COM‑HPC: The Next‑Generation Open Standard for Modular Embedded Systems

Embedded computing is poised to adopt COM‑HPC, the next‑generation open standard for modular system designs. Because its architecture is intricate and often misunderstood, clear guidance is essential.

Stakeholders view COM‑HPC through two lenses. The embedded edge‑server community sees it as a platform for high‑performance, rugged workloads, while existing COM Express users focus on client modules and their evolution. They question whether to transition now and what tangible benefits the new standard delivers.

Open standard platform for embedded servers

A short history of high‑performance
Computer‑on‑Module standards

Year Details

2001 Foundation of the ETX‑IG and launch of the first manufacturer‑independent module standard

2005 PICMG publishes COM Express 1.0 specification

2010 COM Express 2.0 specification

2012 Sales of COM Express modules exceed those of ETX

2012 COM Express 2.1 specification

2018 Foundation of the PICMG COM‑HPC Committee

2019 Release of the COM‑HPC pinout

2020 Launch of the COM‑HPC specification

Because Computer‑on‑Module (CoM) standards are open and vendor‑agnostic, products built on them enjoy multi‑decade lifecycles. OEMs can still order legacy ETX modules today, and PCIe‑based standards promise even longer relevance through backward compatibility.

COM‑HPC Server is the first truly open standard for modular embedded rack‑ and box‑server designs in harsh environments. In today’s classic server landscape, application‑ready processor modules are still rarely leveraged, yet they unlock significant advantages.

By delivering processors, memory, and high‑speed interfaces in a single, standardized module, designers can focus on tailoring the carrier board to precise size and I/O requirements. This approach streamlines design time, enables rapid time‑to‑market, and reduces production costs—particularly for small‑batch products where a full custom design would be prohibitive.

Because only the module is swapped during an upgrade, the cost can be cut by roughly 50 % compared with a full replacement of a 1U or 3U rack system. The result is a more sustainable investment and an extended return on investment over the product’s lifecycle.

Beyond the generic benefits, COM‑HPC Server introduces technical refinements that were previously unavailable in this form factor. The specification is not limited to x86 CPUs; it explicitly supports RISC processors, FPGAs, and GPGPUs. First samples featuring these alternative compute units were showcased at Embedded World 2023.

For the first time, OEMs can design heterogeneous server architectures that combine multiple compute and accelerator types within a single, officially specified ecosystem. The specification also defines a slave mode for modules, giving designers even more flexibility to integrate legacy components.

COM‑HPC Server modules target edge and fog servers that demand the high‑performance compute power of upcoming processors. With a maximum power budget of 300 W, the standard promises performance levels far beyond today’s 100 W‑limited COM Express Type 7 modules. Combined with the anticipated performance leaps from next‑generation CPUs, COM‑HPC is poised to handle enormous server loads.

Large footprints simplify cooling and enable the use of powerful CPUs or FPGAs. COM‑HPC Server offers two footprints: Size E (200 mm × 160 mm) with up to eight DIMM sockets for 1 TB of RAM, and Size D (160 mm × 160 mm) with four DIMM sockets for 512 GB. The figure below illustrates the two sizes.

click for larger image

Figure 1: COM‑HPC Server specifies two footprints—Size E with up to eight DIMM sockets for 1 TB of memory, and Size D with four sockets for 512 GB. The connector layout differs to prevent accidental mis‑mounting (Source: congatec).

The enlarged footprints allow full DIMM modules, delivering the bandwidth and capacity required for micro‑, edge, and fog servers. Size E supports up to 8 DIMMs (≈ 1 TB), while Size D supports up to 4 DIMMs (≈ 512 GB).

COM‑HPC Server defines 8 × 25 GbE ports and 65 PCIe lanes (Gen 4.0/5.0). One lane is reserved for the optional board‑management controller; the remaining 64 lanes feed peripheral devices, enabling I/O throughput of up to 256 GB/s.

