Advantech, a company in the industrial IoT sector, is collaborating with Qualcomm Technologies to enhance edge AI capabilities through the SKY-641E3 server. At the centre of this collaboration is the Qualcomm Dragonwing AI on-prem appliance solution, which is integrated into Advantech’s edge server via the Qualcomm Cloud AI 100 Ultra accelerator.
This partnership aims to support edge processing efficiency by using a specialised PCIe switch backplane. This provides the high-bandwidth connectivity needed to support high-density AI inference across enterprise and industrial applications.
The SKY-641E3 is designed for high inference throughput and differs from standard GPU servers. It supports up to 12 single-slot FH/FL accelerators within a single 4U chassis, offering a high density-to-footprint ratio. The architecture is intended to support concurrent edge workloads such as city-wide video analytics and large-scale retail AI applications, while reducing reliance on power-intensive training GPUs.
The system includes dual high-performance processors, with a dual-socket configuration supporting CPUs up to 400W TDP. It also features ten PCIe 4.0 x16 slots and two PCIe 4.0 x8 slots, supporting accelerator deployment. Power delivery per slot is rated at 75W, with scalability up to 350W for up to four slots. The system is equipped with a 1+1 redundant 2700W AC PSU to support continuous operation.
With the Qualcomm Cloud AI 100 Ultra integrated into the SKY-641E3, the system supports data centre-class inference performance alongside edge processing. Key hardware specifications include a PCIe FH3/4L form factor, 150W TDP, 870 TOPS (INT8) ML capacity, 128 GB on-card DRAM with 548 GB/s bandwidth, a PCIe Gen 4 host interface with 16 lanes, and 64 dedicated AI cores.
The combination of Advantech’s server and Qualcomm’s technology is intended to support on-premise AI deployments across several sectors, including smart manufacturing, retail and hospitality, healthcare, and smart cities and logistics.
