Q&A with Christian Eder, congatec
Technology innovation is advancing rapidly in the embedded market. In this Q&A with Christian Eder of congatec, learn how the 11th Gen Intel Core processors and upcoming PICMG COM-HPC standard are able to take new developments into the future.
Q: Coinciding with the embedded launch of the 11th Gen Intel Core processors (Tiger Lake UP3) congatec has brought the first module based on upcoming COM-HPC standard to market. What is driving the need for this new standard?
A: The PICMG COM Express specification for high-end embedded computing is 15 years old now. It was immensely innovative and urgently needed at the time because the parallel PCI bus was becoming legacy and the PCI Express (PCIe) serial interface was to be the new measure of all things. Also, Ethernet was about to make the leap from 100Mbit Ethernet to 1Gbit Ethernet. The situation is very similar today with COM-HPC. The new specification is designed to integrate up to 8x 25Gbit Ethernet.
PCIe has already reached the 4th generation today, and the 5th with up to 32GT/s is in sight. Compared to the 1st generation of PCIe with a maximum of 2.5 Gtransfers/s (GT/s), it offers another major performance boost and is also covered by COM-HPC. In all likelihood, COM-HPC will even be able to cover the 6th generation of PCIe, which is still in preparation and expected to bring a further doubling of bandwidth using pulse-amplitude modulation with 4 voltage levels (PAM4).
This can only be tested when the PCIe specification is ready, of course. In addition to the performance per pin, COM-HPC also increases the total bandwidth from 440 to 800 pins, allowing more complex edge server processors to be integrated. This is a crucial aspect because edge computing, along with networking and storage equipment, is becoming a growing development market for embedded systems, which is gaining increasing importance fueled by the IoT trend with its industrial clouds and real-time capable fog computing.
Q: What does the new Intel Core processor technology bring to the table for this form factor?
A: Thanks to the new microarchitecture and 10nm SuperFin technology, the 11th generation Intel Core processors provide 23% more single-thread, and 19% more multi-thread performance compared to Intel Core Embedded processors of the 8th generation. What is more, the new Intel Iris Xe graphics also features the 10nm++ manufacturing process, which results in a 50% increase in packing density with 48 and 96 execution units (EUs), yielding nearly three times higher graphics performance compared to what the Intel Core Embedded processors of the 8th generation deliver. Beyond these significant performance increases, they are the first x86 embedded processors supporting the high-bandwidth PCIe Gen 4.0 and USB 4 interfaces. And it is exactly these features that make these processors an ideal candidate for our first COM‑HPC Computer-on-Modules conga-HPC/cTLU.
Q: What are the most demanding new applications for these new COM‑HPC modules?
A: The edge computing trend dominates new application areas. These include industrial and tactile IoT, machine vision and situational awareness, real-time control, collaborative robotics, and autonomous vehicles.
Also, all applications requiring real-time edge analytics and artificial intelligence (AI) with inference workloads are ideally served by the ‘total compute’ capabilities of the module’s CPU and GPU. And thanks to the massively increased bandwidth of the new Intel Iris Xe graphics, the modules are also a perfect fit for graphics intensive medical imaging and immersive digital signage sectors, as well as the industrial machine vision sector. Furthermore, the massive Ethernet and PCIe bandwidth for extremely fast data exchange predestines these COM-HPC modules also for computing nodes, network hubs, local fog data centers, and infrastructure appliances in the core network.
Robust central cloud data centers for critical government applications are also among the target applications. For all these applications, our COM‑HPC modules with the new Tiger Lake processors enable more powerful connected real-time systems in a rugged, durable and maintenance-free design.
Q: You have also made the new processors available on COM Express, conga-TC570. Won’t developers prefer to stick with the familiar form factor instead of making the leap to something new? Can you briefly explain the advantages of COM‑HPC? What are the main differences to COM Express?
A: COM‑HPC is the standard for new high-speed COM and carrier designs. It offers the necessary compliance for high-speed interfaces. COM Express does not do so, yet. COM‑HPC provides the option to implement nearly twice as many interfaces as COM Express. The classic COM Express connector is specified only up to the 3rd generation of PCIe. We can therefore use the PCIe bus of the 4th generation, which is becoming available with the new Intel Tiger Lake processors, only in PCI Gen 3 compatibility mode. There are, however, ongoing efforts to develop a new COM Express connector, with the same physical specifications but more powerful electronics. We have not tested it yet.
Q: Do developers have to fear that they’ll need to give their current projects a complete overhaul because they can no longer use COM Express?
