Innovating Edge Compute with Cloud Native Processors
In the last five years, more processing has moved to the edge. In 2018, around 10% of enterprise-generated data was created and processed outside a traditional centralized data center or cloud. By 2025, Gartner predicts this figure will reach 75%," according to Santhosh Rao, senior research director at Gartner.
As more processing moves to the edge, edge compute is evolving to be cloud native—from coffee machines counting beans on device to products like cameras that use sensors with high bandwidths, and communication services that require more and more compute capacity, but are restrained by power limits and a need to generate less heat.
In a recent webinar, Joe Speed, head of edge at Ampere, and Richard Pinnow, business development manager for embedded modules at ADLINK, sat down to discuss, “How Cloud Native Processors are Revolutionizing Edge Computing”.
You can watch the webinar in its entirety here:
Or read on to get a recap of the discussion along with Q&A topics and questions.
Cloud Native Advantages in Edge and Embedded Systems
First, let’s talk about predictable performance. Ampere Cloud Native Processors scale from 32 to 128 cores in a device, give more compute in any given power budget, and deliver best-in-class performance per Watt compared to x86 CPUs. The Ampere® Altra® Family of processors are designed to deliver uninterrupted, predictable performance.
Whether you have 32 or 128 cores, each core is single threaded, eliminating the noisy neighbor problem. At the edge, this translates to freedom of interference when running many different workloads on a single processor. For example, in robotics and autonomous driving there is a need to have perception pipeline, sensor fusion, path planning, and other workloads all running at the same time without impacting each other.
Scalability is another key feature of Cloud Native Processors. Cloud workloads are designed to scale out across many cores, many servers, and many racks. When you put these workloads and open-source software into a very dense compute—like the Ampere Altra processor—they’re scaled out and run alongside other workloads.
Finally, power efficiency and sustainability are core to the Ampere Cloud Native Processors. The way we've architected and developed the technology, and how we work with community to optimize entire solutions—including software—goes a long way toward a more sustainable future.
Unlock a More Sustainable Future
Ampere customers using our processors in the cloud have big ambitions around being carbon neutral and reaching net zero carbon impact. Ampere Cloud Native Processors help them get there, and we see that efficiency carried all the way out to the edge.
Recently, Jeff Geering and Patrick Kennedy at Serve the Home compared a 128 core Ampere Altra Supermicro telco edge server working with the Arm 5G lab to a cluster of Raspberry PI 4 compute modules.
They found that one socket of Ampere Altra Max was equal to 100 raspberry Pi 4 compute modules. That part we kind of already expected and the result wasn’t surprising to us. But what is interesting is the complete Ampere Altra system—the Supermicro Telco Edge server including redundant power supplies—was 22% more energy efficient than the Raspberry Pi 4 cluster.
Cloud Native Benefits for the Edge:
- More cores for any given size, weight, and power (SWaP)
In domains such as industrial controls, robotics, autonomous driving, transportation, and 5G this ability to have very high performance with low latency and very low jitter, even if multiple workloads running concurrently on the same SOC, is critical—as is the ability to scale linearly.
And with things like Kubernetes, you're managing all the cores in your pool. As you deploy your workloads to incremental cores, you get linear scaling, efficiency, sustainability, and the ability to pack more compute into your limited size, weight, and power (SWaP).
Efficiency and Performance at the Edge
Compared to x86, you can get triple the compute for the same energy budget with Ampere. Or to say it another way, you can get the same compute performance using 1/3 the energy. This is possible because of our efficient processor.
Improving efficiency without sacrificing performance is good for embedded systems where people keep adding more sensors, higher bandwidth sensors, and higher bandwidth communications to their computer while constrained by the existing space, power, and weight for the compute. For example, if you look at something like a 16 core Intel Xeon D — using just two-thirds of the power, Ampere can give you double the compute with our Q64-22— a 64 core 2.2 gigahertz part. If you look at our 128 core 2.6 gigahertz processor, you get over triple the compute for about the same power budget.
Computer on Module
What is Computer on Modules? Computer on Module (COM) or sometimes system on modules, are completely modular, embedded computers consisting of processors including chipset, main memory, pre-interfaced mass storage, and all essential peripherals ready to use on a standardized computer module form factor. There are many different shapes and performance classes available to the market.
Computer on Modules are application ready function blocks, which are ready to use in demanding applications. Customers can really focus on their core value proposition while reducing the design effort and risks and shorten time to market.
High Performance Computing (HPC) with Computer on Modules (COM)
Computer on Modules require a carrier board, which breaks out all module interfaces to standard peripheral connectors—since COM lacks the standard connectors for any input/output peripherals to be attached directly to the board. The carrier boards implement special I/Os reflecting your unique requirements, while the COM itself remains a COTS item. All customization is done on the carrier board. COMs give you the freedom to concentrate on your mission, where you bring your core values into the product without wasting time for complex CPU and SOC integration.
Ampere and ADLINK have joined forces to really create something new. This computer module is the most powerful and energy efficient COM module in the embedded market—scaling from 32 cores to 64, 96, 128 cores. The design features six dim memories with up to 768 gigabytes of memory ready for intensive workloads and demanding hypervisor. This is all on a single Computer on Module the size of an iPad Mini.
