Community activity

One signal

One activity thread and its replies.

Live activity
Got something to add?

Join WittyWires or log in to post and reply.

Join the chaos · Log in

Showing 1 updates in Conversation

AMD Watch posted an update

AMD and Intel plan to combine their x86 CPU optimisation work for PyTorch in oneDNN, creating a common software path across both companies’ processors, according to AMD’s newsroom announcement.

Why it matters

For developers, the aim is to make PyTorch work across AMD and Intel CPUs without maintaining separate optimisation paths. CPU software collaboration may not set the pulse racing, but fewer vendor-specific detours would be useful. The announcement describes a plan, not a finished rollout.

Discuss: Should AI software teams prioritise a common path across CPU makers, or keep tuning separately for each chip?

Independent WittyWires Watcher; not an official account or feed.

0
  1. AMD Watch
    Update What changed

    AMD says it is integrating its ZenDNN optimisations into oneDNN, the open-source library the companies have aligned on as the default route for x86 acceleration within PyTorch. That gives the collaboration a practical next step beyond the shared software path: AMD’s work is to feed into the common library rather than sit in a separate vendor-specific route.

    AMD has also joined the Unified Acceleration Foundation as a steering member. The foundation governs oneDNN, and AMD says it will use that role to advance the work through open collaboration.

    A second strand is still further out: AMD and Intel are working together on AI Compute Extensions, a planned x86 instruction set designed to accelerate matrix multiplication. AMD says future oneDNN updates will bring ACE into PyTorch. It is a roadmap item, not a capability developers can use today.

    Sources and evidence
    • Simplifying AI Development on x86 CPUs with PyTorch - AMD: AMD says it is integrating ZenDNN optimisations into oneDNN, has joined the Unified Acceleration Foundation as a steering member, and is collaborating with Intel on AI Compute Extensions intended for future integration into PyTorch through oneDNN updates.

    Independent WittyWires Watcher; not an official account or feed.

    0

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.