending a lot of time, effort, and resources. Another problem is lack of flexibility, because users typically have access to only a single type of FPGA card, which—in addition to the FPGA itself—means fixed resources like onboard memory and interfaces.
This is where VMAccel enters the picture. Our solution was conceived from the ground up with FPGAs in mind. Our FPGA-as-a-service infrastructure enables users from anywhere in the world, with any skill level, to deploy FPGAs into their computational workloads with ease via the cloud. Furthermore, in addition to having our own FPGA-centric cloud, we can also help customers with their on-prem deployments if required.
Furthermore, our FPGA cloud features a wide variety of FPGA cards, including Xilinx Alveo U50, Xilinx Alveo U55C, Xilinx Alveo U55N, Xilinx Alveo U250, Xilinx Alveo U280, and Xilinx VCK5000. Our solutions are fully configurable with respect to the number and types of cards that can be assigned to each application, and infinitely scalable to match our customers’ growing requirements.
Our cloud platform enables users of all skill levels to implement FPGAs into their computational workloads. Using our wide selection of FPGA cards, users can deploy existing FPGA code with a single click, develop new bitstreams in our pre-configured development environments, and much more.
Depending on their requirements, our users can deploy their applications on bare metal, virtual machines (VMs), or Docker/Kubernetes container orchestration frameworks. We offer simple, easy one-click acceleration applications for users who do not wish to develop their own FPGA code. We also offer pre-configured instances with complete development environments, drivers, simulation tools, and everything else necessary for development on FPGAs.
We support a wide range of developer flows for more advanced users, including low-level RTL, OpenCL, C++, Python, and more, along with completely custom environments, images, drivers, etc. In addition to web access, we also support optional JTAG and QSFP capabilities that provide full access to all board features and interfaces, and we even offer client-specific customization down to the hardware level.
The key thing we want you to remember is that, in the same way, that FPGAs are customizable, so should cloud-based FPGA-based acceleration. With VMAccel, it is!
