CSE539 - Advanced Comp. Arch. @PSU - Term project presentation

Step-by-Step Tutorial: Understanding Memory Processing Approaches in Computer Architecture

1. Introduction to Memory Processing Approaches:

   • The project focuses on memory processing approaches in computer architecture.
   • There are two significant issues: memory technology develops slower than processor technology, and architecture growth in scale and memory capacity.

2. Performance Gap and Research Directions:

   • The performance gap between processor and memory technology leads to latency issues and bandwidth limitations.
   • Current research directions are shifting towards Memory-Centric Computing approaches for increased performance and energy efficiency.

3. Classification of Memory Wall Solutions:

   • Works to solve the memory wall are classified into three categories: memory processing, near memory processing, and processing in memory approaches.
   • Researchers optimize commercial memory technology, exploit new memory technologies, offload computations to memory components, and conduct computations inside memory components.

4. Comparison of Approaches:

   • In-place approaches show promising energy efficiency but have constraints.
   • Utilizing bit lines in SRAM technology for computations and rearranging existing elements for logical operations.

5. Operant Locality and Cache Operations:

   • Addressing operant locality challenges by extending cache and conducting operations with regular access behaviors.
   • Implementing logical and functional operations like copy, search, and compare using cache operations.

6. Stream-Based Memory Access Specialization:

   • Exploring richer ISA semantics for memory access patterns, focusing on stream-based approaches for performance acceleration.
   • Implementing stream configurations, stream steps, and stream ends in ISA extensions for efficient stream processing.

7. Stream Specialized Processor Architecture:

   • Compilation process involves recognizing streams, selecting stream candidates, and generating code for stream-based operations.
   • Implementing stream specialized processor architecture with iteration maps, stream load/store buffers, and stream engine for efficient stream processing.

8. Memory Processing for Irregular Workloads:

   • Addressing irregular workload challenges by conducting computations where data resides in memory hierarchy.
   • Introducing memory services and memory service elements for invoking computations in near-memory components.

9. Stream Floating for Centralized Computing:

   • Extending stream-based approaches to proactively conduct computations throughout the memory hierarchy.
   • Implementing stream engines on core, L2, and L3 cache slices to float streams and reduce latency and bandwidth demands.

10. Reconfigurable Computing and Future Directions:

    • Utilizing last-level cache as a lookup table for reconfigurable computing with minimal system changes.
    • Achieving performance speedup and energy efficiency improvements with reconfigurable cache architecture.

11. Conclusion and Future Work:

    • Exploring different approaches of near-memory computation for enhanced performance and energy efficiency.
    • Learning from experiments and papers to further advance memory processing approaches in computer architecture.

12. Acknowledgment:

    • Thank you for listening to the presentation on advanced computer architecture focusing on memory processing approaches.