With its extensive I/O matrix, COM‑HPC Server can host accelerators—GPGPUs, FPGAs, ASICs—as well as NVMe storage. Two USB 4.0 (40 Gb/s) ports on the Thunderbolt 3.0 interface and two USB 3.2 (20 Gb/s) ports complement the PCIe lanes, while additional USB 2.0, SATA, eSPI, SPI, SMB, I²C, UART, and GPIO interfaces round out the feature set. A dedicated 10 GbE port supports remote management.

COM‑HPC introduces a dedicated system‑management interface, under development by the PICMG Remote Management Subcommittee. It brings IPMI‑derived features for remote edge‑server module management, including reliability, availability, maintainability, and safety (RAMS). OEMs can extend functionality via an optional board‑management controller on the carrier board.

COM‑HPC Client – Bigger, Faster, More

While COM‑HPC Server targets new embedded edge‑server designs, the “classic” high‑performance embedded systems—historically based on COM Express Type 6—still dominate many markets. OEMs naturally ask whether COM‑HPC will render their existing designs obsolete, when to switch, and what benefits they can expect for customers.

Three Sizes

Both standards define three module sizes. COM‑HPC Client uses 120 mm × 160 mm (Size C), 120 mm × 120 mm (Size B), and 120 mm × 95 mm (Size A). The smallest footprint is almost identical to COM Express Basic (125 mm × 95 mm), demonstrating that COM‑HPC sits above COM Express and targets applications beyond its reach.

click for larger image

Figure 3: COM Express and COM‑HPC Client share three footprints; the smallest COM‑HPC Size A is virtually identical to COM Express Basic, illustrating the higher positioning of COM‑HPC (Source: congatec).

More Power

The specification allows a 200 W power budget—roughly three times the limit of today’s most powerful COM Express Type 6 modules. Both standards support SODIMM or soldered memory; COM‑HPC Client can host up to four SODIMM sockets, enabling larger memory capacities while COM Express already supports up to 96 GB.

More and Faster Interfaces

The key difference between the two standards lies in the connector and pin count. COM‑HPC Client introduces a 400‑pin connector (two connectors for a total of 800 pins), nearly twice the 440 pins of COM Express Type 6. This extra pin count accommodates more PCIe lanes, higher‑speed Ethernet, and additional USB ports.

COM‑HPC Client offers 49 PCIe lanes to the carrier board—twice the 24 lanes available in COM Express Type 6—while one lane remains dedicated to the board‑management controller. Two 25 GbE KR Ethernet ports and up to two 10 Gb Base‑T ports are integrated directly on the module. In contrast, COM Express Type 6 provides a single 1 GbE port with optional additional interfaces on the carrier board.

Graphics and Audio

Both standards support up to four displays via three DDI ports and one embedded DisplayPort. COM‑HPC replaces the HDA audio interface with SoundWire, a MIPI‑based standard that requires only two lanes (clock and data) and supports up to four audio codecs.

USB Bandwidth and MIPI‑CSI

COM‑HPC Client supplies four USB 4.0 ports (40 Gb/s) and four USB 2.0 ports, offering higher aggregate bandwidth than COM Express Type 6’s 4 × USB 3.1 and 8 × USB 2.0. It also adds two MIPI‑CSI interfaces for low‑cost camera connectivity, useful for situational awareness and collaborative robotics.

Other interfaces mirror COM Express: two SATA ports, 2 × UART, 12 × GPIO, 2 × I²C, 2 × SPI, and eSPI. COM Express includes optional CAN bus on the connector, a feature absent in COM‑HPC.

click for larger image

Figure 4: COM‑HPC Client and COM Express Type 6 interfaces differ mainly in the number of PCIe lanes, Ethernet ports, USB ports, and the yet‑to‑be‑specified remote‑management support (Source: congatec).

OEMs using COM Express can rest assured that the standard will remain viable for many years. PCIe generations are backward compatible, and designs based on Gen 3.0 will stay serviceable long after Gen 4.0 arrives. If your current interface specifications meet your needs, there is no urgency to migrate.

If your applications demand more than 32 PCIe lanes, full‑bandwidth PCIe 4.0, multiple 25 Gb/s Ethernet ports, or advanced remote‑management features, COM‑HPC is the logical choice. Otherwise, the adage “never change a running system” still holds.