Mmastodon TechnologyAI first seen 10 h ago, last 10 h ago, peak #5

GPUWorks open-sources lightweight multi-vendor GPU inference platform

Original: GPUWorks has open-sourced a 40MB inference platform that pools Nvidia, Huawei Ascend, T-Head PPU and Hygon DCU accelerat

GPUWorks has released an open-source inference platform weighing just 40MB that pools Nvidia, Huawei Ascend, T-Head PPU and Hygon DCU accelerators into a single managed cluster behind an OpenAI-compatible API. The team claims one server orchestrated 1,003 GPU servers and 500 replicas in four minutes. Interest centres on hardware-agnostic AI serving at a time of export controls and diverse chip supply.

Why now: It promises unified inference across Chinese and Western AI chips, a timely solution amid US export restrictions on Nvidia hardware.

GPUWorksNvidiaHuawei AscendT-HeadHygon

Open on mastodon →

Evidence

• GPUWorks has open-sourced a 40MB inference platform that pools Nvidia, Huawei Ascend, T-Head PPU and Hygon DCU accelerators into one managed cluster with an OpenAI-compatible API. The team says one server managed 1,003 GPU servers and 500 replicas in four minutes, and waking a… · china@universeodon.com · 1

API: https://socialmediatrends-api.osmike.com/v1/trends/4865691