A new research paper introduces Space Filling Curves (SFC) as a method to simplify and optimize Communication-Avoiding General Matrix Multiplication (CA-GEMM) for high-performance computing and deep learning. This approach achieves platform-oblivious and shape-oblivious matrix multiplication schemes with improved data locality. The SFC-CA GEMM algorithm demonstrates provable asymptotic communication optimality and outperforms vendor libraries by up to 5.5x on x86 and Arm platforms. Its application in LLM inference prefill shows speedups of up to 1.85x, and in distributed-memory matrix multiplication, it achieves speedups of up to 2.3x. AI
IMPACT Optimizes LLM inference prefill, potentially accelerating deployment and reducing compute costs for large language models.
RANK_REASON Research paper detailing a novel algorithmic approach to matrix multiplication. [lever_c_demoted from research: ic=1 ai=1.0]
- Arm Holdings
- Deep Learning
- Evangelos Georganas
- GEMM
- high-performance computing
- LLM inference
- SFC-CA GEMM
- Space Filling Curves
- x86
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →