The sheer amount of sensor data consumed by modern embedded computing systems has necessitated a shift towards heterogeneous computing. Rather than do everything with a general purpose processor, FPGAs, GPUs and other special purpose processors can each complete tasks they are best suited for.
Current drawbacks to this approach can include the space consumed by individual boards or chips and the inefficiencies in interconnects between the special purpose chips. This is exactly why the new Xilinx Versal ACAP was created. These all-in-one ACAP chips have taken a typical FPGA fabric and added an array of special purpose processing functions all in one package with a high performance interconnect known as a network on chip (NoC). In fact, all current chips in the 7nm Xilinx Versal family are ACAPs which demonstrates Xilinx’s dedication to this new computing paradigm.
So, what exactly is an Adaptive Compute Acceleration Platform and what benefits does it have over traditional FPGAs?
What Is an Adaptive Compute Acceleration Platform?
An Adaptive Compute Acceleration Platform—or ACAP—is a heterogeneous computing system-on-chip that includes a large FPGA fabric. In addition, there are general purpose and real-time ARM cores, scalar processors suited for artificial intelligence and machine learning math operations, and hard silicon IP blocks dedicated to common I/O functions. The hard silicon blocks include things like Ethernet interfaces, high-performance PCIe DMA, and DDR memory controllers. All of the special purpose processing elements in the ACAP are coupled with a high performing NoC and delivered in a single chip. Essentially, the ACAP provides many useful processing functions in a small area at a reduced power. This makes the ACAP an excellent choice for any embedded high-performance computing application.
Three major advantages of an ACAP:
There are three main reasons to choose an ACAP over a more traditional FPGA:
- Size, Weight, and Power–Traditionally, tasks might require multiple chips to perform the same functions of an ACAP, and these might even be spread across multiple boards. With an ACAP, all the processing takes place within a single chip, with the NoC tightly coupling the processing elements together. In addition, moving common interface IP cores into hard silicon provides a more efficient, lower power design.
- Commonly Used IP Now Offered in Hard Silicon–By providing IP Cores like Ethernet, PCIe DMA, and DDR memory controllers in hard silicon, more programmable logic resources are available for the user and the interface logic is more efficient. This lowers the power utilization.
- Software Programmability–Xilinx ACAP allows users to benefit from the latest software abstraction layers. This means users can deploy their algorithms and applications quickly and efficiently through standard software.
Who Uses Adaptive Compute Acceleration Platforms?
Adaptive Compute Acceleration Platforms are used in any industry that requires high performance heterogeneous computing. They are especially well suited for small form factor applications at the edge with limited SWaP requirements. This can include 5G, AI, ML, and other sensor-processing applications. In today’s Military/Aerospace world, everything needs to be a mobile datacenter with the ability to process vast amounts of sensor data from multiple platforms in real-time. The Xilinx ACAP is an ideal solution for these applications.
Even though the newest applications require this capability, they also need to communicate with older technology. The programmable logic in the ACAP makes it possible to upgrade this older tech with new equipment and capabilities while requiring no changes to the existing equipment on that platform. This allows for transparent high-bandwidth and low-latency functionality with new COTS equipment without needing to start from scratch.
How do ACAPs Affect the Future of FPGAs?
Many industries and applications already utilize FPGAs for their robust reliability, long life-spans, and superior ability to parallel process multiple tasks. FPGAs will remain a competitive processor option for the foreseeable future.
However, many users will appreciate the increased capabilities of the ACAP architecture with multiple processor functions in one package. The SWaP benefits of the ACAP will likely push many users down that path. Xilinx’s latest Versal offering shows their commitment to the heterogeneous computing architecture. Those at the bleeding edge of computing applications will benefit from the ACAPs smaller yet more powerful profile. Along with the advantages listed above, the ACAP lends itself to deeper implementation in compute-heavy applications where more than the standard FPGA is required.
Why Choose New Wave DV for Your Adaptive Compute Acceleration Platforms?
If you think an ACAP device is right for you, New Wave DV offers leading-edge rugged hardware featuring Xilinx Versal ACAPs. New Wave DV also provides interface IP that allows you to expedite development as you focus on your high-level application and algorithms. Load your custom IP onto our hardware or license IP for your existing hardware or new design.
New Wave offers two Adaptive Compute Acceleration Platform options
The main difference between these two options is that the V1161 has a Nvidia Mellanox ConnectX-5 ethernet chip—the industry leader in bandwidth—while the V1163 does not.
Learn More About ACAPs & FPGAs With New Wave DV
Still have questions about ACAP vs FPGA board solutions, and how they can fit into your applications? Get in touch today to start a discussion with an expert.