Google's new Tensor G6 processor, built on ARM architecture, significantly underperforms Qualcomm's Snapdragon 8 Elite Gen 5 in traditional benchmarks, yet represents a fundamental shift in how mobile devices handle artificial intelligence. The Tensor G6 trails by 39% in single-core tasks and 67% in multi-core workloads according to Geekbench 6 testing, yet Google is deliberately pursuing a different strategy that prioritizes AI efficiency over raw computational speed.

The performance gap becomes even more dramatic in graphics-intensive tests. The Snapdragon 8 Elite Gen 5 achieved a score over 128% higher than the Tensor G6 in the 3DMark Wild Life Extreme Stress Test, while maintaining frame rates more than twice as high. In the AnTuTu v11 benchmark, Qualcomm's chip scored nearly 3.9 million points compared to the Tensor G6's 1.8 million, a 110% overall performance advantage.

Yet these numbers tell only part of the story. Google has explicitly stated that Tensor silicon is not engineered to win synthetic benchmark wars. Instead, the company prioritizes heterogeneous computing, a design philosophy where the CPU, GPU, Image Signal Processor (ISP), and specialized neural processing unit work in harmony to execute specific, complex tasks hyper-efficiently.

What Makes ARM's Approach to Mobile AI Different?

The true innovation in the Tensor G6 lies in its 5th-generation Tensor Processing Unit (TPU), a purpose-built neural engine designed specifically for running complex artificial intelligence models directly on the device. This specialized processor is what enables Google's signature "software magic," powering features like Live Translate, Magic Eraser, Real Tone computational photography, and on-device generative AI capabilities.

This represents a broader trend in ARM-based chip design. Rather than chasing maximum clock speeds, ARM architecture is increasingly being optimized for AI workloads through dedicated neural processing units. The NXP i.MX 95 processor, featured in Silex Technology's new EP-200N System-on-Module, exemplifies this shift with its hexa-core ARM Cortex-A55 architecture paired with an 8 eTOPS neural processing unit AI engine.

How ARM Architecture Is Powering Edge AI Beyond Smartphones?

• Industrial Automation: ARM-based processors with dedicated neural engines are enabling real-time machine vision and robotic control systems that can process AI models locally without cloud connectivity.
• Medical Imaging: Healthcare applications including portable medical devices and medical imaging systems are leveraging ARM's efficient neural processing units to perform diagnostic AI tasks with minimal power consumption.
• Intelligent Gateways: Edge AI systems built on ARM architecture are serving as intelligent intermediaries between IoT sensors and cloud infrastructure, processing data locally to reduce latency and bandwidth requirements.

The EP-200N module integrates 8 GB of LPDDR5 memory and 32 GB eMMC storage, operating across a temperature range of minus 40 degrees Celsius to plus 85 degrees Celsius. It includes Ethernet, CAN-FD, PCIe Gen3, and multimedia interfaces designed for demanding industrial and medical environments.

"The EP-200N combines Silex's expertise in embedded connectivity with NXP's latest i.MX 95 applications processor to provide OEMs with a scalable foundation for secure edge AI applications," stated Keith Sugawara. "Designed for healthcare and industrial automation applications, the EP-200N helps accelerate the development of intelligent edge systems requiring secure, reliable performance, and long-term availability."

Keith Sugawara, President and CEO of Silex Technology America, Inc.

ARM's neural processing unit strategy extends beyond consumer smartphones. The NXP i.MX 95 processor powers applications ranging from human-machine interfaces to intelligent gateways and machine vision systems. This diversification suggests that ARM architecture is becoming the foundation for distributed AI computing, where processing happens closer to where data originates rather than in centralized data centers.

Evaluation kits for the EP-200N will be available in September 2026, with developers able to request early access through the company's Early Access Program. Silex Technology will showcase the platform at Embedded World North America 2026, scheduled for September 22 through 24 in Anaheim, California.

The divergence between Qualcomm's raw-power approach and Google's AI-optimized strategy reflects a broader industry recognition that mobile and edge AI success depends less on absolute processing speed and more on architectural choices that align with how modern AI models actually run. ARM's flexibility as an instruction set architecture allows manufacturers to add specialized neural processing units tailored to specific workloads, creating a more efficient path to AI deployment than relying solely on general-purpose computing power.