All News | Boards | Chips | Devices | Software | Archive | About | Contact | Subscribe
Follow LinuxGizmos:
Twitter Facebook Pinterest RSS feed
*   get email updates   *

Zynq UltraScale+ SBC sells for $259, offers Baidu AI tools

Jul 1, 2020 — by Eric Brown 3,337 views

MYIR’s $259 “FZ3 Card” SBC runs Linux with Baidu AI tools on a Zynq UltraScale+ via its MYC-CZU3EG module with 4GB DDR4 and 8GB eMMC. Specs include mini-DP, GbE, 2x USB, PCIe, 2x 40-pin GPIO, and CSI and BT1120 cam links.

MYIR has launched a 100 x 70mm, -40 to 85°C tolerant development board built around its MYC-CZU3EG module, which integrates Xilinx’s FPGA-equipped Zynq UltraScale+ MPSoC. The FZ3 Card, which is also referred to as the Deep Learning Accelerator Card, is available in an affordably priced $259 FZ3 Kit. Applications include AI cameras and computing devices, robotics, intelligent car, intelligent electronic scale, patrol UAV, and other intelligent embedded applications.

FZ3 Card with (left) and without optional active heatsink
(click images to enlarge)

The FZ3 Card is the third carrier board option for the MYC-CZU3EG module and by far the cheapest. In addition to the full-featured, $659 MYD-CZU3EG Development Board that was announced with the module, MYIR recently released a $599 Vision Edge Computing Platform (VECP) kit that combines the module with a smaller MYD-CZU3EG-ISP dev board that is just slightly larger than the FZ3 Card, but with a different layout and feature set. The VECP kit is equipped with an 8.42MP Sony IMX334LLR camera plus 4K@30 ISP, GigE Vision, and USB3 Vision IP cores.

Baidu PaddlePaddle and OIN membership


Each of the three MYC-CZU3EG kits has its own Linux stack. This time around, MYIR supplies Xilinx’s PetaLinux 2019.1. The board is also compatible with Baidu’s PaddlePaddle deep learning AI framework, which is part of Baidu’s open source OpenEdge edge computing platform. PaddlePaddle can tap the FPGA’s up to 1.2-TOPS neural processing performance with support for MobileNet at up to 100fps, says MYIR.

FZ3 Kit (left) and software architecture with PaddlePaddle
(click images to enlarge)

PaddlePaddle (PArallel Distributed Deep LEarning) is compatible with Baidu Brain AI development tools like EasyDL, AI Studio, and EasyEdge. Now available in a PaddlePaddle 1.8.2 version, PaddlePaddle received its last major upgrade last November with PaddlePaddle 1.6, which updated its IoT-focused Paddle Lite variant to v2.0. Paddle Lite 2.0 is compatible with PaddlePaddle models and pretrained models from other sources including new support for ResNet-50. It also adds support for edge-based FPGAs such as the Zynq UltraScale+.

This week Baidu, which owns the largest Chinese language search engine and is a leading AI vendor, joined leading Linux patent defense organization Open Invention Network (OIN). The non-aggression network protects companies that use Linux and open source technologies from patent attacks by agreeing to cross-license patents with other members without royalties. A decade ago, OIN battled Microsoft over Linux-related patent issues, but in recent years Microsoft joined the group along with Huawei and other tech giants.

MYC-CZU3EG module

On MYIR’s FZ3 Card, the 60 x 52mm MYC-CZU3EG module defaults to a mid-range Zynq UltraScale+ XCZU3EG variant with 154K logic cells. MYIR “may” provide other options with higher and loewr-end FPGAs, presumably only for volume orders. All the UltraScale+ models feature 4x Cortex-A53 cores, which are clocked to 1.2GHz on the XCZU3EG model and up to 1.5GHz on higher-end models. Other standard features are dual 600MHz Cortex-R5 cores, and on the XCZU3EG or higher models, a 667MHz Mali-400 MP2 GPU.

MYC-CZU3EG detail view (left) and available Zynq UltraScale+ MPSoC models
(click images to enlarge)

The MYC-CZU3EG module is loaded with 4GB DDR4, 32MB QSPI flash, and 4GB or 8GB eMMC, with the latter capacity available on the FZ3 Card. The 3.3V module also has a GbE controller, PMIC, watchdog, clock generator, and 4x LEDs.

FZ3 Card and Kit

The FZ3 Card is equipped with coastline GbE, USB 3.0, USB 2.0, and micro-USB serial debug ports. There is also a mini-DP port for up to 4K@30fps video. For imaging input you get MIPI-CSI and a digital HDTV BT1120 interface, both via FPC connectors.

FZ3 Card detail views
(click images to enlarge)

For expansion, the SBC provides a single-lane PCIe 2.1 interface and dual 40-pin GPIO connectors. You also get a 6-pin JTAG header, FPGA and system reset buttons, a boot switch, and an RTC with battery socket. The board runs on a 12V/2A terminal block input and the system operates at about 5W static power.

The FZ3 Kit adds a 12V/2A adapter, 16GB microSD card, product disc, and micro-USB and mini-DP-to-HDMI cables. For $19 more, you get an active heatsink with fan.

Unlike MYIR’s Zynq-7000-based Z-turn and Z-turn Lite SBCs, this does not appear to be an open source hardware offering. However, there is a casual mention of referring to the schematics, suggesting that perhaps the carrier board is open spec. As usual with MYIR, you get detailed documentation, and further details are said to be available upon request.

Further information

The FZ3 Kit with FZ3 Card, MYC-CZU3EG module, and accessories is available for $259. More information may be found in MYIR’s announcement and product page.

(advertise here)

Print Friendly, PDF & Email

Please comment here...