SYCL* is a promising programming model for performance and functional portability across different architectures and vendors. Intel® toolkits provide a collection of useful tools to develop, build, and optimize heterogeneous applications.

This video demonstrates a generic C++ implementation of a finite difference method-based thermal solver.

1. The binary is built and run to generate the baseline solution (temperature) and solver time.
2. The Intel® Advisor offload modeling feature is run to check for GPU suitability and possible offload sites. The resulting HTML file is then described at a high level.
3. Based on offload recommendations from Step 2, a SYCL version is produced with minimal changes.
4. Functional and performance portability on Intel® Xeon® CPU Max Series (formerly code named Sapphire Rapids) with SYCL versus C++ is confirmed by comparing the resulting temperature and solver time with those from Step 1.
5. Functional and performance portability are confirmed on Intel® Data Center GPU Max Series (formerly code named Ponte Vecchio). A gain on this GPU of around 8.1x versus Intel Xeon CPU Max Series is demonstrated.
6. Functional and performance portability are confirmed on an NVIDIA* A100 (80 GB). An Intel Data Center GPU Max Series gain of around 14 percent versus A100 is demonstrated.

Speaker

Dr. Amarpal Kapoor, software technical consulting engineer, Intel

Subscribe Now

Stay in the know on all things CODE. Updates are delivered to your inbox.

Sign Up

@IntelDevTools