[Paper Review] Open SYCL on heterogeneous GPU systems: A case of study

Rocío Carratalá-Sáez, Francisco J. Andújar+3 more|arXiv (Cornell University)|Oct 10, 2023

Parallel Computing and Optimization Techniques4 citations

TL;DR

This paper evaluates Open SYCL for heterogeneous GPU systems using the Finite Time Lyapunov Exponent computation as a case study, demonstrating that SYCL offers performance comparable to CUDA and HIP while significantly reducing host code development effort and enabling seamless multi-vendor GPU utilization across NVIDIA and AMD devices.

ABSTRACT

Computational platforms for high-performance scientific applications are becoming more heterogenous, including hardware accelerators such as multiple GPUs. Applications in a wide variety of scientific fields require an efficient and careful management of the computational resources of this type of hardware to obtain the best possible performance. However, there are currently different GPU vendors, architectures and families that can be found in heterogeneous clusters or machines. Programming with the vendor provided languages or frameworks, and optimizing for specific devices, may become cumbersome and compromise portability to other systems. To overcome this problem, several proposals for high-level heterogeneous programming have appeared, trying to reduce the development effort and increase functional and performance portability, specifically when using GPU hardware accelerators. This paper evaluates the SYCL programming model, using the Open SYCL compiler, from two different perspectives: The performance it offers when dealing with single or multiple GPU devices from the same or different vendors, and the development effort required to implement the code. We use as case of study the Finite Time Lyapunov Exponent calculation over two real-world scenarios and compare the performance and the development effort of its Open SYCL-based version against the equivalent versions that use CUDA or HIP. Based on the experimental results, we observe that the use of SYCL does not lead to a remarkable overhead in terms of the GPU kernels execution time. In general terms, the Open SYCL development effort for the host code is lower than that observed with CUDA or HIP. Moreover, the SYCL version can take advantage of both CUDA and AMD GPU devices simultaneously much easier than directly using the vendor-specific programming solutions.

Motivation & Objective

• Address the growing challenge of programming heterogeneous GPU clusters with diverse vendors and architectures.
• Reduce the complexity and portability issues inherent in vendor-specific programming models like CUDA and HIP.
• Evaluate Open SYCL as a high-level, portable alternative that supports both single and multi-vendor GPU deployments.
• Assess the trade-off between performance and development effort in real-world scientific HPC workloads.
• Demonstrate the feasibility of using SYCL for production-level scientific applications on mixed GPU environments.

Proposed method

• Implement the Finite Time Lyapunov Exponent (FTLE) computation using Open SYCL, CUDA, and HIP for direct comparison.
• Use the Open SYCL compiler to generate kernels targeting both NVIDIA and AMD GPU devices from a single codebase.
• Profile kernel execution times and host code complexity across all three implementations.
• Measure performance on heterogeneous systems with mixed GPU vendors (NVIDIA and AMD) to evaluate portability and load balancing.
• Analyze host code structure and lines of code to quantify development effort differences.
• Ensure feature parity across implementations to isolate performance and productivity differences.

Experimental results

Research questions

• RQ1 Does Open SYCL introduce significant performance overhead compared to vendor-optimized CUDA and HIP implementations?
• RQ2 How does the development effort for host code in Open SYCL compare to that of CUDA and HIP?
• RQ3 To what extent can Open SYCL efficiently utilize multiple GPU devices from different vendors in a single application?
• RQ4 Can SYCL achieve performance portability across diverse GPU architectures without sacrificing efficiency?
• RQ5 What is the practical feasibility of using Open SYCL for high-performance scientific computing in heterogeneous environments?

Key findings

• Open SYCL achieves GPU kernel execution times comparable to those of CUDA and HIP, with no significant performance overhead observed.
• The development effort for host code in Open SYCL is substantially lower than in CUDA or HIP, due to higher-level abstractions and reduced boilerplate.
• SYCL natively supports heterogeneous GPU systems, enabling efficient use of both NVIDIA and AMD GPUs in the same application with minimal code changes.
• The SYCL implementation demonstrated better portability across different GPU vendors compared to vendor-specific solutions.
• The performance of the SYCL-based implementation remains competitive across both single-GPU and multi-GPU configurations.
• The study confirms that SYCL can be a viable alternative for scientific HPC applications requiring multi-vendor GPU support and maintainable codebases.