The proliferation of AI workloads across diverse hardware has created a fragmented ecosystem with vendor-specific libraries and duplicated optimization efforts.
The UXL Foundation leverages Intel's oneAPI to create a unified, open-source library suite that ensures portability and performance across different hardware architectures.
A breakdown of oneDNN for accelerating deep learning primitives and oneMKL for high-performance mathematical functions like BLAS and FFT.
An overview of oneTBB for task scheduling, oneCCL for multi-node communication, and oneDAL for managing data analysis and machine learning pipelines.
The complete architecture shows how UXL libraries abstract hardware and accelerate operations for frameworks during training, inference, and serving.
An open-source, unified stack increases developer productivity, fosters community innovation, and ensures software evolves in lockstep with new hardware releases.
A call for developers to contribute kernels, join the UXL Foundation working groups, and share workloads to help grow the open ecosystem.
19:39Andrew Waafa
35:16Ankit Patel
26:25Stephan Gillich
31:05Ankit Patel, Matt White, Philipp Schmid, Lucie-Aimée Kaffee & Andreas Blattmann
23:09Gian Marco Iodice
28:09Jan Oberhauser, Mirko Novakovic, Alex Laubscher & Keno Dreßel
23:04Matt White
26:47Alejandro Saucedo
Jobs that call for the skills explored in this talk.
