Kỹ hơn về CPU của Panther Lake: nhân E-Core mạnh hơn, tăng thời gian dùng pin?

Key Concepts

• Intel Panther Lake: The new generation of Intel chips discussed in the video.
• Compute Tile: The main processing unit on the Panther Lake chip, containing CPU, GPU, and IPU.
• Intel 18A: The new manufacturing process used for the Panther Lake compute tile.
• Performance-cores (P-cores): High-performance CPU cores.
• Efficient-cores (E-cores): Power-efficient CPU cores.
• Kuga Curve: Codename for the new P-core architecture.
• DM (Diamond Mesa): Codename for the new E-core architecture, also used in server CPUs.
• Skyman & Crasman: Previous E-core architectures.
• Low Power Island: A dedicated cluster of E-cores designed for low-power tasks.
• Intel Thread Director: A technology that helps the OS scheduler allocate tasks to appropriate CPU cores.
• SOC Hybrid Core Configuration: Intel's term for the hybrid architecture combining P-cores and E-cores.
• Intel Xe-LPG: The architecture for integrated graphics in Panther Lake, referred to as Intel Xe3.
• Intel Xe3P: A future version of Xe3 for discrete GPUs.
• DirectX Cooperative Vector: A new technique for integrating AI into games.
• Intel XeSS: A supersampling technology for improving graphics performance.
• Multiframe Generation: A new technology to increase FPS in supported games.
• Pre-combine: A cloud-based system for pre-compiling shaders to reduce game loading times.
• Intelligence Bias Control Version 2: A driver feature to optimize GPU power for gaming.
• Tops (Tera Operations Per Second): A unit of measurement for computing performance, particularly for AI tasks.

CPU Enhancements in Panther Lake

Manufacturing Process and Core Architecture

• Intel 18A Process: The entire compute tile, including both P-cores and E-cores, is manufactured using the new Intel 18A process.
• P-cores (Kuga Curve):
  • Larger physical size compared to previous generations.
  • Each P-core has its own dedicated L2 cache for faster data access and reduced latency.
  • The L3 cache for the performance cluster (P-cores and E-cores) has been increased to 18MB, up from Lunalake, further reducing latency and improving fetch efficiency.
  • Improvements in prefetch and branch prediction to reduce redundant calculations and data fetching.
  • Focus on "reliable performance," meaning the ability to sustain high performance for longer durations.
• E-cores (DM - Diamond Mesa):
  • Based on previous E-core architectures (Skyman and Crasman) but with microarchitectural design improvements for better efficiency and performance.
  • Each E-core has access to a wider L2 cache compared to previous generations.
  • Improvements in prefetch and branch prediction for E-cores as well.
  • Low Power Island: A dedicated cluster of E-cores designed for very low-power tasks. This is a significant enhancement over Lunalake, which only had P-cores and a general E-core cluster.
    • Functionality: Handles lightweight applications like Word, Excel, PowerPoint, web browsing, chat, and email.
    • Benefit: Significantly reduces power consumption, leading to improved battery life. The Low Power Island E-cores consume even less power than the E-cores within the performance cluster.
  • Shared L2 Cache for E-cores: Four E-cores share a 4MB L2 cache.

Performance Gains and Power Efficiency

• Single-Threaded Performance: At the same power level, Panther Lake offers approximately a 10% performance increase over Lunalake.
• Power Efficiency: At the same performance level, Panther Lake can consume up to 40% less power than Lunalake and even ADOLc H-series chips.
• Multi-Threaded Performance:
  • At the same performance level, Panther Lake can achieve over 50% power savings compared to previous generations.
  • At the same power level, performance can increase by over 30%.

SOC Hybrid Core Configuration

• P-cores: Each P-core has its own dedicated L2 cache.
• E-cores: Four E-cores share a 4MB L2 cache.
• Performance Cluster: P-cores and E-cores within a performance cluster share the L3 cache.
• Low Power Island E-cores: In addition to L2 cache, these E-cores have a "memory size cache" (a secondary cache layer) before accessing RAM, further reducing latency.

