The Stack Overflow Podcast Planning to Arm mobile devices with chips that handle AI
13 snips
Sep 12, 2025 Geraint North, an AI and developer platforms fellow at Arm, discusses the future of chip design and its importance for mobile devices. He dives into the challenges of optimizing large language models for edge computing, ensuring performance and efficiency. The conversation highlights how Arm's new Lumex CSS Platform empowers developers, particularly in mobile gaming, by integrating AI without compromising user experience. North emphasizes the need for collaboration between chip designers and game developers to push the boundaries of mobile technology.
AI Snips
Chapters
Transcript
Episode notes
Heterogeneous Cores For Mobile Balance
- Mobile SoCs mix big high-clock cores for peak work and smaller efficient cores for mid-range workloads.
- Designers choose combos that balance performance and battery for specific mobile use cases.
Area Is The Primary Mobile Constraint
- Mobile SoC designers allocate limited silicon area among CPU, GPU, NPU and optional modem blocks.
- Those area trade-offs determine which workloads the device will excel at and which it will sacrifice.
Memory Bandwidth Dominates LLMs On Mobile
- LLM workloads on mobile are often memory-bandwidth bound, not raw compute-bound.
- As a result, CPU runs of LLMs can match GPU/NPU performance if memory and DRAM speed are the bottleneck.