IT House, Feb 12 — On February 11, Zhipu formally launched its new-generation giant mannequin GLM-5.Moore Threads has accomplished full-process adaptation and validation on launch day (Day‑0) for its flagship AI coaching and inference GPU, the MTT S5000, primarily based on the SGLang inference framework.
Leveraging the intensive operator protection and robust ecosystem compatibility of its MUSA structure, Moore Threads totally enabled the end-to-end inference pipeline for the mannequin. It unlocked the native FP8 acceleration of the MTT S5000, sustaining mannequin accuracy whereas vastly lowering video reminiscence utilization, delivering high-performance inference for GLM‑5.
As the newest milestone within the GLM collection, GLM‑5 is positioned as a top-tier coding mannequin, with general efficiency improved by 20% over the earlier technology.Its key breakthrough lies in Agentic Engineering capabilities: it not solely excels in code technology but additionally handles advanced system engineering and long-range agent duties, enabling end-to-end growth from necessities to software.
The MTT S5000 is a full-featured AI computing GPU designed for large-model coaching, inference, and high-performance computing, constructed on the 4th‑technology MUSA “Pinghu” structure.It delivers as much as 1000 TFLOPS of AI computing energy per card, with 80 GB of reminiscence, 1.6 TB/s reminiscence bandwidth, and 784 GB/s inter-card bandwidth, totally supporting full-precision computing from FP8 to FP64.
Supply: IT house
Elevate your perspective with NextTech Information, the place innovation meets perception.
Uncover the newest breakthroughs, get unique updates, and join with a worldwide community of future-focused thinkers.
Unlock tomorrow’s developments as we speak: learn extra, subscribe to our publication, and change into a part of the NextTech group at NextTech-news.com
