AMD’s Adaptive and Embedded Computing Group (formerly Xilinx) has upgraded its Versal Premium series platform with AI Engines to bring more digital signal processing (DSP) power and serial bandwidth in service of applications in the aerospace and defense and wireless test and measurement markets.
Use cases in these segments, including radar applications, signals intelligence, wireless system testing and wireless device testing are “super signal processing-intensive applications,” said Mike Thompson, Senior Product Line Manager, High-End ACAP Devices, Adaptive and Embedded Computing Group at AMD.
AMD has addressed the growing demands of these applications with the new Versal Premium with AI Engines 7 nm platform that combines advanced signal processing from the company’s already available Versal AI Core series with the massive DSP compute capacity and serial bandwidth capacities of the existing Virtual Premium series, a platform which currently is shipping to early access customers. This results in major performance gains over earlier 16nm Xilinx devices and competing products, Thompson said.
The combination of capabilities provides a 4x increase in signal processing capacity compared to the last-generation Xilinx Virtex UltraScale+ VU13P FPGA. It also eliminates I/O bottlenecks with up to 9Tbps serial bandwidth, and offers significantly reduced size, weight and power through heterogeneous, power-optimized integration of hardened, ASIC-like cores such as 100G/600G Ethernet cores, 400G High-Speed Crypto Engines, DDR memory controller, and integrated PCIe Gen5 blocks.
“We've seen historically intensive demand for signal processing from these markets, and we see that trend accelerating with a massive increase in these requirements for signal processing compute,” Thompson said, “But there are some significant bottlenecks. For example, for any processors that are doing the signal processing for these types of radar and wireless test systems, one of the key bottlenecks has been the ability to get massive amounts of data in and out of these devices to correlate with the amount of signal processing that's really necessary. And so we're seeing serial bandwidth limitations preventing higher channel densities that are really needed by these systems channel densities in terms of the number of radar antennas, or for wireless test, the number of devices under test. Particularly on the aerospace and defense side, but also on the wireless test side we're seeing really highly constrained swap size, weight and power. And it's those challenges in the market that led us to architect Versal Premium with AI Engines the way that we did.”
The size constraints have led AMD to achieve a smaller footprint and lower power for the device, but without sacrificing performance. In radar beamforming applications, for example, the heterogenous compute engines of the platform enable 67% smaller footprint and up to 43% lower power, but with twice the beamforming performance of the company’s previous platform.
The Versal Premium series with AI Engines is expected to begin shipping in the first half of next year, but customers already can leverage prototype designs by using the existing Versal Premium and Versal AI Core Evaluation Kits and devices that already are available, Thompson said