← All issues
Edge AI Performance Is Moving to the Module Level

Edge AI Performance Is Moving to the Module Level

· By Mansa Muhammad

The density of compute at the edge is shifting from general-purpose processors to specialized, high-TOPS silicon integrated into small form factors. ADLINK’s new COM-HPC-mPTL module demonstrates this shift by bringing Intel Core Ultra Series 3 "Panther Lake" architecture to a COM-HPC R1.3 Mini computer-on-module.

The hardware targets high-performance workloads through significant increases in neural processing unit (NPU) and graphics capabilities. The top-tier configuration utilizes the Intel Core Ultra X7 358H processor, which features 16 cores and 12-core Xe3 graphics capable of 122 TOPS. When paired with a 50 TOPS NPU, this module provides the raw throughput necessary for complex edge AI tasks.

The architecture is not limited to a single performance tier. The module lineup includes:

  • Intel Core Ultra 5 325: An 8-core processor providing 40 TOPS from Xe3 graphics and 47 TOPS from the NPU.
  • Intel Core Ultra 7 356H: A 16-core processor delivering 40 TOPS from Xe3 graphics and 50 TOPS from the NPU.
  • Intel Core Ultra X7 358H: The high-end 16-core option with 122 TOPS from Xe3 graphics and 50 TOPS from the NPU.

This compute density is supported by a high-bandwidth memory subsystem, offering up to 64GB of LPDDR5x at speeds reaching 7467 or 8533 MT/s. Connectivity remains a priority for industrial and edge deployments, with the module exposing I/Os through a standard 400-pin high-density connector. This includes support for up to 16x PCIe lanes across Gen4 and

Subscribe to The Mansa Report

Strategic intelligence on AI, business building, and the future of technology. Delivered Monday through Friday.