At Intel Vision 2022, Codeplay Demonstrates Strategies
for Migrating Software Assets from Proprietary CUDA to Open Standards and
multi-platform SYCL and oneAPI
Using automated tools, open-source software, the SYCL programming model and oneAPI to enable multiple processor targets were topics covered by Codeplay experts at Intel Vision 2022. This includes accelerator processor targets from Nvidia, Intel, and beyond. The two-day event was hosted live May 10-11 outside Dallas, TX, and virtually to a global online audience.
In a recorded presentation, the Codeplay team explains how to future-proof software development strategies for high performance computing (HPC) and artificial intelligence (AI) in systems ranging from data center and automotive to healthcare and FinTech, by migrating code from the proprietary constraints of CUDA to SYCL, an open-standard alternative, embraced by multiple hardware vendors. At present, a majority of HPC and AI solutions are bound to Nvidia via CUDA, working only with Nvidia chipsets.
A key mechanism to help accomplish the SYCL migration is the Intel® DPC++ Compatibility Tool (Intel® DPCT), which can automatically translate up to 90 percent of the CUDA code; the remainder is then analyzed and modified manually.
Avoiding ‘vendor lock’ for a Flexible FutureThe Codeplay demo shows the simplicity of migrating CUDA code to SYCL while continuing to run on Nvidia GPUs. By adopting SYCL and oneAPI, users will enjoy a royalty-free, open standard alternative to CUDA that’s defined and supported by industry experts on a growing range of processors, including Intel, Nvidia, and AMD.
Industry MomentumMoving to open standards software offers organizations improved portability and performance, and perhaps, more importantly, choice through access to a broad ecosystem. Made up of industry leaders, this extensive ecosystem offers familiar development tools, verified libraries and frameworks, along with benchmarks.