HPC and AI on Arm architecture: Observations, learnings and path forward
Workshop Date: December 18, 2023
Overview
This workshop is aimed at providing a “birds-eye” view of the latest contributions and progress in industry and academia for furthering Arm architecture adoption for HPC and AI. Specifically, it is important to highlight several systems that are available today based on Fujitsu’s A64FX, AWS Graviton3, Ampere Altra processors. In addition, the Arm HPC/ AI ecosystem has mature software tools and applications. New designs based on Arm architecture such as “Rhea” CPU by SiPearl, ETRI and “Grace” CPU by Nvidia reinforce the value proposition of Arm for HPC/ AI.
In this workshop, we will have industry experts/ users highlight some of the experiences using an Arm based system for HPC/ AI problem solving. We will have performance results comparing different architectures and systems, portability concerns, performance analysis tools and compiler code-gen/ maturity. We will cover case-studies optimizing applications for Arm, NEON/ SVE vs. AVX vectorization, amongst several other topics. In addition, we plan to have a panel discussion on the workshop theme.
Program Schedule
|
Time
|
Topic
|
Duration
|
|
2:00 pm |
Openings remarks
|
5 mins |
|
2:05 pm |
Talk 1: Architecture innovation and the future of Arm processors Speaker: Ashok Bhat, ARM
|
35 mins |
|
2:40 pm |
Talk 2: Deep-dive into AWS Graviton processors Speaker: Santosh Kumar, AWS
|
35 mins |
|
3:15 pm |
Talk 3: Demo – Running HPC Applications on AWS Speaker: Bhagyaraju Kasina, AWS
|
15 mins |
|
3:30 pm |
Break
|
15 mins |
|
3:45 pm |
Talk 4: Fujitsu and its game changing HPC AI optimized processors Speaker: Priyanka Sharma, Fujitsu
|
35 mins |
|
4:20 pm |
Talk 5: The future of Compute for AI Speaker: Vinod Kumar, Ampere Computing
|
35 mins |
|
4:55 pm |
Closure |
5 mins |
Topics
- Experiences on GPU-Accelerated Arm-based HPC Platforms.
- Adoption of ARM in the HPC community: Benchmarks and Applications.
- HPC on Arm based offerings in the cloud: AWS Graviton, Ampere Altra, others.
- Performance comparison of Open-source HPC apps on Arm vs. x86 architecture.
- Machine Learning and AI on Arm Servers: Libraries, BF16, optimized ML models.
- Porting to Arm, usage of intrinsic functions: real world experiences.
- Tools availability for porting and performance tuning on Arm.
- SVE and compiler auto-vectorization opportunities.
- HPC and AI application case-studies.
Organizers, Speakers
Workshop Chair, Co-Chair:
Rama Malladi, Amazon Web Services, India
Bhagyaraju Kasina/ Santosh Kumar, Amazon Web Services, India
Tentative Speakers
Ashok Bhat, Arm, India
Sanjay Tiwary/ Bhagyaraju Kasina/ Santosh Kumar, Amazon Web Services, India
Vinod Kumar, Ampere, India
Priyanka Sharma, Fujitsu, India