Zynq UltraScale+ SBC and AI Box support PaddlePaddle AI tools
Sep 17, 2020 — by Eric Brown 1,395 viewsMYIR’s $699 and up “FZ5 Card” SBC and $729 and up “FZ5 EdgeBoard AI Box” run Linux on a Zynq UltraScale+ MPSoC ZU5EV with 2.4-TOPS AI, 4-8GB DDR4, 32GB eMMC, GbE, HDMI-in, DP, 4x USB 3.0, and more.
In July, MYIR launched a $259, Raspberry Pi like FZ3 Card SBC built around its MYC-CZU3EG CPU Module with optional MYD-CZU3EG Development Board based on Xilinx’s FPGA-enabled Xilinx Zynq UltraScale+ ZU3EG. Now, the company has returned with a larger FZ5 Card — AI Accelerator Card SBC and FZ5 EdgeBoard AI Box embedded system based on it that uses the higher end Zynq UltraScale+ ZU5EV.


FZ5 Card — AI Accelerator Card (left) and rear view of FZ5 EdgeBoard AI Box
(click images to enlarge)
Like the ZU3EG, the ZU5EV SoC offers 4x Cortex-A53 cores, 2x 600MHz Cortex-R5 cores, a 667MHz Mali-400 MP2 GPU, and FPGA fabric. However, it clocks the -A53 cores to 1.5GHz instead of 1.2GHz and adds a video control unit (VCU) that supports H.264/H.265 video support. The ZU5EV supplies a higher-end FPGA with 256K logic cells.
Other SBCs that support the ZU5EV include iWave’s Pico-ITX iW-RainboW-G36S Corazon AI and Avnet’s Ultra96. MYIR’s earlier MYC-CZU3EG module also powers MYIR’s GigE Vision and camera equipped Vision Edge Computing Platform.
— ADVERTISEMENT —
The 107 x 96mm’s FZ5 Card’s FPGA enables up to 2.4-TOPS neural processing performance (vs. 1.2-TOPS with the FZ3 Card’s ZU3EG) and can run ResNet-50 at 55fps for in “actual measurement” at 10W to 20W consumption, says MYIR. It also supports 8- to 16-channel video decoding and 4- to 8-channel intelligent analysis.

FZ5 Card software architecture
(click image to enlarge)
Like the FZ3 Card, the FZ5 Card runs Xilinx’s PetaLinux 2019.1 and is compatible with Baidu’s PaddlePaddle deep learning AI framework. PaddlePaddle supports Baidu Brain AI development tools like EasyDL, AI Studio and EasyEdge. (For more on PaddlePaddle, see our FZ3 Card report.)
The FZ5 Card ships with 4GB or 8GB DDR4, 64MB QSPI, 32GB eMMC, and a microSD slot with 32GB card. Major ports include GbE, mini-DP, HDMI-in, 4x USB 3.0, and a micro-USB debug port.


FZ5 Card front and back detail views
(click images to enlarge)
Other features include MIPI-CSI, CAN, 2x serial, and a 100-pin GPIO header. The 12V board supports -40 to 85°C temperatures and offers an RTC and JTAG. Schematics are included.
The FZ5 EdgeBoard AI Box extends the FZ5 Card with all major ports exposed. The 120 x 100 x 50mm embedded system has a slightly lower -40 to 70°C operating range.


FZ5 Card top view with fan (left) and multiple views of the FZ5 EdgeBoard AI Box
(click images to enlarge)
Specifications listed for the FZ5 Card SBC include:
- Processor — Xilinx Zynq UltraScale+ MPSoC ZU5EV(4x Cortex-A53 @ 1.5GHz); FPGA with up to 256K logic cells; up to 2.4-TOPS AI; 2x Cortex-R5 MCUs @ 600MHz; Mali-400 MP2 GPU @ 677MHz; VCU with H.264/H.265 support
- Memory/storage:
- 4GB or 8GB DDR4-2400 RAM
- 64MB QSPI flash (PL)
- 32GB eMMC
- MicroSD slot with 32GB card
- Networking – Gigabit Ethernet port
- Media I/O:
- Mini-DisplayPort at up to 4K/30fps
- HDMI input port
- MIPI-CSI 40-pin FPC interface
- Other I/O:
- 4x USB 3.0 host ports
- Micro-USB debug port and cable
- RS-232, RS-485, CAN via terminal blocks
- JTAG header (PL)
- Expansion – 100-pin (2x 50) FPFA I/O interface (5 x PS_MIO, 69 x PL_IO)
- Other features — RTC with battery holder; system and FPGA reset buttons; 5x LEDs; optional fan; schematics included
- Power — 12V DC jack via terminal block; 12V/3A adapter and cable
- Operating temperature — -40 to 85°C
- Dimensions — 107 x 96mm
- Operating system — Xilinx PetaLinux 2019.1 with Baidu PaddlePaddle AI Framework support
Further information
The FZ5 Card — AI Accelerator Card sells for $699 (4GB RAM) or $759 (8GB) and the FZ5 EdgeBoard AI Box sells for $729 (4GB) or $789 (8GB). More information may be found in MYIR’s announcement and the FZ5 Card and FZ5 EdgeBoard AI Box product and shopping pages.
Please comment here...