Monthly Archives: February 2024

For 30 years, PICMG has been shaping the embedded computing industry with standards. However, the organization is by no means resting on its laurels. Quite the opposite: with ModBlox7, some member companies have developed the first global standard for industrial box PCs.

Under the umbrella of the PCI Computer Manufacturers Group (PICMG), several manufacturers have been working on a new standard for modular industrial PCs: ModBlox7. It is intended to combine the advantages of scalable systems such as CompactPCI with cost-saving box PCs. As a flexible box PC system, ModBlox7 is based on units that are defined in terms of height and depth and can be extended in width as required. At the Markt&Technik roundtable, experts explained why a modular standard is needed in view of the large installed base and the long history of use of box PCs.

Matthias Beer, Managing Director at Ci4Rail, played a key role in the implementation of the new standard alongside Bernd Kleeberg, Managing Director at EKF Elektronik. When asked about the background to the idea of ModBlox7, Beer cites the increasing demand for small, space-saving, and lightweight industrial box PCs. The challenge for developers is interoperability between different PCs and a lack of flexibility. 

“Developers have to work with devices they get in shops, regardless of how they are configured. We want to combine the advantages of modular systems with those of a box PC,” explains Beer. Standardizing box PC technology is a completely new idea; often, only the technology used in the box PC is standardized, for example, COM Express modules. For this reason, it was easy to start something new, says Beer.

Thomas Kaminski, Director Product Sales and Marketing Management at Advantech, draws a comparison with the so-called Intel NUC standard. It is the only example where a company has attempted to develop a standard for industrial computers that addresses several vertical markets. “Other companies are more focused on a specific market with a standard product family,” continues Kaminski. 

Interoperability as a Major Advantage

The wide range of industrial computers available on the market raises the question of what the advantages of ModBLox7 are compared to off-the-shelf box PCs. Matthias Beer says: “What all standards achieve is to provide interoperability for the user, as well as for suppliers who manufacture products according to the new standard. This means that developers can put together a system from several suppliers. Conversely, manufacturers of ModBlox7 components can exchange and interact with each other—not everyone has to design every component themselves,” explains Beer.

Advantech is one of the largest manufacturers of industrial computer technologies such as box PCs or computer modules. The company manufactures proprietary or custom industrial PCs, but so far not in the new ModBlox7 form factor. Thomas Kaminski explains the reasons for this: “Firstly, we serve completely different markets with different requirements, and we also want to be flexible when it comes to system configuration. We also have to meet the demands of the market, for example in terms of certifications or thermal management. This often leads to a certain size, a certain housing, and a certain shape of PC. I very much doubt that all of this can be accommodated in a single standard for the various markets. However, we are keeping a very close eye on standardization,” says Kaminski.

For Matthias Beer, it was never the intention to force every application into a specific performance class or internal standard, as this was simply not possible. “It was never the intention to revolutionize the market for box PCs,” explains Beer. “Rather, we want to appeal to users of control or diagnostic systems, especially for the rail sector, which is where Ci4Rail comes from,” reports Beer.

Increase Availability with ModBlox7

Another point that is currently the subject of heated debate in the field of industrial computers is processor technology. This is because they are becoming ever more powerful and energy-efficient, which is leading to new innovations on the market. “With ModBlox7, we are addressing low-end to high-end Arm as well as x86 processors up to 20-25 W. However, the standard does not provide for clusters in which two or three CPUs can be stacked. ModBlox7 is more about redundancy, for example in terms of availability,” says Beer. 

Timo Korhonen, Chief Engineer Control Systems Division at European Spallation Source ERIC, shares his experience from scientific applications: “There is always a gap between systems with high and low performance requirements,” says Korhonen. You end up with lots of different box PCs or programmable logic controllers (PLCs) in the lab that don’t fit together, he explains. Matthias Beer likes the description: “ModBlox7 fits perfectly if someone wants to do general-purpose computing and have a normal operating system”.

Final Steps of the Specification Process

Bernd Kleeberg explains which points of the specification are currently being worked on and which points still need to be discussed. “With regard to ModBlox7, we have reached the end of the specification process. The technical work is complete. We just need to take the final administrative steps and expect to be ready in the first quarter of 2024. We would like to present the finished specification to the community at embedded world 2024.”

When asked about the service life of ModBlox7, Matthias Beer says it depends very much on the market. “We are active in the rail business, where a service life of at least 15 years is required. We have to be able to supply and change our product for ten years or more,” adds Beer, unlike in the automation sector, for example, where innovation cycles are shorter. The standard helps here because it offers the possibility of iterating individual parts.

There is also the question of how to achieve higher performance from ModBlox7 PCs and whether documents need to be expanded or provided for this. Matthias Beer says: “What the standard covers are additional cooling options. We are currently talking about including graphics processing units (GPUs) up to a certain level—not the 90 or 200 W versions, but up to about 30 watts, which are suitable for smaller computer vision applications or similar. We want to cover edge AI applications with this,” explains Beer.

