Press "Enter" to skip to content

China’s Moore Threads launches advanced MTT S4000 for AI and data center workloads

Moore Threads, a prominent Chinese GPU manufacturer, has recently unveiled the MTT S4000, a high-performance graphics card specifically engineered for AI applications and data center operations. This new flagship GPU, according to a recent report by Tom’s Hardware, is set to be a key component in the KUAE Intelligent Computing Center, which will feature clusters each equipped with 1,000 units of the S4000. Additionally, Moore Threads is collaborating with several major Chinese firms, including Lenovo, to strengthen its KUAE ecosystem, encompassing both hardware and software solutions.

The MTT S4000 marks a significant technological advancement over its predecessors, the S2000 and S3000. While complete specifications are yet to be disclosed, the S4000 demonstrates remarkable improvements in key performance areas. It significantly outperforms the S2000 in terms of FP32 and INT8 metrics, offers increased VRAM, and boasts enhanced memory bandwidth. The S4000 is powered by the advanced third-generation MUSA architecture, a step up from the first-generation architecture used in earlier models.

When compared to Nvidia’s product line, the S4000 surpasses the older Turing-based Tesla server GPUs but remains behind the more recent Ampere and Ada Lovelace series. Nevertheless, the S4000’s robust memory specifications make it a strong contender for AI-focused tasks and processing large language models.

The KUAE Intelligent Computing Center: A hub for AI innovation

In conjunction with the S4000 launch, Moore Threads also introduced the KUAE Intelligent Computing Center. This center represents a holistic approach, merging cutting-edge hardware with sophisticated software. At the heart of the KUAE clusters are the MCCX D800 GPU servers, each equipped with eight S4000 GPUs. The software infrastructure of KUAE is designed to support leading AI models and frameworks, with Moore Threads’ MUSIFY tool ensuring compatibility with Nvidia’s CUDA ecosystem.

To navigate the complexities of developing advanced GPU technologies and software ecosystems, Moore Threads has established strategic alliances with other Chinese tech companies, notably including Lenovo. This collaboration is crucial for Moore Threads, especially in light of the challenges posed by U.S. sanctions that limit China’s access to high-end GPUs from foreign suppliers.

While Moore Threads may not be directly competing with global GPU giants, its latest developments are crucial for China’s burgeoning semiconductor industry. The S4000 and the KUAE platform could potentially carve out a significant market share within China, especially given the restrictions on Nvidia’s China-specific GPU offerings.

Source: ReadWriteWeb