A: Certainly not. The new high-speed Intel Core and AMD Ryzen processors will continue to be available on COM Express. This means that existing designs can be upgraded on the basis of COM Express for a long time to come. Especially if a new connector for COM Express is launched. Nevertheless, COM‑HPC offers clear advantages in terms of the wider functionalities and high-speed interface support and hence the long-term future perspective. Also, migrating from COM Express Basic to COM‑HPC Client Size A—the smallest module size of the new standard—will be fairly straightforward in terms of footprint.
Customers should therefore always check whether COM‑HPC is a better choice for them. After all, the extended remote management functions are highly attractive and increasingly important for IoT-connected distributed devices. Here, too, COM‑HPC sets new standards that are not defined for COM Express. However, customers must remember that COM‑HPC also requires new carrier boards.
So, the initial effort is invariably higher. But we are happy to advise OEMs without obligation on the design decision between COM‑HPC and COM Express. We have recently published whitepapers on this topic, which is available on the congatec landing page for the launch of the 11th generation of Intel Core processors at https://congatec.com/en/technologies/intel-tiger-lake-modules
Q: Now, a distinction is being made between COM‑HPC Server and Client modules. What exactly is the difference between the two module types, and what does this mean for the interfaces?
A: In general, we assume two application fields: Servers, which are usually headless, and clients, which offer comprehensive graphics support. These are the two main differences, which also affect the interfaces. While client modules need many signal pins for the graphics, we can use these in the server modules for more PCIe and Ethernet.
Next to the different interface specifications, the modules also vary in the form factor. The COM‑HPC Client modules are available in three sizes: Size A, B and C. The COM‑HPC Server modules are available in Size D and E. While the COM‑HPC specification allows all combinations, a Size E module with client pinout makes no sense today.
With increasing size, the space requirements to accommodate more memory go up, so that we can integrate up to 8 full DIMM memory modules for currently up to 1.0 Terabyte RAM in the largest server modules, while normal Size A assemblies will usually only offer space for two SODIMM sockets for 32GB DDR4 modules. So, a total of 64GB RAM – which is a comparable performance to COM Express Basic.
Q: Are there advantages that speak for certain processor architectures, or can developers choose freely between them?
A: The have a free choice. They just need to adhere to the specification. It includes only standard interfaces that are not specific to the different processor architectures. The COM‑HPC specification even allows a PCIe device mode, so it is equipped with a PCIe clock input. This makes it possible to develop for instance GPU or FPGA boards in an industrial COM‑HPC design. Even several CPU modules can communicate directly and with highest performance via PCIe.
Q: Many developers are focused on AI and IoT these days. How does COM‑HPC support this trend?
A: AI is often inextricably linked to parallel data processing, and this is exactly where the master/slave functions of the COM‑HPC specification can provide value by supporting the operation of several CPUs in parallel. Additionally, massively parallel computing units such as GPGPUs or FPGAs can be connected via PCIe lanes. As AI is often based on video data sources, COM‑HPC also provides native support of up to two MIPI-CSI interfaces.
IoT is supported with more Ethernet, and more complex remote management functions. Another important area are the more powerful multicore processors, which can host many virtual machines in parallel, and where we will also be able to consolidate numerous applications. What makes the congatec offering particularly attractive here, is the comprehensive support of the real-time RTS Hypervisor from our subsidiary Real-Time Systems.
Q: As mention earlier with the 11th Intel Core processor generation, developers can choose either COM‑HPC modules or COM Express for their designs, as congatec has equipped both module families with the same 11th generation Intel Core processor variants. When should a developer switch to COM‑HPC?
A: OEMs should use COM‑HPC if they want to start a new design or use interfaces that are not available on COM Express. In the case of the current 11th generation Intel Core processors, the interfaces that are not available on COM Express but supported by COM‑HPC are mainly USB 4.0, 2.5 Gbit Ethernet, SoundWire, and MIPI-CSI. Anyone who may need a still larger or more powerful PCIe or Ethernet interface selection with up to 25 Gbit Ethernet in the future, should also opt for COM‑HPC.
For developers of high-speed embedded systems, it’s worth bearing in mind that they can scale down more easily if they use a single standard, for example implement everything in COM‑HPC. And let’s not forget the extended remote management interface that will become available in the upcoming COM‑HPC launches. All good reasons for COM‑HPC. Otherwise – never change a running system. Also, because we are still to test a new connector for COM Express, which is expected to be available soon.
This congatec Decision Guide explains the differences between COM Express and COM‑HPC in detail. It is available for download here.
congatec | www.congatec.com