AI Applications and ModBlox7: Do They Go Together?

Edge AI is a good keyword, as more and more AI applications are entering the market for embedded computing technology. And the question arises as to whether the opportunities on the AI market are not being taken away by standardization, as the necessary flexibility could be lost in the process. Brandon Lewis, Marketing Officer at PICMG, says: “When we talk about AI, everyone immediately thinks of GPUs, and the first thing that comes to mind is Nvidia. Of course there are other GPUs, and the question is whether GPUs can thermally fit into a small form factor standard like this,” says Lewis. Matthias Beer notes that applications are increasingly networked and that the connection between the cloud and edge devices is much more stable. This makes it very easy for developers to distribute AI applications between the two, explains Beer. 

Thomas Kaminski sees the opportunities in AI as one of Advantech’s biggest goals for the next few years. First and foremost, this is about capturing information and processing it at the edge. The next generations of Intel processors are very well suited for this. They have a Neural Processing Unit (NPU), scaling of the core components for more computing power, and AI efficiency. This opens many doors for edge computing. And Intel is also starting to launch its GPU cards. “We have specialized in AI at the edge with the latest generation of Intel and AMD technology,” explains Kaminski. 

MicroTCA Receives Update

In addition to ModBlox7, there are other PICMG standards that are along the same lines, including CompactPCI Serial and MicroTCA. MicroTCA emerged in 2006 from its predecessor AdvancedTCA and was mainly developed for the telecommunications industry for systems with high power density. In 2024, the question is to what extent the standard will play a role at all.

Timo Korhonen can speak from the perspective of scientific applications. Here, MicroTCA is used in areas that primarily process high-frequency (HF) signals, measure electromagnetic fields, or similar. He also sees a growing area of application in image processing. Jess Isquith, President of PICMG, says: “The scientific community has adopted the specification worldwide and new applications are constantly being added. A new version of the specification was even published in 2023. There is a high level of acceptance of the standard, particularly in Asia and Europe,” says Isquith. Bernd Kleeberg disagrees: “We don’t use the standard. One of the main reasons for this is that the administrative effort and associated costs are too high. That’s why we prefer other standards in industrial applications.” 

As MicroTCA was originally developed for telecommunications applications, the question currently arises as to whether it is necessary to revise the standard with regard to 5G and 6G. Jess Isquith says: “There is a problem with the connectors, because each connector can only transmit a certain amount of data. So we have to change the specification in terms of the connectors. This is a new specification because the backwards compatibility is no longer given. So we are creating a new set of specifications,” summarizes Isquith. 

There has also been a MicroTCA Technology Lab for several years, which is funded by the German government. “The employees are very committed and develop frameworks, standard products, and reference designs,” Isquith adds. 

Timo Korhonen adds: “I have been working on this project committee for ten years. We rely on a lot of input from outside, from industry, and from our partner laboratories across Europe. The dilemma here is that, on the one hand, you want to be at the cutting edge of technology, but on the other hand you have to be able to maintain the system because you can’t replace everything when something new comes along. So the longevity of the standard is very, very important,” says Korhonen.

CompactPCI Serial as a Perennial Favorite

Another PICMG standard also aimed at modular systems is CompactPCI Serial, which was ratified in 2011. In contrast to classic CompactPCI, CompactPCI Serial no longer uses the parallel Peripheral Component Interconnect (PCI) bus, but instead relies on modern high-speed serial interfaces.

The specification is soon to be extended to 4th generation PCI Express, explains Bernd Kleeberg, one of the initiators of the standard. Kleeberg adds: “The new specification is almost ready and we will publish it in the near future. Among other things, it includes an extension of the backplane connections to 25 or 40 Gigabit Ethernet and the USB ports to up to 10 Gigabit Ethernet. The connection for the CPU card will be changed, but the form factor will remain Horizontal Pitch (HP), so developers can use peripheral cards as before,” explains Kleeberg. In addition, work is already underway on the next version of the specification, which will support up to PCIe Gen 5. However, the form factor will have to be adapted somewhat due to the connector size, says Kleeberg.

“CompactPCI Serial is an interoperable modular standard that fulfills all the objectives of PICMG,” explains Jess Isquith proudly. However, the standard is not as widespread as expected. With the new version of the specification, PICMG has the opportunity to conquer the market anew. The standard is alive and only needs to be technically revised. 

There is also a new organization, the “Open Group”, which is working on new applications for CompactPCI Serial. “120 companies and organizations have come together to develop a forum for open process automation and to consider how we can introduce standards in this industry. New members such as Exxon Mobil, Schneider Electric, and a handful of others who have never been involved with PICMG are leading by example. We are seeing more and more potential applications for CompactPCI Serial. Other members are also getting involved and contributing their expertise to the new group. This is a very interesting new experience for all PIGMG member companies,” Jess Isquith concludes the discussions.

