The Circuit EP 134: Apple iPhones and Tensor Cores, Synopsis Earnings, Hock's Comments at Goldman
12 snips
Sep 14, 2025 Innovation takes center stage with Apple's latest iPhone launch, showcasing the integration of tensor cores for enhanced AI capabilities. The hosts dive into Synopsys's disappointing earnings, exploring market reactions and the impact on their future. Discussions of Nvidia's new CPX system reveal the intense competition in AI processing. Additionally, the shift towards in-house connectivity chips highlights Apple’s strategy to reduce supplier reliance, all while navigating an increasingly volatile semiconductor market.
AI Snips
Chapters
Transcript
Episode notes
Apple Adds Tensor Cores To GPU
- Apple added neural (tensor) cores to its GPU across the A19 lineup to accelerate on-device AI matrix multiplication.
- This architectural shift lets developers run larger AI workloads on-device with lower latency and streamlined memory use.
GPU-Based AI Reduces Data Movement
- Running AI on the GPU avoids costly data movement between GPU and NPU and improves inference throughput.
- GPUs with tensor cores trade higher power for much faster, larger on-device inference than a small NPU alone.
Hands-On With The Thin iPhone Air
- Ben described holding the new iPhone Air and called it an engineering marvel that felt like the original iPhone 4 in design impact.
- He emphasized its extreme thinness and how most of the body is battery with the PCB in a small camera bump.