Dev Kit Weekly: Xilinx Kria KV260 Vision AI Starter Kit
July 30, 2021
You heard that right. Xilinx is no longer just a provider of programmable logic devices. They’ve officially entered the embedded module business, and not just with low-cost evaluation-only hardware.
The company’s Kria line of high-performance yet cost-optimized SoMs, including the Kria K26 platform at the heart of the KV260 kit, were built with volume-based production in mind and are available in versions rated for commercial or industrial use.
I guess this shouldn’t come as too much of a surprise given that Xilinx started shipping data center acceleration cards a couple years ago, but it’s certainly an indicator that the electronics landscape is changing. If you want to find out more about these trends, check out he links to our recent Embedded Insiders podcast that features excerpts from an interview with Xilinx CEO Victor Peng.
But enough of all that. On to the kit!
While the 119 mm x 140 mm x 36 mm KV260 itself isn’t rated for production deployment, it can get you well on your way thanks to a package that includes a carrier card and this module outfitted with a fan and heat sink. The 7.5 to 15 W system needs that thermal management solution in certain applications because the SoM is designed around a high-performance Zynq UltraScale+ MPSoC device that integrates:
- A 1.5 GHz quad-core Arm Cortex-A53 MPCore
- Dual-core Arm Cortex-R5F MPCore that clocks at 600 MHz
- 667 MHz Mali-400 MP2 GPU
- 1-32 stream 4Kp60 video codec unit
- 256K system logic cells
- And 1,248 DSP slices
That represents 1.36 TOPS of machine learning throughput. And aside from that, there’s even a 13 Megapixel ON Semi AP1302 image signal processor onboard.
All of this is accompanied by 26.6 Mb of on-chip SRAM, 4 GB of 64-bit DDR4 on the module, and another 16 GB of eMMC flash. It supports multiple cameras with up to eight interfaces that can be dedicated to that purpose, including 2 MIPI interfaces, a Raspberry Pi camera interface, one DisplayPort 1.2a, one HDMI 1.4, and four USB 3.0 interfaces. There are also two IAS module connectors for integrating ON Semi cameras into your design, as well as Gigabit Ethernet, support for multiple PCIe generations, and a Pmod connector for additional expansion.
Why all that compute, memory and interface support on an embedded SoM? Well, for the same reason Xilinx decided to make an AI vision module in the first place. Because computer vision is demanding. And it’s hard to do. Really hard.
To simplify it, Xilinx has developed a “Getting Started Guide” that its engineers claim will have you up and running with the kit in under an hour – even if you don’t have any prior experience with these Xilinx technologies or vision applications. The five-step process walks you through the board, setting up the SD card image, getting everything connected, and booting the kit – which can be done from the SD card or onboard 512 Mb QSPI and is supported by the MPSoC’s hardware root of trust or the on-module Infineon TPM 2.0.
The tutorial proceeds to the final step that introduces accelerated applications, or pre-built programs available from Xilinx's online App Store that configure the kit hardware for end use cases but still allow customization of things like AI models, sensors, and so on. The application outlined in the tutorial is for a smart camera, but there are also defect detection, object tracking and re-identification, license plate recognition, visual keyword spotting, and more. From there, users can advance into a variety of free or paid training courses from Xilinx that can earn you a certificate of completion.But
If you are more advanced than all of this, don’t worry. The Kria Starter Kit works with the Vivado hardware design studio, PetaLinux SDK, or Vitis software development environment that already includes base implementations for video capture and decode on the KV260. And if you’re looking for a place to start once you’ve fired up Vitis, you can run benchmarks through Xilinx’s xmulti platform management utility by simply executing the “platformstats” command to measure performance characteristics like power consumption, throughput, and latency.
Obviously there’s tons of great stuff here in Xilinx’s first foray into embedded SoMs. To learn more about the Xilinx Kria family of SoMs, checkout the new system-on-module section of the Xilinx website at www.xilinx.com/products/som/kria.html.
As mentioned, the Kria KV260 Vision AI Starter Kit comes with the module and carrier, and thermal management solution for $199. The power supply is sold separately, and available for $25 on its own or as part of a $59 accessory pack that includes an AR1335 13 MP camera module, microSD card, and various cables.
But one of you will be able to waive at least part of that cost by winning this week’s raffle. All you have to do to enter is fill out the form that’s linked to on the screen, and you could take home this very KV260 free of charge. Well, actually we’ll mail it to you but it’s still free.
That’s it for this week. Good luck in the raffle and we’ll see you next week on Dev Kit Weekly.