This article was originally published in Markt und Technik, issue 7/2024: https://wfm-publish.blaetterkatalog.de/frontend/mvc/catalog/by-name/MUT?catalogName=MUT2407D

PICMG: Can you describe your work outside of PICMG as well as the role you played in development of the COM-HPC specification? 

JENS: I am currently engaged in research at Bielefeld University, focusing on the development of heterogeneous and reconfigurable computing technologies for a wide range of applications. These include the Internet of Things (IoT), edge computing, cloud computing, and high-performance computing (HPC).

My involvement with the early stages of the COM-HPC specification centered on leading the development of the Platform Management Interface Specification and expanding the Embedded EEPROM Specification for COM-HPC. 

PICMG: You recently raised awareness about the CXL standard within the COM-HPC community. What is CXL and why is it relevant for PICMG COM-HPC developers and users?

JENS: CXL, or Compute Express Link, is a high-speed, high-capacity interconnect standard that facilitates efficient communication between CPUs, memory, and peripherals using the PCIe physical layer. Its support for cache coherency, disaggregation, and scalable architectures makes it a compelling choice for modular form factors like COM-HPC, driving its popularity among developers and users seeking advanced computing solutions.

PICMG: Why is CXL 3.1 significant in the context of COM-HPC? What use cases or capabilities will it drive in the COM-HPC ecosystem?

JENS: CXL introduces features that cater to the demanding requirements of cloud and high-performance computing systems. Its emphasis on scalable architectures, disaggregation, and cache coherency is particularly relevant for COM-HPC, offering the potential to revolutionize the way modular computing platforms are designed and utilized. 

The integration of CXL into COM-HPC could facilitate the development of more sophisticated computing solutions, enabling the seamless coupling of specialized accelerators and the establishment of cache-coherent multi-socket systems. These advancements promise to unlock new possibilities for COM-HPC applications, ranging from data-intensive analytics to AI and machine learning workloads, driving innovation in modular computing technologies.

PICMG: Given that CXL targets PCIe, has it been compatible with COM-HPC to date?

JENS: The relationship between CXL and COM-HPC is fundamentally influenced by CXL’s reliance on the PCIe physical layer for connectivity. This means that while direct compatibility between previous versions of CXL and COM-HPC has not been explicitly defined, the architectural underpinnings allow for potential integration. 

The absence of CXL in the current COM-HPC specification, coupled with the lack of support in existing modules, suggests that the integration of CXL represents a forward-looking opportunity for enhancing COM-HPC. Such integration is anticipated to require minimal modifications to the specification, paving the way for future advancements in modular computing.

PICMG: What does the COM-HPC community need to know about the CXL market or technical requirements to capitalize on the opportunity?

JENS: To fully leverage the potential that CXL brings to the COM-HPC community, it is crucial to understand the intricacies of CXL’s market dynamics and technical specifications. This involves a deep dive into the architecture of CXL, including its device types—such as Type 1 for I/O devices, Type 2 for cache-coherent devices, and Type 3 for memory expander devices. Additionally, understanding the topology options that CXL supports, including switch-based topologies for larger, more complex systems, can empower developers to design COM-HPC solutions that are both innovative and future-proof. 

Staying abreast of the evolving CXL specifications and market trends will enable the COM-HPC community to identify new opportunities for integration and application, ensuring that COM-HPC modules remain at the forefront of technological advancement.

PICMG: What are you and Bielefeld University doing with respect to CXL today?

JENS: We are working on integrating CXL within the RISC-V ecosystem, a venture that holds promising implications for the future of computing architectures. Our work focuses on the development of innovative bridge technologies that facilitate communication between the RISC-V Coherent Hub Interface (CHI) and CXL, using FPGA-based modules, which we refer to as microservers. 

This endeavor is not just about bridging two technical standards; it’s about creating a foundation for next-generation computing platforms that can seamlessly integrate diverse processing and memory resources. By developing these bridges, we aim to enable more efficient, scalable, and flexible computing architectures that can cater to the demanding requirements of modern applications, ranging from AI and machine learning to big data analytics.

PICMG: Where can interested parties go to find more information on CXL?

JENS: The CXL Consortium website serves as the primary repository of knowledge. This platform not only provides access to the official CXL specifications and technical documents, but also offers insights into the latest developments, industry adoption stories, and educational resources.

More Information:
• Compute Express Link: https://computeexpresslink.org
•*PICMG COM-HPC Overview: https://www.picmg.org/openstandards/com-hpc
•*PICMG Platform Management Interface Specification: https://www.picmg.org/product/com-hpc-platform-management-interface-specification