Vendor-Neutral GPU Programming in Chapel
summary
Chapel compiles the same high-level code for CPUs and GPUs and runs unchanged on a laptop, AWS T4 nodes, or Frontier, claiming vendor-neutral support for NVIDIA and AMD (Intel planned). Examples show simple array promotion becoming kernel launches via do on here.gpus and parallel tasking with coforall, verbose GPU diagnostics reporting kernel counts and bytes moved, multi‑GPU runs across four T4s, and a MATLAB-to‑Chapel convolution that drops runtimes from days to ~10–20 minutes.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 1:45Chapel’s GPU role
- 3:32GPU support and apps
- 7:27CPU to GPU conversion
- 13:09Single to multi-GPU scaling
- 17:24Ocean biodiversity with GPUs
- 20:58Frontier performance
- 22:02Getting started with Chapel
- 23:10Q&A: Chapel vs CUDA
related talks
29mBeyond CUDA: GPU Accelerated Computing on Cross-Vendor Graphics Cards with Vulk...
59mUsing Template Magic to Automatically Generate Hybrid CPU/GPU-Code
48mBridging the Gap: Writing Portable C++ Programs for CPU and GPU
53mIntroduction to C++ AMP (GPGPU Computing)
01h 06mFoundation of GPU Programming
25mLeveraging AI Ecosystem for Portable and Sustainable GPU Kernels in HPC
50mFrom Magic to Mechanics: Demystifying GPU Programming
34mGPU as a Service Over K8s: Drive Productivity and Increase Utilization