Read original ↗

paperarXivTrust 82 · PrimaryPublished 1mo agoLive · 1mo ago

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 weight matrices for coalesced global memory access, and (3) a fused MatMul+ReLU kernel that eliminates intermediate global-memory round-trips. Experiments on an NVIDIA Tesla T4 (CUDA 13.0) across three dataset sizes show that the fully optimized implementation achieves a 1.41x speedup over the baseline CUDA version on the

Lineage graph

Paper → model → repo connections mined from source citations (Tier-1 exact match).

Why these links exist

Every edge carries a method, confidence, and the source snippet that justified it — so bad links are debuggable.

LinkedLinked via unknownGoing from single GPU to dual GPU is nice but not in the way I expected →
LinkedLinked via unknownShow HN: NanoEuler – GPT-2 scale model in pure C/CUDA from scratch →
LinkedLinked via unknownUbuntu, CUDA, llama.cpp , nvcc versioning →
LinkedLinked via unknownHow NVIDIA’s Inference Software Stack Powers the Lowest Token Cost →
PossiblePossibly related (embedding) · 47%NexusGPU/tensor-fusion →
PossiblePossibly related (embedding) · 50%NVIDIA/cuvs →
PossiblePossibly related (embedding) · 47%pytorch/pytorch →
PossiblePossibly related (embedding) · 48%NVIDIA/physicsnemo →

Covers

newsGoing from single GPU to dual GPU is nice but not in the way I expected newsShow HN: NanoEuler – GPT-2 scale model in pure C/CUDA from scratch newsUbuntu, CUDA, llama.cpp , nvcc versioning

Covers (incoming)

newsHow NVIDIA’s Inference Software Stack Powers the Lowest Token Cost newsUltra budget 20GB vram with 448GB/s for $100 bucks.newsAlternative(s) to run CUDA on non-Nvidia hardware newsPyTorch model running 170x slower on T4 vs A100. What could cause a bottleneck this extreme? [D]

Implements (incoming)

repoNexusGPU/tensor-fusion repoNVIDIA/cuvs repopytorch/pytorch repoNVIDIA/physicsnemo repoNVIDIA/TransformerEngine repouccl-project/uccl repoNVIDIA/raft repoBBuf/how-to-optim-algorithm-in-cuda

Related across the graph

repouccl-project/uccl repoNVIDIA/cuvs newsUltra budget 20GB vram with 448GB/s for $100 bucks.repoNVIDIA/TransformerEngine newsHow NVIDIA’s Inference Software Stack Powers the Lowest Token Cost repoNVIDIA/physicsnemo repoNVIDIA/raft repoNexusGPU/tensor-fusion newsUbuntu, CUDA, llama.cpp , nvcc versioning repopytorch/pytorch repoBBuf/how-to-optim-algorithm-in-cuda newsPyTorch model running 170x slower on T4 vs A100. What could cause a bottleneck this extreme? [D]newsShow HN: NanoEuler – GPT-2 scale model in pure C/CUDA from scratch newsAlternative(s) to run CUDA on non-Nvidia hardware newsGoing from single GPU to dual GPU is nice but not in the way I expected

Topics