NVIDIA GTC 2022 is coming soon (March 21-24, virtual event) and Arm and our partners will be presenting on a wide range of interesting topics. Whether you are looking for application performance tuning tips, Arm porting advice, or how to make best use of the latest Arm-based cloud instances, GTC 2022 has something for you. In this blog I highlight several sessions being presented by Arm and by our Arm Neoverse ecosystem partners. As a reminder, registration and attendance of NVIDIA GTC 2022 is free.
Arm’s Director of HPC Engineering, John Linford, is well regarded as an expert in tuning HPC applications for maximum performance on Arm platforms. In his talk, “Port, Profile, and Tune HPC Applications for Arm-based Supercomputers”, John will present a methodology and software tools for porting and tuning HPC applications on any Arm-based system. These include NVIDIA’s upcoming “Grace” CPU, as well as AWS Graviton2, Ampere® Altra®, Fujitsu A64FX, SiPearl “Rhea” and others. In addition, John will discuss performance tuning for Arm-hosted NVIDIA A100 and NVIDIA V100 GPU-based systems.
Next up, Beau Paisley, Principal Solutions Architect for HPC Tools at Arm, will present “Ensuring Program Correctness in CUDA-based HPC Applications”. Beau will present a methodology and set of tools for finding bugs and ensuring program correctness in CUDA-based HPC applications. This includes a hands-on overview of Arm DDT, Arm’s cross-platform, integrated environment for debugging parallel codes at any scale. He will also demonstrate how to detect and fix correctness issues on CUDA devices and how to use tools to analyze CUDA applications.
A key Arm advantage is our ecosystem of partners, and on that front there are many interesting partner talks available.
From NVIDIA, “Getting started with Arm software development: 86 the x86 dependencies in your code”. Jeff Hammond, Principal Engineer at NVIDIA will show how to break away from architecture-specific software patterns so your application can run on any CPU architecture, especially the Arm architecture. Using examples from scientific applications in bioinformatics and chemistry, Jeff will show the common mistakes that break portability and how to fix them. The simplest issues include compiler flags and dependence on specific compilers or vendor libraries, but Jeff will also go into SIMD intrinsics and atomic operations.
Eric Lequiniou, Altair’s VP of Radioss Development & Altair Solver HPC, will address the question “Leveraging Arm Processors in Data Centers: The Chicken-or-egg Problem Finally Solved for Commercial Applications?” Eric will answer this question using the example of the Altair Radioss solver, an industry-proven crash code. As demand for accurate simulations continues to rise, Arm chip architecture is emerging as an attractive alternative, providing hardware cost optimizations and energy efficiency while handling compute-intensive applications. Eric will summarize Altair’s porting experience and discuss the continuous performance improvement observed. He will also take a deep dive on the Arm Neoverse platform with latest results achieved with Radioss on the Ampere Altra processor.
“First hands-on experiences using the NVIDIA Arm HPC Developer Kit” is presented by NVIDIA (Filippo Spiga, Developer Relations Manager) and by several leading HPC national labs and universities, including:
The NVIDIA HPC developer kit allows the HPC community to prepare for future NVIDIA Arm CPUs and other Arm CPUs-plus-GPU deployments. Come hear from these early adopters as the present their results – positive and negative – share wisdom, and help you understand what to do next.
Oracle Distinguished HPC Cloud Architect, Kevin Jorissen, will present “Arm’s performance and impact on HPC in the cloud”. Oracle entered the market for Arm-based servers in 2021 with a dual-socket Ampere Altra-based instance shape. Priced at just $0.01 per core-hour, it turned out to be one of Oracle’s most popular instance launches. Kevin will discuss why users typically choose Arm, how they start, and what to expect. And he’ll also present optimized numerical performance libraries (BLIS) that were developed to boost performance specifically for the Ampere Altra platform.
From Alibaba Cloud, Ming Yang, Engineer II, will present “The Future Gaming Trend: Cloud Gaming with Aliyun Elastic Cloud Phone Solution”. Based on NVIDIA’s T4 card and native Arm platform, Alibaba cloud has designed and implemented a high-density and high-performance cloud mobile gaming solution – Elastic Cloud Phone.
Stefano Marzani, Worldwide Tech Leader for Autonomous Driving, Amazon Web Services will present “Accelerate Innovation in Autonomous Vehicle Systems with Amazon EC2 G5g Instances”. By using G5g instances – powered by Arm-based Graviton2 processors and NVIDIA T4G Tensor Core GPUs – embedded developers can now run their workloads with unprecedented parity between the cloud and the edge while also taking advantage of the elastic scaling in the AWS Cloud.
And last but not least (for this blog anywhere … there are yet more Arm-related sessions in the NVIDIA GTC 2022 catalog), Ramesh Narayanaswamy, Principle Engineer Silicon Realization Group, Synopsys and Magnus Ekman, Director of Architecture, NVIDIA will co-present “Large-scale IC Simulation Workload on Arm Server CPU”. They will describe their experience porting Synopsys’ VCS Logic Simulator to Arm architecture, and will present the workload behavior and efficiency of using Arm-based servers for logic simulation.
By the way, I’ve only covered a handful of Arm Neoverse-related talks here. There is still more to learn based on Arm’s client, automotive, and IoT products.
Join Arm at GTC 2022