Intel Thread Director Enhancements

• Role: Acts as a suggestion layer for the OS scheduler (Windows, Linux, Chrome OS), not a replacement. It provides "hints" to the OS on how to best allocate tasks.
• Customization: Laptop manufacturers can further tune Thread Director based on their specific hardware and software optimizations.
• Improved Task Classification Model: The new model uses AI and other techniques to more effectively classify tasks (e.g., demanding, moderate, power-saving) and allocate them to the appropriate cores.
• Enhanced Power Distribution: Improved power allocation to individual cores and better distribution of power between the CPU and integrated GPU.
• Heterogeneous Task Execution: Thread Director in Panther Lake is more efficient at running multiple different types of cores (P-cores, E-cores, Low Power Island E-cores) simultaneously.
• Task Allocation Flow:
  1. Priority 1: Low Power Island E-cores: Tasks are first attempted on the Low Power Island E-cores.
  2. Priority 2: E-cores in Performance Cluster: If the task requires more performance, it's moved to the E-cores within the performance cluster.
  3. Priority 3: P-cores: For the most demanding tasks, they are allocated to the P-cores.
• Comparison to Previous Generations:
  • Lunalake: Had P-cores and Low Power Island, but no dedicated E-cores within the performance cluster.
  • ADOLc: Had P-cores and E-cores, but no Low Power Island.
  • Panther Lake: Combines the strengths of both, with P-cores, E-cores in the performance cluster, and the Low Power Island.
• Real-world Use Cases:
  • Microsoft Teams: Primarily utilizes Low Power Island E-cores, with occasional shifts to regular E-cores.
  • Pro Office Benchmarks: Starts on Low Power Island E-cores, then moves to regular E-cores, and finally to P-cores for optimal performance.
  • Cinebench Benchmarks: Utilizes all core types: E-cores, P-cores, and Low Power Island E-cores.

Intelligence Experience Optimizer Software

• Functionality: A software tool for upcoming Panther Lake laptops that provides a performance slider.
• Slider Options:
  • Battery Life Priority: Favors E-core allocation.
  • Performance Priority: Favors E-cores in the performance cluster or P-cores.
• Performance Improvement: Can improve performance by up to 19% compared to not using the optimizer.

Integrated Graphics Enhancements in Panther Lake

Intel Xe3 (Xe-LPG Architecture)

• Architecture: Based on the Intel Xe3 architecture, referred to as Intel Xe3 in the video.
• Configurations: Available in two configurations: 4-core GPU and 12-core GPU.
  • 8-core CPU variants will have 4-core GPUs.
  • 16-core CPU variants will have 4-core GPUs.
  • The top-tier 16-core CPU variant will feature a 12-core GPU.
• Scalability: Designed to be highly scalable, meaning performance increases proportionally with the addition of GPU cores.
• Xe3P: A future version of Xe3 designed for Intel's discrete GPUs.

Key Improvements in Xe3

• Vector Engine: Enhanced for AI and graphics tasks requiring vector calculations, with a 25% increase in processing threads.
• Ray Tracing Units: Upgraded compared to previous generations.
• AI Performance: Faster processing of AI tasks, including large language models, on the integrated GPU.
• Tops Performance: Can reach up to 120 tops.
• Power Efficiency: More power-efficient than Lunalake's integrated graphics.
• Peak Performance: Up to 50% higher peak performance compared to Lunalake.
• Performance at Same Power: 40% higher performance compared to ADOLc H-series integrated graphics.

New Graphics Technologies

• DirectX Cooperative Vector: A new technique for easier integration of AI into games. Game developers can write AI features once, and the system will optimize them for different hardware (Intel integrated, Nvidia discrete, AMD integrated). This saves development time.
• Intel XeSS: A supersampling technology (similar to Nvidia DLSS) that renders at a lower resolution and uses AI to upscale the image, improving performance.
• Multiframe Generation: A new technology that generates one or more additional frames based on an existing frame, significantly increasing FPS in supported games. Intel claims it can double FPS or more in supported AAA titles.

Shader Compilation and Optimization

• Pre-combine: Intel is introducing a cloud-based system to pre-compile shaders for their GPUs.
  • Process: Shaders for specific games are compiled in the cloud by Intel. Users then download these pre-compiled shaders instead of compiling them on their device the first time they run the game.
  • Benefit: Reduces initial game loading times, allowing users to start playing faster.
• Intelligence Bias Control Version 2:
  • Functionality: The GPU driver detects if a game is running and monitors GPU utilization.
  • Optimization: If GPU utilization is low, the driver signals the firmware, which then informs the OS scheduler to reallocate tasks and power to the GPU for a better gaming experience.
• Increased Power for Heavy Tasks: The integrated GPU in Panther Lake will receive more power when running demanding tasks, leading to higher and more sustained performance.

Author's Perspective on Integrated Graphics

• Current Assessment: The author rates Intel's current integrated graphics as "average."
• Capabilities: Sufficient for basic esports and casual gaming.
• Limitations: Struggles with demanding AAA titles.
• Comparison: Falls short compared to AMD's integrated graphics and significantly behind Apple Silicon's integrated graphics.
• Expectations for Panther Lake: The author is cautiously optimistic about potential improvements but does not have overly high expectations.

Conclusion

The Intel Panther Lake generation introduces significant advancements in both CPU and integrated graphics. The new Intel 18A manufacturing process, coupled with the Kuga Curve P-cores and DM E-cores (including the Low Power Island), promises substantial improvements in performance and power efficiency. Intel Thread Director has been refined to better manage these hybrid architectures, leading to more intelligent task allocation. On the graphics front, the Xe3 architecture brings enhanced AI capabilities, ray tracing, and new technologies like DirectX Cooperative Vector and Multiframe Generation, aiming to improve gaming experiences. However, the author emphasizes that real-world performance, especially for gaming, will need to be tested once the chips are released.