PulseAugur
EN
LIVE 10:01:02

New CUDA optimization strategies yield 1.41x speedup in neural network training

By PulseAugur Editorial · [2 sources] ·

This research paper details a comparative study of CUDA optimization strategies for shallow neural networks, focusing on forward and backward propagation. The study evaluated three stacked optimizations: tiled shared memory, pre-transposed weight matrices for coalesced memory access, and a fused MatMul+ReLU kernel. On a large dataset, the fully optimized implementation achieved a 1.41x speedup compared to the baseline CUDA version, reducing execution time significantly. AI

IMPACT Demonstrates significant speedups in neural network training through optimized GPU parallelization, potentially accelerating research and development.

RANK_REASON Research paper detailing optimization strategies for neural network training on GPUs.

Read on arXiv cs.LG →

AI-generated summary · Google Gemini · from 2 sources. How we write summaries →

New CUDA optimization strategies yield 1.41x speedup in neural network training

COVERAGE [2]

  1. arXiv cs.LG TIER_1 English(EN) · Rania Zitouni, Nadine Bousdjira, Sarah Hasnaoui, Amel Sadoun, Fatma Salhi ·

    GPU Parallelization Strategies for Forward and Backward Propagation in Shallow Neural Networks: A CUDA-Based Comparative Study

    arXiv:2606.30497v1 Announce Type: cross Abstract: We present a comparative study of CUDA optimization strategies applied to forward and backward propagation in a shallow neural network. Three stacked optimizations are evaluated: (1) tiled shared memory with bank-conflict eliminat…

  2. arXiv cs.LG TIER_1 English(EN) · Fatma Salhi ·

    GPU Parallelization Strategies for Forward and Backward Propagation in Shallow Neural Networks: A CUDA-Based Comparative Study

    We present a comparative study of CUDA optimization strategies applied to forward and backward propagation in a shallow neural network. Three stacked optimizations are evaluated: (1) tiled shared memory with bank-conflict elimination via +1-column padding, (2) pre-transposed weig…

RELATED ENTITIES

RELATED TOPICS