Advantech has expanded its collaboration with Qualcomm Technologies to speed up deployment of generative artificial intelligence at the edge, focusing on enterprise and industrial use cases. The partnership centers on integrating the Qualcomm Dragonwing artificial intelligence on-prem appliance solution, powered by the Qualcomm Cloud AI 100 Ultra accelerator, into Advantech’s SKY-641E3 4U high performance edge server. A specialized PCIe switch backplane is used in the design to provide high bandwidth connectivity, which is described as necessary to unlock the full potential of high density artificial intelligence inference for edge workloads.
The SKY-641E3 is positioned as an alternative to standard dual slot GPU servers, with an architecture specifically tuned for maximum inference throughput rather than training. The system supports up to 12 single-slot FH/FL accelerators within a single 4U chassis, and this configuration is presented as delivering a superior density to footprint ratio for inference heavy tasks. By prioritizing inference accelerators, the design aims to avoid the thermal overhead typically associated with more power hungry training GPUs.
This high density layout is optimized for high concurrency edge scenarios such as city wide intelligent video analytics and large scale retail artificial intelligence agents that require many simultaneous inferences. The server is described as capable of handling the simultaneous processing of hundreds of real time data streams, while maintaining efficiency within the physical and thermal constraints of a 4U chassis. Together, Advantech and Qualcomm Technologies are pitching the SKY-641E3 with Qualcomm Dragonwing as a way for enterprises and industrial operators to deploy scalable generative artificial intelligence inference closer to where data is generated.
