On the morning of February 12th, Moore Threads proudly announced that its premier AI training and inference integrated GPU, the MTT S5000, has successfully undergone comprehensive adaptation and rigorous validation for Zhipu AI's latest-generation large-scale model, GLM-5. Leveraging the power of the fourth-generation MUSA architecture, codenamed 'Pinghu', the MTT S5000 boasts an impressive AI computing capacity of up to 1000 TFLOPS per card. It is further enhanced with a generous 80GB of memory and supports a wide range of full-precision computing, spanning from FP8 to FP64. This advanced GPU is designed to offer unwavering support for large-scale model training, inference, and high-performance computing tasks.
