VCK5000 Versal Development Card for AI Inference
by: Xilinx, Inc
The Xilinx VCK5000 Versal development card is built on the Xilinx 7nm Versal ACAP architecture and is designed for designs requiring high throughput AI inference and signal processing compute performance.
Overview
Product Description
The Xilinx® VCK5000 Versal® development card is built on the Xilinx 7nm Versal ACAP architecture and is designed for designs requiring high throughput AI inference and signal processing compute performance.
Delivering 100X greater compute power than today’s server-class CPUs and better MLPerf inference performance than today’s server-class GPU, with standardized development flows, the VCK5000 is an ideal development platform for cloud acceleration and edge computing applications.
VCK5000 is fully supported by Vitis™ AI development environment which consists of optimized IP, tools, libraries, models, and example designs. It is designed with high efficiency and ease-of-use in mind, unleashing the full potential of AI acceleration on VCK5000. You can run Tensorflow, Pytorch, Caffe models using Python or C++ APIs in minutes without any prior hardware knowledge about FPGA.
Key Features & Benefits
Superior AI Inference Platform
- 145TOPS (Int8) @ $2,495
- AI inference with AI Engine outperforms nVidia T4 on MLPerf 1.0
Solutions Out-of-Box
- Vitis AI 1.3 with 2 DPU variants (384 vs 288 AIE cores)
- 50+ model zoo support
- Natural Language (BERT) demo
- MLPerf submission ResNet-50 demo (11% better than T4)
- Getting started with Vitis AI, Board user guide
Cloud <-> On-Premises
- Deploy solutions on the cloud or on-premises interchangeably
- Scalable to application requirements
Product Information
Board Features
| Card Specifications | VCK5000 | |
|---|---|---|
| Device | VC1902 | |
| Compute | Active | Passive* |
| INT8 TOPs (peak) | 145 | 145 |
| Dimensions | ||
| Height | Full | Full |
| Length | Full | 3/4 |
| Width | Dual Slot | Dual Slot |
| Memory | ||
| Off-chip Memory Capacity | 16 GB | 16 GB |
| Off-chip Total Bandwidth | 102.4 GB/s | 102.4 GB/s |
| Internal SRAM Capacity | 23.9 MB | 23.9 MB |
| Internal SRAM Total Bandwidth | 23.5 TB/s | 23.5 TB/s |
| Interfaces | ||
| PCI Express | Gen3 x 16 / Gen4 x 8 | Gen3 x 16 / Gen4 x 8 |
| Network Interfaces | 2x QSFP28 (100GbE) | 2x QSFP28 (100GbE) |
| Logic Resources | ||
| Look-up Tables (LUTs) | 899,840 | 899,840 |
| Power and Thermal | ||
| Maximum Total Power | 225W | 225W |
| Thermal Cooling | Active | Passive |
* We will ship the Active board only. If you remove the fans from the VCK5000, following the Hardware Installation Guide in the VCK5000 Versal Development Card Secure Site, it becomes Passive.
Resources
Default
Default
Title
Document Type
Date
Follow the steps below to get started with the VCK5000:
Step 1: Install VCK5000 in your X86 host machine following the Hardware Installation Guide in the VCK5000 Versal Development Card Secure Site.
Step 2: Install Vitis AI 1.3.1 docker in your X86 host machine
docker pull xilinx/vitis-ai-cpu:latest
- For more info about Vitis AI, please visit https://github.com/Xilinx/Vitis-AI
Step 3: Run 51 VCK5000 AI models, following the the instructions in the VCK5000 Versal Development Card Secure Site.
For any questions regarding VCK5000 getting started, please contact vck5000-aie_sponsor@xilinx.com.