Fair CUDA Streams benchmark for pinned memory, asynchronous copies, events, overlap and Nsight Systems.
-
Updated
Aug 31, 2026 - Cuda
Fair CUDA Streams benchmark for pinned memory, asynchronous copies, events, overlap and Nsight Systems.
The MNIST classification problem is a fundamental machine learning task that involves recognizing handwritten digits (0- 9) from a dataset of 70,000 grayscale images (28x28 pixels each). It serves as a benchmark for evaluating machine learning models, particularly neural networks.
Multi-stream GPU edge detection over whole image directories — CUDA NPP plus two custom kernels, 103 images / 46 MP in 0.64 s on an A100, verified against an independent CPU reference that agrees on 99.9998% of pixels.
CUDA C++ practice project for RTX 4070 SUPER — explore GPU concurrency, pinned memory, and Nsight profiling. Includes SAXPY and 2D blur kernels to train optimization, stream overlap, and timing analysis for NVIDIA Developer Technology Engineering skillset.
Batch image enhancement & edge detection on the GPU: custom CUDA kernels (histogram equalization, separable Gaussian blur, Sobel) over 141 USC-SIPI images with multiple CUDA streams and a bit-exact CPU reference.
CUDA-accelerated multi-restart K-Means clustering with custom kernels and CUDA-stream concurrency, benchmarked against a CPU baseline (~22–55× speedup on a Tesla T4).
A CUDA C++ demo showing how to overlap data transfer and kernel execution using multiple streams and pinned (page-locked) host memory. This project illustrates asynchronous memcpy, event timing, and performance benefits of concurrent GPU execution — essential for building high-throughput pipelines.
Profile-first CUDA lab for killing GPU pipeline bubbles: stream overlap, CUDA Graphs, torch.compile, spec-decode overlap — Nsight-verified
To associate your repository with the cuda-streams topic, visit your repo's landing page and select "manage topics."