As we continue to churn through our news from last week’s 2024 Embedded World conference, it’s clear that one of the pervasive themes of the show was artificial intelligence (AI), particularly in edge computing devices.
Last week, industry leaders like Arm and AMD announced new computing solutions, while smaller startups like Semidynamics showed off innovative takes on AI subsystems.
Intel’s Dan Rodriguez (corporate VP and general manager—left) and Bassel Haddad (VP and general manager of NEX core and graphics and entry—right) at Embedded World 2024.
Here’s a look at some of the most noteworthy processing solutions announced at this year’s event by Google, Intel, and Imagination Technologies.
Google Debuts Axion Processors
Google has launched its new Axion processors to boost the performance and efficiency of its data centers. Built on Arm’s high-performance Neoverse V2 architecture, Axion CPUs boast up to 30% better performance than the fastest existing Arm-based cloud instances. They also offer substantial improvements in energy efficiency and performance—up to 50% and 60% better, respectively—over current generation x86 processors.
The Titanium workflow. Image used courtesy of Google
Axion integrates Google’s custom system, Titanium, to handle essential platform operations like networking and security, freeing up more resources for primary processing tasks. Titanium includes the custom-designed Titan security microcontrollers. The Titanium adaptor offloading card manages acceleration and virtualization services, and the Titanium offload processors offload network and I/O processing from the host. This integration allows Axion to efficiently address demanding applications such as machine learning, extensive data analytics, and multimedia processing, making it a powerhouse for general-purpose computing needs.
According to Google, Axion is set to be available to Google Cloud customers later this year.
Intel Releases Suite of Edge-Optimized Processors
Intel used this year’s Embedded World to announce a slew of different edge-optimized processing solutions, including the Intel Core Ultra, Intel Core, and Intel Atom processors—each designed to handle demanding AI, media, and graphics workloads at the edge.
The Intel Core Ultra processors feature an integrated system-on-chip (SoC) that merges the Intel Arc GPU and a neural processing unit (NPU) for improved generative AI and graphics performances in edge applications. Intel engineered these processors to deliver up to 5.02x better image classification inference performance compared to their predecessors.
The Intel Core processors combine an LGA socket with 13th-Gen Intel Core mobile processors and 3x more graphics units to achieve 2.57x improved graphics performance compared to the 13th-Gen Intel Core.
For more detailed deployments, the Intel Atom x7000C and x7000RE series processors offer substantial enhancements in deep learning inference capabilities, catering to industrial and telecommunications needs. These chips can handle more complex processing tasks efficiently, facilitating real-time AI applications without compromising power efficiency. For example, the x7000RE series is said to include up to 32 graphics execution units in a 6 W–12 W BGA package, enabling 9.83x improved image classification compared to the x6000RE.
Finally, the Intel Arc GPU for edge emphasizes flexibility and performance, allowing legacy Intel Core systems to boost their AI and graphics capabilities. This aligns with Intel’s strategy to provide robust, versatile solutions that support the increasing demands of edge computing and AI deployment across various sectors.
Imagination Adds to CPU Line for RISC-V-Based Devices
Joining Google and Intel’s processor releases, British semiconductor company Imagination Technologies unveiled the latest addition to its Catapul CPU lineup: the APXM-6200 CPU.
Designed to enhance the performance and security of RISC-V-based devices, the APXM-6200 CPU is a 64-bit, in-order application processor equipped with an 11-stage, dual-issue pipeline. The device’s architecture supports single-, dual-, and quad-core configurations, providing flexible scalability for various performance demands. Each core has individual power control, enhancing overall system efficiency and cache coherency, crucial for maintaining the speed and reliability of the processor during complex computations.
APXM-6200 performance benchmarks. Screenshot from Imagination Technologies
The device supports RISC-V vector extensions for AI capabilities. This feature, coupled with fast data-coupling for AI accelerators, positions the APXM-6200 as a powerful option for next-generation consumer and industrial devices, including smart TVs and smart home hubs.
Imagination claims its design results in a 65% improvement in normalized performance and a 2.5x improvement in normalized performance density compared to similar CPUs currently available on the market.
A Push for Edge AI Processing
Embedded World has always been a platform for companies to release their latest processing tech, and this year was no exception. With new solutions from Google, Intel, Imagination Technologies, and many others, the state of the computing industry is moving toward greater performance and efficiency to bear heavy AI workloads.
Check Out More Embedded World News
Renesas Rolls Out Entry-Level MCU With ‘Best-in-Class’ Power Consumption
