Advanced Computing in the Age of AI | Saturday, July 20, 2024

Exploring the AMD ROCm Ecosystem: Webinar to Feature Key Strategies for Portability and Efficiency 
Sponsored Content by AMD

A tech dots and lines background, in blue and pink colors.As high performance computing (HPC) continues to tackle the most intricate real-world problems in science and industry, AMD platforms are driving the powerful compute, efficiency, and adaptability requirements for large scale HPC deployment.

ROCm™ is the AMD open source software stack designed for GPU computing that consists of a collection of drivers, development tools, and APIs that enable GPU programming from low-level kernel to end-user applications.

The ROCm platform is compatible with many of the most popular HPC programming languages and frameworks, including those that parallelize operations across multiple GPUs and servers, handle memory hierarchies, and solve linear systems.

AMD invites HPC enthusiasts to join a webinar dedicated to the latest version of this software, ROCm 6.0, that will take place on Thursday, July 17 at 11:00 a.m. EDT. The latest ROCm updates will be presented by Joe Schoonover, Founder & Research Software Engineer at Fluid Numerics.

Explore New ROCm Updates Supporting AMD Instinct™ MI300 GPUs

The two-part webinar’s first segment will delve into recent updates introduced in ROCm 6.0.0 and 6.0.2 related to MI300 support and changes to supported operating systems and GPUs. The webinar will also analyze performance implications for those transitioning from ROCm 5.7 to 6.0 with a special focus on the AMD CDNA™ 2 architecture for real-world applications such as the Spectral Element Library in Fortran, an API for solving partial differential equations.

AMD says the ROCm 6.0.x enhancements are primarily aimed at bolstering support for the company’s AMD CDNA 3 architecture, the dedicated compute architecture underlying its AMD Instinct MI300 Series accelerators. The webinar will compare CDNA 2 and CDNA 3 architectures, along with APU (MI300A) and discrete GPU (MI300X) architectures, with an emphasis on memory spaces and the potential nuances related to memory management, particularly with MI300A.

Driving Portability with HIP

ROCm helps to address the common HPC constraints of vendor lock-in and ease of migration with its emphasis on portability. With ROCm, users can port CUDA®-based code into the AMD native, open source Heterogeneous Computing Interface for Portability (HIP), allowing developers to run their code in both NVIDIA and AMD environments with minimal changes.

The second part of the webinar will take a deep dive into HIP and its significance within the ROCm ecosystem. The presentation will outline the historical evolution of ROCm concerning HIP and the various AMD compiler aliases which will guide users in facilitating portability of HPC applications across GPUs from diverse vendors.

Dr. Schoonover and other HPC experts from AMD will also be available at the end of the webinar for a live Q&A with attendees.

This fascinating webinar offers a unique opportunity to stay ahead of the curve as AMD redefines the HPC landscape with its versatile approach to flexible and efficient computing, which is paving the way for the groundbreaking research and discovery the HPC community is known for. Sign up for the July 17 webinar today at this link.

Learn with the AMD HPC User Forum

This exciting ROCm webinar is presented in partnership with the AMD HPC User Forum, a member-led community of customers and users of AMD processor-powered HPC systems. Members can share technical content, experiences, and best practices on a forum that provides them the opportunity for direct connection with AMD leadership and to influence future AMD HPC products and solutions. Learn more about the AMD HPC User Forum here.

EnterpriseAI