Eurotech, a long-time leading provider of embedded computing systems and leader in Internet of Things (IoT) enablement, has launched a high performance edge computing system (HPEC) for the transport industry to enable the deployment of advanced AI-based analytics capabilities at the edge. The DynaCOR 40-36 is fanless and uses liquid cooling capabilities, making it suitable for the harshest of transportation environments.
With up to 10240 CUDA cores and 1280 Tensor cores, the DynaCOR 40-36 delivers up to 237 TFLOPS for deep learning precision (up to 29.6 TFLOPS for single precision and up to 14.8 TFLOPS for double precision), enabling the training of AI models to happen at the edge rather than in a data centre or cloud. This reduces latency and enables real-time responses, making it ideal for monitoring, inference, training, and general purpose GPU-accelerated workloads on vehicles and at the edge.
Moreover, the combination of high-performance computing, data logging and networking performance enables the DynaCOR 40-36 to support low-latency 5G network applications.
With an automotive grade power supply (24VDC; 12 and 48VDC factory option), and certifications for automotive and heavy-duty use cases (ISO 16750-2, EN60068-2-27) the DynaCOR 40-36 is suitable for operation under the most demanding conditions, including extreme operating temperature. Thanks to liquid cooling, the system is largely decoupled from environmental temperature and can be installed in recesses with little or no ventilation.
Pierfrancesco Zuccato, senior product manager at Eurotech, commented, “The DynaCOR 40-36 sets a new record for AI performance in truly rugged applications. It is a great platform for the most computation intense inference models and supports double precision for accurate calculations and approximations, for advanced requirements and training on the field.”
The DynaCOR 40-36’s features include:
- 16 core Interl Xeon D-2183IT – powerful CPUs that allow server-class processing while maintaining high energy efficiency and good thermal management – ideal for embedded applications.
- Up to two NVIDIA GV100 GPUs – enabling the processing of multiple tasks in parallel. The integration of GPUs allows faster processing of image data and the application of AI computational models, allowing machine and deep learning at the edge.
- Four high speed NVME units with one additional 512GB NVME to host the OS – advanced storage drives to enable ultra-fast data logging, making it immediately available for applications that require real-time responses. The dedicated unit for the OS allows system files to be written on a separate unit, allowing more efficient management of the data among the core units.
- Dual 100GbE, quad 10GbE and two GbE ports – these provide a high bandwidth, low latency feed for advanced sensors and allow the implementation of clustered architectures (built on two or more centrally managed systems used to perform the same task to double the computing power and reduce the time it takes to complete the task).
- High performance frame grabber, CAN, USB3 and serial interfaces – making it simple to connect the DynaCOR 40-36 to cameras, the vehicle, other peripherals and edge devices.
The DynaCOR 40-36 is compatible with the other Eurotech HPEC systems, including the DynaCOR 40-35 (high performance data logger), DynaNET 100G-01 (100GbE spine switch) and the DynaNET 10G-01 (10GbE leaf switch), and will be available in Q4 of 2021.