Moortech S5000 GPU Breakthrough Powers China Mobile's Jiutian AI Model
At the upcoming 9th Digital China Summit, China Mobile's self-developed "Jiutian" 35B general-purpose large language model will make its official public debut. As a significant advancement for the domestic computing ecosystem, Moore Threads recently announced that its flagship, full-featured GPU, the MTT S5000, has completed full-process adaptation and inference verification for this model.
The core of this adaptation lies in deep integration. Leveraging its proprietary MUSA software stack and the SGLang-MUSA high-performance inference engine, Moore Threads successfully implemented the entire inference pipeline for the "Jiutian" 35B model. Through collaborative optimization of the MUSA C development framework, the muDNN computing library, and the open-source MATE operator library, the MTT S5000 has been finely tuned for the specific attention mechanisms and long-sequence inference requirements of large models. This ensures efficient and stable performance when processing lengthy texts and handling high-concurrency requests.

The MTT S5000 computing card, serving as the technical foundation for this adaptation, has demonstrated exceptional capabilities. Built on the fourth-generation MUSA "Pinghu" architecture, this GPU delivers a maximum AI dense computing power of up to 1000 TFLOPS per card. Its hardware configuration features 80GB of high-capacity VRAM with a memory bandwidth of 1.6 TB/s, supporting full-precision computing from FP8 to FP64. Furthermore, a high inter-card interconnect bandwidth of 784 GB/s ensures excellent scalability in complex intelligent computing scenarios.
This collaboration not only validates the reliability of domestic GPUs in supporting core large models from central state-owned enterprises but also highlights Moore Threads' maturity in high-performance operator optimization and software ecosystem development. With the official launch of the "Jiutian" 35B model, this "domestic large model + domestic computing power" combination provides a highly relevant practical case for achieving independent and controllable computing infrastructure.
Related article
DeepMind CEO Hassabis: I sleep six hours a day, usually feel energetic around 1 a.m.
Fortune recently featured an interview with Demis Hassabis, CEO of Google DeepMind, revealing his unconventional approach to rest and productivity. Hassabis disclosed that he sleeps very little, structuring his waking hours into two distinct work blo
OpenAI, Anthropic Vie for Market Share Despite Revenue Shortfalls
Despite recent reports suggesting OpenAI missed revenue targets, creating pressure on tech stocks this Tuesday, private AI lab investors remain resilient. Seasoned backers have confirmed they will not reduce investment despite negative media coverage
California AV Compliance: A New Era of Tickets, Geofences, and 1M Miles
Guident operates an AuveTech shuttle in South Florida, managing a four-mile route in West Palm Beach and a one-mile route in Boca Raton using its remote monitoring technology. | Credit: GuidentCalifornia is redefining the regulatory landscape for dri
Related Special Topic Recommendations
Comments (0)
0/500
At the upcoming 9th Digital China Summit, China Mobile's self-developed "Jiutian" 35B general-purpose large language model will make its official public debut. As a significant advancement for the domestic computing ecosystem, Moore Threads recently announced that its flagship, full-featured GPU, the MTT S5000, has completed full-process adaptation and inference verification for this model.
The core of this adaptation lies in deep integration. Leveraging its proprietary MUSA software stack and the SGLang-MUSA high-performance inference engine, Moore Threads successfully implemented the entire inference pipeline for the "Jiutian" 35B model. Through collaborative optimization of the MUSA C development framework, the muDNN computing library, and the open-source MATE operator library, the MTT S5000 has been finely tuned for the specific attention mechanisms and long-sequence inference requirements of large models. This ensures efficient and stable performance when processing lengthy texts and handling high-concurrency requests.

The MTT S5000 computing card, serving as the technical foundation for this adaptation, has demonstrated exceptional capabilities. Built on the fourth-generation MUSA "Pinghu" architecture, this GPU delivers a maximum AI dense computing power of up to 1000 TFLOPS per card. Its hardware configuration features 80GB of high-capacity VRAM with a memory bandwidth of 1.6 TB/s, supporting full-precision computing from FP8 to FP64. Furthermore, a high inter-card interconnect bandwidth of 784 GB/s ensures excellent scalability in complex intelligent computing scenarios.
This collaboration not only validates the reliability of domestic GPUs in supporting core large models from central state-owned enterprises but also highlights Moore Threads' maturity in high-performance operator optimization and software ecosystem development. With the official launch of the "Jiutian" 35B model, this "domestic large model + domestic computing power" combination provides a highly relevant practical case for achieving independent and controllable computing infrastructure.
DeepMind CEO Hassabis: I sleep six hours a day, usually feel energetic around 1 a.m.
Fortune recently featured an interview with Demis Hassabis, CEO of Google DeepMind, revealing his unconventional approach to rest and productivity. Hassabis disclosed that he sleeps very little, structuring his waking hours into two distinct work blo
OpenAI, Anthropic Vie for Market Share Despite Revenue Shortfalls
Despite recent reports suggesting OpenAI missed revenue targets, creating pressure on tech stocks this Tuesday, private AI lab investors remain resilient. Seasoned backers have confirmed they will not reduce investment despite negative media coverage





Home






