chiprook
← Hardware
HardwareSeptember 24, 2026, 14:38

Moore Threads MTT S5000 matches mainstream GPUs in embodied RL (r=0.976)

Moore Threads published embodied-AI training data for its MTT S5000 GPU, reporting training curves that track international mainstream GPUs with a correlation coefficient of 0.976. Support entered the open RLinf framework from version 0.3, single-step time dropped from 2,549 to 1,888 seconds, and dual-arm real-robot assembly reached about 92% success.

Moore Threads MTT S5000 matches mainstream GPUs in embodied RL (r=0.976)
#MooreThreads#MTTS5000#RLinf
Read next
Hardware

CXMT gains ground in DRAM as Samsung, SK Hynix shift capacity to HBM

Hardware

XFEON unveils Physical-AI matrix: Xinghe S1 edge chip, AGLobe brain, Xinghui Token workstations

Hardware

Imec and TSMC expand cloud access to advanced-node chip design

Hardware

Huawei unveils Atlas 960E SuperPoD: 4,096 NPUs, 8 EFLOPS FP8