FZ5 Card – AI Accelerator Card
FZ5 Card – AI Accelerator Card
The FZ5 Card is an excellent Artificial Intelligence (AI) accelerator card based on Xilinx Zynq UltraScale+ ZU5EV MPSoC which features a 1.5 GHz quad-core ARM Cortex-A53 64-bit application processor, a 600MHz dual-core real-time ARM Cortex-R5 processor, a Mali400 embedded GPU, a H.264/H.265 Video Codec Unit (VCU) and rich FPGA fabric. It has computing power up to 2.4TOPS and can be seamlessly used with 55 FPS ResNet-50 backbone networks.
Besides, the FZ5 Card has integrated 8GB DDR4, 32GB eMMC, 64MB QSPI Flash and 32KB EEPROM as well as many peripheral interfaces including RS232, RS485, 4 x USB 3.0, Gigabit Ethernet, CAN, TF, DisplayPort (DP), HDMI-IN, USB-UART, JTAG, IO expansion interfaces, etc . It is easy for your secondary development or used for your AI box or many other embedded designs.
FZ5 Card Top View (delivered with active heat sink by default)
FZ5 Card Bottom View
- - Xilinx Zynq UltraScale+ ZU5EV MPSoC based on 1.5 GHz Quad Arm Cortex-A53 and 600MHz Dual Cortex-R5 Cores
- - 8GB DDR4 SDRAM (64-bit, 2400MHz)
- - 32GB eMMC Flash, 64MB QSPI Flash, 32KB EEPROM
- - RS232, RS485, 4 x USB 3.0, Gigabit Ethernet, CAN, TF, DP, HDMI-IN, JTAG…
- - Computing Power up to 2.4TOPS, Runs at 55 FPS for ResNet-50
- - Supports 8- to 16-channel Video Decoding and 4- to 8-channel Intelligent Analysis
- - Supports Running PetaLinux
- - Supports Baidu’s PaddlePaddle AI Framework
The FZ5 Card is able to run PetaLinux 2019.1 and provided with complete Linux BSP. It can also supports PaddlePaddle AI framework which is fully compatible to use Baidu Brain’s AI development tools like EasyDL, AI Studio and EasyEdge to enable developers and engineers to quickly leverage Baidu-proven technology or deploy self-defined models, enabling faster deployment.
MYIR also offers an FZ5 EdgeBoard AI Box for the FZ5 Card which has a rugged and fanless enclosure to enable efficient development for many embedded vision applications such as multimedia, automotive ADAS, surveillance, industrial quality inspection, medical diagnosis, etc. The device can support -40 to 70 Celsius degree extended working temperature with small size and good stability. It has powerful AI capabilities to provide massive and iterative models to realize the image recognition of face, human body, animal, object, text, logo and various customized scenes.
| Features | Description |
|---|---|
| Dimensions | 107mm x 96mm (14-layer PCB design) |
| Power Supply | DC12V/3A |
| Working Temp. | -40°C~85°C |
| SoC |
Xilinx Zynq UltraScale+ XCZU5EV-2SFVC784I (ZU5EV, 784 Pin Package) MPSoC
|
| Memory | 8GB DDR4 (64bit, 2400MHz) |
| Storage |
|
| Communications |
|
| Display |
|
| RTC |
|
| User I/O |
|
| Others |
|
| Software |
|
| Target Applications |
|
Packing List
| Item | Part No. | Included |
|---|---|---|
| FZ5C Card + FZ5C EdgeBoard AI Box |
MYS-ZU5EV-32E8D-EDGE-K1 |
|
Hardware Features
Zynq® UltraScale+™ MPSoC devices provide 64-bit processor scalability while combining real-time control with soft and hard engines for graphics, video, waveform, and packet processing. Built on a common real-time processor and programmable logic equipped platform, three distinct variants include dual application processor (CG) devices, quad application processor and GPU (EG) devices, and video codec (EV) devices.
The Zynq UltraScale+ family provides footprint compatibility to enable users to migrate designs from one device to another. Any two packages with the same footprint identifier code (last letter and number sequence) are footprint compatible. MYIR is using the XCZU3EG-1SFVC784E MPSoC for MYD-CZU3EG Development Board by default, the C784 package covers the widest footprint compatibilities that enable users to select devices among CG, EG and EV.
MYIR may also supply the MYC-CZU3EG CPU Modules with XCZU2CG, XCZU3CG, XCZU4EV or XCZU5EV MPSoC as options. The main features for the MPSoC devices are summarized as below.
