Defining Open Standards Programming for AI and HPC - Hosted by the Unified Acceleration (UXL) Foundation
Wednesday, June 11, 2025 2:15 PM to 3:15 PM · 1 hr. (Europe/Berlin)
Hall G1 - 2nd floor
Birds of a Feather
AI Applications powered by HPC TechnologiesCompiler and Tools for Parallel ProgrammingDevelopment of HPC SkillsParallel Programming Languages
Information
The Unified Acceleration (UXL) Foundation was formed to define and implement open standards development for multi-vendor architectures including CPUs, GPUs and other processors used to accelerate software.
The world has moved towards what are known as heterogeneous architectures. These are systems that take advantage of different types of processors to achieve not only the fastest performance, but include chips for specific tasks. This is especially being seen in AI, with CPUs, GPUs, NPUs, TPUs and other AI processors being used to achieve the best performance and energy efficiency.
However, software developers do not want to program in different ways for different processors, in particular across vendors. The challenge developers are currently seeing is that different architectures often typically require unique languages, tools and libraries, adding complexity and limiting code reuse. These are also often closed source and proprietary. This makes it difficult to take advantage of multi-architecture systems and adopt new architectures, where code branching and porting increase effort and maintenance costs.
The UXL Foundation was formed to tackle these challenges, by providing an open, standards-based way to develop software and deploy software across vendor processors. The foundation governs an open specification and open source projects providing accelerated routines that aim to cover the building blocks for the majority of applications. There are seven projects governed by the foundation covering a range of functions for math, AI operators, parallel ISO C++, multi-threading, and others, implemented using C++.
The BoF session will explore the current challenges when developing software for different architectures and ask - is it too late to define open standards for HPC and AI software development? Are the developers already committed to their programming ecosystem? Is there any hope for platform portability? Can polymorphic compilers help? What can the community do to deliver new solutions, especially within the context of what the UXL Foundation is doing?
Organizers:
The world has moved towards what are known as heterogeneous architectures. These are systems that take advantage of different types of processors to achieve not only the fastest performance, but include chips for specific tasks. This is especially being seen in AI, with CPUs, GPUs, NPUs, TPUs and other AI processors being used to achieve the best performance and energy efficiency.
However, software developers do not want to program in different ways for different processors, in particular across vendors. The challenge developers are currently seeing is that different architectures often typically require unique languages, tools and libraries, adding complexity and limiting code reuse. These are also often closed source and proprietary. This makes it difficult to take advantage of multi-architecture systems and adopt new architectures, where code branching and porting increase effort and maintenance costs.
The UXL Foundation was formed to tackle these challenges, by providing an open, standards-based way to develop software and deploy software across vendor processors. The foundation governs an open specification and open source projects providing accelerated routines that aim to cover the building blocks for the majority of applications. There are seven projects governed by the foundation covering a range of functions for math, AI operators, parallel ISO C++, multi-threading, and others, implemented using C++.
The BoF session will explore the current challenges when developing software for different architectures and ask - is it too late to define open standards for HPC and AI software development? Are the developers already committed to their programming ecosystem? Is there any hope for platform portability? Can polymorphic compilers help? What can the community do to deliver new solutions, especially within the context of what the UXL Foundation is doing?
Organizers:
Format
On Site
Targeted Audience
This BoF will attract software developers that design functionality for AI and HPC applications running on multi-vendor hardware architectures. Hardware companies will be drawn to the discussions on how to enable portability and which programming environments attract software developers. It will include UXL Foundation members and the broader community.
