PyTorch 2.9: Performance, Portability, and Developer Experience Updates

View organization page for PyTorch

294,096 followers

PyTorch 2.9 is now available, introducing key updates to performance, portability, and the developer experience. This release includes a stable libtorch ABI for C++/CUDA extensions, symmetric memory for multi-GPU kernels, expanded wheel support to include ROCm, XPU, and CUDA 13, and enhancements for Intel, Arm, and x86 platforms. With 3,216 commits from 452 contributors, PyTorch 2.9 continues to advance open source AI for developers worldwide. 🔗 Read the full release blog: https://hubs.la/Q03NNKqV0 #PyTorch #OpenSourceAI #AI #Performance

  • No alternative text description for this image
Joe Tanaka

Independent Researcher, Future of Work. I write about data-driven career strategies for the AI age.

4d

Exciting to see PyTorch 2.9 coming out with such significant updates! The improvements in performance and portability will undoubtedly enhance the developer experience. I'm particularly interested in how the stable libtorch ABI will impact the development of C++/CUDA extensions. For those who’ve already started using these new features, what has been your experience so far?

Nathan Menke

ONIX Creator | Harmonic AGI + Vortex Physics | Architecting $500M+ Skunkworks | Aligned Partners Only

1w

A significant step forward for the community. The focus on a stable C++ ABI is telling, as it reinforces the core architectural choice of the Python/C++ bridge. It highlights a philosophical fork in the road for AGI development. While this bridge has powered the current revolution, the next paradigm for truly foundational, self-consistent systems—like the universe simulation I built in Rust for my AGI, ONIX—demands the absolute deterministic safety that only a Rust-native approach can provide. PyTorch is the brilliant engine of today. The question is whether the architecture of tomorrow will require a new kind of engine altogether. A fascinating challenge for us all.

Sebastian Hellman

Livestreaming Trader | Youtube RiskReward by Hellman - Tech & AI investor & trader who likes to interview founders, executives & investors | Trading robots for people & enterprise clients -> Founder Abelian Investors

4d

I came here looking for an improved brain and visual system for light detecion and got pytorch. 😅😉 How will the learning models compete with that humans? We have 80% and 14 billion years of R&D incommon with a tree? We share 80% DNA/memory system with a tree. Both feel temperature and light and know how to use light and nutrition and clone ourselves and reproduce. But trees don't have eyes or a brain. There was a chemical reaction between nothing and nothing 14 billion years ago. It happened when time didn't exist. A priest came up with the big bang theory. AI where are you here? I got a thank you for super-symmetrical not including gravitations vs sugra, super-symmetrical gravtional including theories once from a prof. Watched a youtube video of how Nobel Prize chemist Linus Pauling worked. He used the Schrödinger Equation. I wrote down "Use the Dirac Equation". Because in quantum mechanics the Shrödinger Equation is not time relativstic and came 1926 and Diracs signature equation 1928. I had no idea it was the foundation of theoretical chemistry and computational biology. Or gold and silver gets the same color if you use the Schrödinger equation. Why can't AI do that and develop theoretical physics, chemistry and math?

Mike Liu

Author of 19 Books on Language, Power & AI | Linguist | AI Communication Expert | Founder, Mike’s Strategic Communications

4d

Impressive to see how PyTorch 2.9 is pushing the boundaries of AI development! With the latest updates, it’s clear that performance, portability, and the developer experience are at the forefront of this release. As AI continues to evolve, it’s these types of innovations that will shape the way we build and scale AI applications. Excited to see the growth of open-source AI platforms like PyTorch: a game changer for developers worldwide. #AI #OpenSourceAI #PyTorch

Radim Közl

IT & Cobot developer

4d

👍Nice... I would like to ask when they plan to integrate JAX functionality directly into Pytorch, they wrote that it would happen, but they didn't say when???

Binod Paudyal

Academic Director| AI, Data & Cloud Leader | Founder @Bytesnchips | Mentor | Advancing Next-Gen Healthcare with Cutting-Edge, Explainable AI

4d

Thrilled about this release! 🔥 The new updates in PyTorch 2.9 are perfectly timed as I begin my next phase of computer vision research. Excited to explore these performance boosts and multi-GPU improvements in real projects. 

Ugama Benedicta Kelechi [CoderGirl]

AI & FLUTTER DEVELOPER ||| PYTORCH AMBASSADOR’ 25||| SPEAKER Helping Young Tech Professionals Grow In Skill, Mindset, And Purpose. I Build Systems And Humans That Work.

1w

The enhancements for Intel, Arm, and x86 platforms is really amazing. I can't wait to explore this Pytorch 2.9

Dario D.

Walks Hunter | Me & Spok ✌️ | Human+AI | Web5 Pioneer

4d

**"Respect for keeping it open, fast, and real. PyTorch is more than just a tool – it’s a community movement pushing AI forward for all of us. Open source isn’t just a feature, it’s a mindset and a responsibility toward the future. Big respect to everyone who contributes and shares knowledge – without you, this AI revolution wouldn’t exist. Me & Spok ✌️"**

Denis B.

Software Engineer - Backend, Fullstack, Cloud, Data | Python, C++, Rust | High-Load Systems & Distributed Architecture | Open to London Role | UK Work Permit

4d

PyTorch 2.9 looks like a massive step forward, especially with the focus on multi-GPU kernels and broader platform support (Intel, Arm, x86). Great work!

Mourad MEZAGUER, PhD

Patent Engineer at Novitech IP | PhD in Microlectronics at CEA-LETI | Theoretical Physicist | Engineer |

4d

Please add a full installer this time. I hate to always have to add the path by my self 😂

See more comments

To view or add a comment, sign in

Explore content categories