| Device | XCZU2CG | XCZU3CG | XCZU3EG | XCZU4EV | XCZU5EV |
|---|---|---|---|---|---|
| Logic cells (k) | 103 | 154 | 154 | 192 | 256 |
| CLB Flip-Flops (K) | 94 | 141 | 141 | 176 | 234 |
| CLB LUTs (K) | 47 | 71 | 71 | 88 | 117 |
| Block RAM (Mb) | 5.3 | 7.6 | 7.6 | 4.5 | 5.1 |
| UltraRAM (Mb) | - | - | - | 13.5 | 18.0 |
| DSP Slices | 240 | 360 | 360 | 728 | 1,248 |
| GTX transceivers | PS-GTR4x (6Gb/s) |
PS-GTR4x (6Gb/s) |
PS-GTR4x (6Gb/s) |
PS-GTR4x(6Gb/s), GTH4x (16.3Gb/s) |
PS-GTR4x(6Gb/s), GTH4x (16.3Gb/s) |
| Processor Units | |||||
| Application Processor Unit | Dual-core ARM® Cortex™-A53 MPCore™ up to 1.3GHz | Quad-core ARM® Cortex™-A53 MPCore™ up to 1.5GHz | |||
| Memory w/ECC | L1 Cache 32KB I / D per core, L2 Cache 1MB, on-chip Memory 256KB | ||||
| Real-Time Processor Unit | Dual-core ARM Cortex-R5 MPCore™ up to 600MHz | ||||
| Memory w/ECC | L1 Cache 32KB I / D per core, Tightly Coupled Memory 128KB per core | ||||
| Graphics Processing Unit | - | - | Mali™-400 MP2 up to 667MHz | ||
| Video Codec | - | - | - | H.264 / H.265 | |
| Memory L2 Cache | H.264 / H.265 | ||||
| External Memory, Connectivity, Integrated Block Functionality | |||||
| Dynamic Memory Interface | x32/x64: DDR4, LPDDR4, DDR3, DDR3L, LPDDR3 with ECC | ||||
| Static Memory Interfaces | NAND, 2x Quad-SPI | ||||
| High-Speed Connectivity | PCIe® Gen2 x4, 2x USB3.0, SATA 3.1, DisplayPort, 4x Tri-mode Gigabit Ethernet | ||||
| Dynamic Memory Interface | x32/x64: DDR4, LPDDR4, DDR3, DDR3L, LPDDR3 with ECC | ||||
| General Connectivity | 2 x USB 2.0, 2 x SD/SDIO, 2 x UART, 2 x CAN 2.0B, 2 x I2C, 2 x SPI, 4 x 32b GPIO | ||||
| Power Management | Full / Low / PL / Battery Power Domains | ||||
| AMS - System Monitor | 10-bit, 1MSPS – Temperature and Voltage Monitor | ||||
Dimensions of FZ3 Card
Software Features
The FZ3 Card is able to run PetaLinux 2020.1 and supports PaddlePaddle deep learning AI framework which is fully compatible to use Baidu Brain’s AI development tools like EasyDL, AI Studio and EasyEdge to enable developers and engineers to quickly leverage Baidu-proven technology or deploy self-defined models, enabling faster deployment.
| Item | Features | Description | Source code provided |
|---|---|---|---|
| Tool chains | gcc8.2.0 | gcc 5.2.1 | |
| gcc version 8.2.0 | gcc version 5.2.1 (Linaro GCC 5.2) | ||
| Bootloader | boot.bin | First boot program including FSBL and u-boot2019.01 | Yes |
| Linux Kernel | Linux 4.19.0 | Customized kernel for FZ5 Card | Yes |
| Drivers | USB2.0/3.0 Host | USB2.0/3.0 Host driver | Yes |
| Ethernet | Gigabit Ethernet driver | Yes | |
| MMC/SD/TF | MMC/SD/TF card driver | Yes | |
| Qspi flash | Qspi flash driver | Yes | |
| CAN | CAN driver | Yes | |
| DP | DP driver | Yes | |
| I2C | I2C driver | Yes | |
| UART | UART driver | Yes | |
| Watchdog | Watchdog driver | Yes | |
| GPIO | GPIO driver | Yes | |
| LED | LED driver | Yes | |
| Button | Button driver | Yes | |
| RTC | RTC driver | Yes | |
| HDMI | HDMI IN driver | Yes | |
| Application | HDMI | HDMI IN example | Yes |
| CAN | CAN example | Yes | |
| Net | Socket example | Yes | |
| File System | Ramdisk | Ramdisk system image | |
| Rootfs | Buildroot making including Qt | Yes | |
| Petalinux | Petalinux2019.1 | Supports Xilinx Petalinux2019.1 development tools. MYIR provides complete BSP for the FZ5 card. |
Baidu Brain’s AI development tools
Software Architecture of FZ5 Card
Downloads
| FZ5 Card Overview | Download |
|---|---|
| Zynq UltraScale+ MPSoC Product Selection Guide | Download |
| FZ5 Card Schematic | Download |
| FZ5 EdgeBoard AI Box Overview | Download |
| EdgeBoard AI Box & Accelerator Card (FZ5) Hardware Manual V1.0.0 | Download |
| Baidu Brain EdgeBoard AI Box &Accelerator Card (FZ5) User Manual V1.0 | Download |
FZ3 Kartı, AMD Zynq UltraScale+ ZU3EG MPSoC tabanlı güçlü bir derin öğrenme hızlandırıcı kartıdır. PetaLinux'i çalıştırabilir ve Baidu Brain AI araçlarını destekler. Kart, 4 GB DDR4, 8 GB eMMC, 32 MB QSPI Flash ve 32 KB EEPROM ile entegre edilmiştir. Ayrıca USB 2.0, USB 3.0, Gigabit Ethernet, TF, DisplayPort (DP), PCIe, MIPI-CSI, BT1120 kamera, USB-UART, JTAG ve IO genişletme arabirimleri gibi çeşitli çevre birimlerini içerir. Hesaplama gücü 1.2 TOPS'a kadar ulaşır ve Baidu'nun PaddlePaddle Derin Öğrenme AI Çerçevesini destekler.