Description
Mustang-V100-MX8
Accelerate To The Future
Intel Vision Accelerator Design with Intel Movidius VPU.
A Perfect Choice for AI Deep Learning Inference Workloads
Powered by Open Visual Inference & Neural Network Optimization (OpenVINO) toolkit
- Half-Height, Half-Length, Single-Slot compact size
- Low power consumption ,approximate 2.5W for each Intel Movidius Myriad X VPU.
- Supported OpenVINO toolkit, AI edge computing ready device
- Eight Intel Movidius Myriad X VPU can execute eight topologies simultaneously.
OpenVINO toolkit
OpenVINO toolkit is based on convolutional neural networks (CNN), the toolkit extends workloads across Intel hardware and maximizes performance.
It can optimize pre-trained deep learning model such as Caffe, MXNET, Tensorflow into IR binary file then execute the inference engine across Intel-hardware heterogeneously such as CPU, GPU, Intel Movidius Neural Compute Stick, and FPGA.
Applications
 |  |
| Machine Vision | Smart Retail |
 |  |
| Surveillance | Medical Diagnostics |
Overview
Powered by Open Visual Inference & Neural Network Optimization (OpenVINO) toolkit. Half-Height, Half-Length, Single-Slot compact size. Low power consumption ,approximate 2.5W for each Intel Movidius Myriad X VPU. Supported OpenVINO toolkit, AI edge computing ready device. Eight Intel Movidius Myriad X VPU can execute eight topologies simultaneously.OpenVINO toolkit is based on convolutional neural networks (CNN), the toolkit extends workloads across Intel hardware and maximizes performance. It can optimize pre-trained deep learning model such as Caffe, MXNET, Tensorflow into IR binary file then execute the inference engine across Intel-hardware heterogeneously such as CPU, GPU, Intel Movidius Neural Compute Stick, and FPGA.
Features
- Operating Systems
- Ubuntu 16.04.3 LTS 64bit, CentOS 7.4 64bit (Support Windows 10 in the end of 2018 & more OS are coming soon)
- OpenVINO Toolkit
- Intel Deep Learning Deployment Toolkit
- Model Optimizer
- Inference Engine
- Optimized computer vision libraries
- Intel Media SDK
- *OpenCL graphics drivers and runtimes.
- Current Supported Topologies: AlexNet, GoogleNet V1, Yolo Tiny V1 & V2, Yolo V2, SSD300, ResNet-18, Faster-RCNN. (more variants are coming soon)
- High flexibility, Mustang-V100-MX8 develop on OpenVINO toolkit structure which allows trained data such as Caffe, TensorFlow, and MXNet to execute on it after convert to optimized IR.
*OpenCL is the trademark of Apple Inc. used by permission by Khronos
Specification
| Model Name | Mustang-V100-MX8 | Mustang-V100-MX4 | Mustang-MPCIE-MX2 | Mustang-M2AE-MX1 |
| Main Chip | Eight Intel Movidius Myriad X MA2485 VPU | 4 x Intel Movidius Myriad X MA2485 VPU | 2 x Intel Movidius Myriad X MA2485 VPU | 1 x Intel Movidius Myriad X MA2485 VPU |
| Operating Systems | Ubuntu 16.04.3 LTS 64bit, CentOS 7.4 64bit (Support Windows 10 in the end of 2018 & more OS are coming soon) | Ubuntu 16.04.3 LTS 64bit, CentOS 7.4 64bit, Windows 10 64bit | Ubuntu 16.04.3 LTS 64bit, CentOS 7.4 64bit, Windows 10 64bit | Ubuntu 16.04.3 LTS 64bit, CentOS 7.4 64bit, Windows 10 64bit |
| Dataplane Interface | PCI Express x4 | PCIe Gen 2 x 2 | miniPCIe | M.2 AE Key |
| Power Consumption | <30W | 15W | Approximate 7.5W | Approxinate 5W |
| Operating Temperature | 5°C~55°C(ambient temperature) | 0°C~55°C (In TANK AIoT Dev. Kit) | 0°C~55°C (In TANK AIoT Dev. Kit) | 0°C~55°C (In TANK AIoT Dev. Kit) |
| Cooling | Active fan | Active fan | Passive/Active Heatsink | Passive Heatsink |
| Dimensions | Half-Height, Half-Length, Single-width PCIe | 113 x 56 x 23 mm | 30 x 50 mm | 22 x 30 mm |
| Support Topology | AlexNet, GoogleNet V1/V2/V4, Yolo Tiny V1/V2, Yolo V2/V3, SSD300,SSD512, ResNet-18/50/101/152, DenseNet121/161/169/201, SqueezeNet 1.0/1.1, VGG16/19, MobileNet-SSD, Inception-ResNetv2, Inception-V1/V2/V3/V4,SSD-MobileNet-V2-coco, MobileNet-V1-0.25-128, MobileNet-V1-0.50-160, MobileNet-V1-1.0-224, MobileNet-V1/V2, Faster-RCNN | AlexNet, GoogleNetV1/V2, Mobile_SSD, MobileNetV1/V2, MTCNN, Squeezenet1.0/1.1, Tiny Yolo V1 & V2, Yolo V2, ResNet-18/50/101 | AlexNet, GoogleNetV1/V2, Mobile_SSD, MobileNetV1/V2, MTCNN, Squeezenet1.0/1.1, Tiny Yolo V1 & V2, Yolo V2, ResNet-18/50/101 | AlexNet, GoogleNetV1/V2, Mobile_SSD, MobileNetV1/V2, MTCNN, Squeezenet1.0/1.1, Tiny Yolo V1 & V2, Yolo V2, ResNet-18/50/1015% ~ 90% |
| Operating Humidity | 5% ~ 90% | 5% ~ 90% | 5% ~ 90% | 5% ~ 90% |
*Standard PCIe slot provides 75W power, this feature is preserved for user in case of different system configuration
Ordering Information
| Part No. | Description |
| Mustang-V100-MX8-R10 | Computing Accelerator Card with 8 x Movidius Myriad X MA2485 VPU, PCIe Gen2 x4 interface, RoHS |
| Mustang-V100-MX4-R10 | Computing Accelerator Card with 4x Intel Movidius Myriad X MA2485 VPU, PCIe Gen 2 x 2 interface, RoHS |
| Mustang-MPCIE-MX1 | Deep learning inference accelerating miniPCIe card with 2 x Intel Movidius Myriad X MA2485 VPU, miniPCIe interface 30mm x 50mm, RoHS |
| Mustang-M2AE-MX1 | Computing Accelerator Card with 1 x Intel Movidius Myriad X MA2485 VPU,M.2 AE key interface, 2230, RoHS |
| Mustang-M2BM-MX2-R10 | Deep learning inference accelerating M.2 BM key card with 2 x Intel Movidius Myriad X MA2485 VPU, M.2 interface 22mm x 80mm, RoHS |