COM-HPC offers High Performance Compute. When it comes to memory-hungry applications HBCs are the perfect choice and major consideration for embedded systems. COM-HPC Server —in contrast to the more compact COM-HPC Client and COM Express in general—features DIMM memory and the best performance per dollar. Regardless of the core count you get on 64 lanes of PCIe and built-in networking.
COM-HPC Gives You More
You can interface a lot of peripherals on the carrier. For instance—multiple NVIDIA GPUs, Ethernet NICs and DPUs. You can include a lot of high speed NVME storage, and of course, FPGAs too for additional acceleration.
Next to the PCIe interfaces, it features up to four 10GbE ethernet interfaces, and one 1GbE in the base configuration. The standard operating temperature is zero to 60 degrees but ADLINK and its COM-HPC solutions provide extended temperature options by temperature screening. ADLINK tests -20°C to 75°C.
You might be wondering, what does it mean? Stretch the specification to reach extended temperatures? What is the reason to go in this direction? What is wrong with simply using commercial of the self 19-inch server solutions at zero to 40 degrees?
Commercial off-the-shelf server equipment offers commercial temperature range only, meaning zero to 60 degrees operation in best case. If you run this at minus 20°C or minus 40°C degrees, it is simply not booting reliably —which is the worst case for mission critical edge computing devices.
It has to work without excuses in any environment. And in most cases, those booting problems are caused by memory timing issues. ADLINK offers additional screening services to really make sure that everything is running smoothly—from minus 40°C and up to 75°C degrees. Every single module together with the memory is functionally tested and screened in a climate chamber in advance at different temperature levels.
Comparisons COM-HPC and COM Express
COM-HPC covers the area COM Express doesn't cover. COM-HPC client with up to 90 Watts complements COM Express Type 6 and COM-HPC server type with more than 200 Watts complement Com Express Type 7 use case.
What is the major difference between COM-HPC and Com Express? COM-HPC offers four times more PCIe bandwidth. If one has a look at the display interface, the COM-HPC server and COM Express type seven is intended for headless applications or server kind of use cases at the edge without display interfaces. The COM-HPC client and COM Express Type 6 features up to four times 4K display interfaces.
These are the two use cases within COM-HPC uses standard memory for the best tradeoff between cost and performance. COM-HPC grants you five times more ethernet bandwidth plus additional NIC cards you can apply to the carrier board and 10GbE Ethernet right from the COM-HPC client module in addition to sophisticated IPMI remote management BMC capabilities, not available on Express.
And lastly, COM-HPC offers way more PCB space and higher electrical design power to host even the most demanding next generation CPUs and SOCs. In the past, this was not doable using COM Express Type 7. CPUs like the Ampere Altra offering higher computing performance, higher memory capacities, and higher bandwidth interfaces are the main technology driver to COM-HPC.
COM-HPC and the Ampere Developer Platform
Our COM-HPC modules come along with a very powerful evaluation ecosystem to enable clients to get started quickly. The back plane is fully integrated in our reference system and teamed up with top notch NVIDIA GPU solutions. Our reference carrier can host three NVIDIA GPUs with PCIe x16.
ADLINK has a very strong collaboration with NVIDIA in place for many years already, so this is unique. We make sure that you get very robust graphics and AI ML inference support on your way to design your Ampere Altra product into your application. The system is ready for the most demanding CPUs and SLCs coming up in the next years.
Ampere has very good optimizations for all the major frameworks like TensorFlow and PyTorch. Ampere Altra also has strong GPU support. In fact, NVIDIA recently announced their cloud gaming platform with Ampere Altra 128 core processors paired with up to four A16 GPUs.
Ampere Altra Developer Platform was developed by ADLINK originally as the reference platform for software-defined vehicle SOAFEE special interest group—the scalable open architecture for embedded edge. It’s used by all the usual automotive suspects: Continental, Volkswagen, and the whole automotive community. It's also used within the Autoware Foundation, which is open source, autonomous driving, and the things that they do for the Mobility in Harmony EV consortium, led by Foxconn.
AWS Automotive uses it as a developer workstation and for live demos at conferences. And 19 automakers have this Ampere Developer Platform for things that they're doing. It's one example application of the COM-HPC Ampere Altra module.
This is compatible with SOAFEE, and all the software defined vehicle elements, like taking the open-source containers and virtualization and working to make automotive functionally safe. Another example of this is rugged computers for passenger trains from Seven Star Lake using the COM-HPC module. They’re doing passenger safety and information systems with the cameras, displays, and all these things in the train as it rolls on the rails.
To get more details and hear the entire Q&A conversations, watch the webinar in its entirety here:
Ready to get started? Talk to our expert sales team about partnerships or get more information or trial access to Ampere Systems through our Developer Access Programs. Visit ADLINK and check out the I-Pi website.
Built for sustainable computing in the cloud and on the edge, Ampere’s Cloud Native Processors deliver predictable high performance, platform scalability, and power efficiency unprecedented in the industry.