Lecture Notes: M4 Processor Overview and Testing

Introduction

• M4 Arrival: Arrived just half a year after the M3 launch; first launched on the iPad.
• Expectations: Initial thoughts were an enhanced M3 version with improved NPU and N3E process.
• Reality: M4 proves its worthiness as a new name through architecture, performance, and energy efficiency tests.
• Testing Goals:
  1. Assess M4's architecture to gauge new performance limits.
  2. Evaluate energy consumption as the world’s first N3E process chip.

M4 Overview

• Manufacturing: TSMC's second-generation 3nm (N3E) process.
• CPU Improvements:
  • Micro-architecture improvements.
  • Peak frequency up to 4.5GHz on single-core.
  • Small core frequency at 2.88GHz.
• GPU:
  • Slightly optimized version of M3 GPU.
  • Frequency increased from 1.34GHz to 1.47GHz.
• NPU:
  • Significant enhancements claimed by Apple, especially in AI performance.
• Memory:
  • Uses LPDDR5-7500 memory instead of LPDDR5X.
  • Memory latency significantly improved.

CPU Micro-Architecture

• Decoding Unit:
  • Expanded from 9-width (M3) to 10-width.
  • Increased Dispatch Buffer and Scheduler queue size.
• Execution Units:
  • No increase in number; improvements made in instruction throughput.
• Additional Features:
  • SME Unit included, akin to Arm's AVX512.
• Performance:
  • Improved memory and cache performance, with memory latency reduced to 88ns.
  • Single-core frequency potentially at 4.5GHz.

Performance Testing

• SPEC2017:
  • Measured under both normal and extreme conditions (liquid nitrogen cooling).
  • Peak performance shows M4 20% higher than M3.
  • Power consumption at peak: 60% higher than M3 for 20% performance gain.
  • Energy efficiency sees minimal improvement, major gains attributed to architecture.
  • IPC (Instructions Per Cycle) increase close to 8% over M3.
  • Compared against Intel's 14900K, M4 outperforms in single-core by 23%.
• GeekBench 6:
  • Addition of SME support boosts scores.
  • Single-core score: Nearly 4,000 points (26% higher than M3).
  • Multi-core score: Close to 15,000 points.
  • Peak power consumption at 30W.
  • In power-saving modes, M4 shows some energy efficiency improvement.
• Applications in Tech:
  • M4 expected to elevate Macs significantly, iPads benefit in longevity and minor performance boosts.

GPU Analysis

• Performance and Energy Efficiency:
  • Peak power consumption similar to CPU increase (~30W).
  • Low-frequency energy efficiency shows slight improvement over M3.
  • Running mobile-specific tests reveal lower power consumption with competitive performance.
• Gaming Performance:
  • Handles Genshin Impact at high resolution (2560x1440), maintaining 60fps consistently.
  • Struggles with 120fps in sustained states.
  • Honkai Impact: Star Rail shows higher pressure but maintains performance better than M2.
  • Existing 3A games (e.g., Resident Evil 8) locked at 10W, not utilizing full potential.

iPad Pro Specifics

• Heat Dissipation:
  • Better heat dissipation compared to previous models.
  • Surface temperatures lowered, internally improved heat management.
• Battery Life:
  • Significant improvements in battery life, almost doubling in some cases.
  • OLED screens contribute major energy savings compared to mini-LEDs.

Conclusion

• Technological Implications:
  • Highlights the diminishing returns of process improvements at advanced nodes.
  • Apple may need to balance between pushing performance and maintaining energy efficiency.
  • Future impacts on A-series processors and the overarching strategy for M-series in Macs and iPads.
• Final Thoughts:
  • M4 sets a new performance standard but emphasizes the ongoing challenges in semiconductor advances.
  • The future of Apple silicon will likely focus more on architectural enhancements over bare process improvements.