header-langage
简体中文
繁體中文
English
Tiếng Việt
한국어
日本語
ภาษาไทย
Türkçe
Scan to Download the APP
Partnered With:

Nvidia, AMD, Intel Rarely Join Forces: 9 Companies Build SGLang Open-Source Inference Alliance

Beating AI News Flash: At the inaugural SGLang Summit, NVIDIA, AMD, and Intel, normally fierce competitors, uncharacteristically stood on the same side.


The three chip giants, along with Google, Meta, Thinking Machines Lab, Cloudflare, Modal, and RadixArk, became the 9 founding members of the SGLang Alliance, participating in shaping the future roadmap of this open-source AI inference engine. The summit attracted over 2,000 in-person attendees.


SGLang is the underlying software of the large model runtime, responsible for handling user requests, scheduling GPUs, managing memory, and reusing computation results. It directly determines how fast and at what cost a chip can run AI models.


SGLang previously supported multiple hardware types including NVIDIA, AMD, and Intel. The formation of the alliance means chip manufacturers will participate more systematically in joint development, contributing engineers and testing resources, and driving performance optimization for their respective hardware.


Alliance members hold roadmap meetings twice a year, but final technical decisions and code merge authority remain with the SGLang maintainer team.


SGLang officially states that 332 organizations have adopted the project. Meta also disclosed at the summit that after migrating to SGLang, its large-scale production environment saved over half of its GPU resources compared to its previous inference solution.


The summit also announced the development direction of SGLang v1.0, focusing on improving stability and compatibility in production environments, with the official version still in development. Additionally, SGLang has launched public preview support for AWS Trainium chips and partnered with Intel to optimize its CPUs, GPUs, and accelerators. The reinforcement learning framework Miles v0.2 has also been released.

举报 Correction/Report
Correction/Report
Submit
Add Library
Visible to myself only
Public
Save
Choose Library
Add Library
Cancel
Finish