Moore Threads, a renowned Chinese chipmaker, has made headlines with the introduction of its latest graphical innovations, promising significant advancements in gaming and AI performance. This leap forward is embodied in their new GPUs, Lushan for gaming and Huashan for AI, announced at the MUSA 2025 summit.
Unveiling the Flower Harbor Architecture
The heart of these advancements lies in the Flower Harbor architecture. This next-generation design brings a 50% increase in compute density and a 10% boost in energy efficiency. The architecture boasts compliance with FP4 to FP64 compute formats and introduces exclusive MTFP6 and MTFP4 formats for mixed low precision. It also features an asynchronous programming model and an ultra-large-scale interconnect, crucial for the AI-ready Huashan GPUs. With MTLink high-speed interconnect technology, Moore Threads aims to scale over 100,000 GPUs within a cluster.
Lushan: The Future of Gaming GPUs
The Lushan GPU is poised to revolutionize Moore Threads’ gaming and content creation offerings, succeeding the MTT consumer family. While specific product details remain under wraps, the Lushan chip promises to replace the older MTT S80 and MTT S90 models. Expect dramatic improvements with a 15x enhancement in AAA gaming performance, 64x in AI compute, 16x in geometry processing, and a 50x increase in RT capabilities. Additionally, it is now fully compliant with the latest APIs, including DirectX 12 Ultimate, addressing previous shortcomings in the MTT consumer lineups. The new architecture introduces AI generative rendering under the UniTE unified rendering architecture, alongside a cutting-edge hardware ray-tracing engine for future-ready Neural Rendering and Path Tracing.
Memory and Availability
The new GPUs are expected to offer four times the memory capacity of their predecessors. Given that the MTT S80/S90 models feature 16 GB GDDR6 memory, the upcoming solutions may provide up to 64 GB. Moore Threads plans to launch the Lushan gaming GPUs in 2026, with the AI products possibly arriving in the same timeframe.