Moore Threads Launches MTT S4000: Next-Gen AI GPU with 48GB Memory and Gen5 Compatibility
Chinese company Moore Threads has introduced the MTT S4000, a new GPU accelerator for Large Language Models (LLMs). The MTT S4000 features a third-generation MUSA architecture and is fully compatible with the CUDA framework through the self-developed MUSIFY translation tool. With 48GB of GDDR6 memory, the accelerator is designed to meet the high memory capacity requirements of LLMs. It is intended for data-center applications and supports up to 96 video decoding streams at 1080p resolution. Moore Threads is targeting larger customers and also offering the KUAE Smart Computing Center Solution, a GPU-based stack for large-scale computing. The company has yet to provide more details on the S4000 GPU's specifications.