1 source 4m ago
AMD and Intel Move PyTorch x86 Optimization Into oneDNN
AMD and Intel said they will fold PyTorch optimization work for x86 CPUs into the open-source library oneDNN, making it the default acceleration path for both companies' processors.
Reporting from 1 source: ASCII.jp.
Why it matters
A single default library removes the split optimization path that forced PyTorch developers to choose between AMD and Intel CPUs, and it puts both companies behind one instruction set plan for AI workloads.
Key Facts
- AMD and Intel will integrate PyTorch x86 CPU optimization work into the open-source oneDNN library.
- AMD will integrate its ZenDNN optimization library into oneDNN as the default x86 acceleration path.
- AMD has joined the Unified Acceleration Foundation, which manages oneDNN, as a steering member.
- AMD and Intel are cooperating on AI Compute Extensions, an x86 instruction set for matrix multiplication.
- The effort follows recommendations from the x86 Ecosystem Advisory Group.
AMD will integrate its ZenDNN library into oneDNN and has joined the Unified Acceleration Foundation as a steering member. The two companies are also working together on AI Compute Extensions, a next-generation x86 instruction set aimed at matrix multiplication.
The change reaches past existing code. AMD and Intel are also working on AI Compute Extensions, an instruction set intended to speed up matrix multiplication, the core operation in deep learning workloads. PyTorch is expected to reach those features through future oneDNN updates.
AMD's entry into the Unified Acceleration Foundation as a steering member gives it a formal role in the body that manages oneDNN, so the arrangement has a venue for continued joint development rather than a one-time handoff. Jeff McVeigh, VP and GM of Intel's Data Center Strategy Programs, said developers should not have to be aware of separate software paths to get the best performance out of x86 CPUs. Robert Hormuth, Corporate VP of Architecture & Strategy at AMD, stressed the value of the industry uniting around common foundational technology.
Written with AI by Yomimono from the 1 cited source below, including Japanese-language reporting where cited. See our editorial policy.
Add Yomimono as a preferred source on Google