Title

AWS re:Invent 2022 - Run large-scale graphics workloads with AWS, featuring Mircom and Snap (CMP208)

Summary

• Samantha Pham, a principal product manager at AWS, introduces the session on running large-scale graphics workloads on AWS.
• AWS offers over 500 EC2 instances, including those optimized for graphics workloads with processors from Intel, AMD, and AWS Graviton.
• AWS provides hardware accelerators like NVIDIA and AMD GPUs, media accelerator cards, and chips for ML inference and training.
• G4AD, G5, G4 GN, and G5G instances are highlighted for graphics workloads, with G5G being the first GPU-based instance paired with ARM processors.
• VT1 instances are designed for live, low-latency transcoding workloads.
• AWS also offers services like Nice DCV for remote display and managed services for various industries.
• Graphics applications are being used across industries for immersive experiences, productivity, education, collaboration, and retail.
• Four main graphics use cases are remote workstations, game/pixel streaming, rendering, and video transcoding.
• Customer references include Ubitis for cloud gaming, Vario for virtual reality, and DioLive for real-time video streaming.
• Eli Semery from Snap discusses the evolution of Bitmoji avatars from 2D to 3D, the architecture of creating and rendering avatars on AWS, and the use of G4DN and G5 instances for rendering over 1.7 billion 3D images daily.
• Brian Lung from Mircom presents their journey from on-premise hardware to AWS cloud for their OpenGN platform, which monitors fire alarm systems and integrates with other building technologies. They have refactored and re-platformed to EC2 G5G instances, resulting in a modernized OpenGL experience and a digital twin IoT platform.

Insights

• AWS's broad portfolio of EC2 instances and managed services provides a robust infrastructure for running graphics-intensive applications, which is critical for industries like gaming, media, and entertainment.
• The transition from on-premise hardware to cloud-based solutions, as demonstrated by Mircom, can lead to significant improvements in scalability and cost-effectiveness, although it may require initial investments in refactoring and re-platforming.
• Snap's use of AWS for Bitmoji avatars showcases the potential of cloud computing in handling massive scale real-time rendering workloads, which is essential for social media platforms with millions of daily active users.
• The evolution of Bitmoji from 2D to 3D avatars on AWS indicates a trend towards more immersive and customizable user experiences in social media and gaming.
• The success stories of Ubitis, Vario, and DioLive illustrate the versatility of AWS's graphics instances in supporting diverse use cases, from cloud gaming to virtual reality to real-time video streaming.
• The combination of AWS's EC2 instances with specialized hardware accelerators and managed services enables customers to optimize their workloads for performance and cost, which is a key consideration for companies operating at scale.
• The advancements in AWS's GPU instances, such as the G5G paired with Graviton2 processors, reflect AWS's commitment to innovation and performance optimization in the cloud computing space.