oneAPI with SYCL gives Software Portability including Nvidia
summary
OneAPI uses SYCL as a single-source C++ standard and offers 90% automated conversion from CUDA via the dpct / open-source cyclomatic tool, with the remaining ports flagged for manual work. Codeplay supplies oneAPI plugins for NVIDIA and AMD GPUs so SYCL code runs back on existing CUDA/HIP hardware; performance on many kernels is roughly equivalent. The stack also targets CPUs, FPGAs and RISC‑V RVV vector accelerators for heterogeneous deployments.
AI-written summary · about
chapters (12)
- 0:00Introduction to Codeplay
- 0:34The Cuda ecosystem
- 1:30Introduction to SYCL
- 2:29SYCL and hardware support
- 3:33oneAPI framework
- 4:42Scalability and use cases
- 5:40Migrating from Cuda
- 7:27Incremental porting strategy
- 9:06Tools and plugin support
- 11:00Performance benchmarks
- 13:12Industry open standards
- 16:05Conclusion and call to action
related talks
41mSYCL building blocks for C++ libraries
01h 00mHeterogeneous Programming in C++ with SYCL 2020
01h 07mHeterogeneous Modern C++ with SYCL
58mKhronos APIs for Heterogeneous Compute and Safety: SYCL and SYCL SC
25mDeep Dive: Accelerating Neural Networks using RVV and Open Standard Software
25mLeveraging AI Ecosystem for Portable and Sustainable GPU Kernels in HPC
43mTensorflow on open source GPUs
42mOpen Source is Better Together- GPU Python Libraries Unite