Moore Threads just announced two new GPUs at their summit. The gaming-focused one is called Lushan, and the AI model is named Huashan. Both use their next Flower Harbor architecture. The company claims massive performance jumps, saying Lushan will offer fifteen times the gaming speed and fifty times better ray tracing over their old cards. It will also support modern standards like DirectX 12 Ultimate, which earlier products lacked.
The Huashan AI processor uses a chiplet design with eight HBM memory sites. Moore Threads compared its specs directly to Nvidia's Blackwell, suggesting competitive floating-point compute and bandwidth. They also mentioned scaling over a hundred thousand GPUs in a cluster using their proprietary MTLink tech. Memory capacity across the new lineup is supposed to quadruple, potentially reaching sixty-four gigabytes.
They provided some numbers for their current MTT S5000 AI GPU, claiming it hits a thousand tokens per second on certain tasks. The first products based on the new Lushan and Huashan designs are slated for a 2026 launch.
The Huashan AI processor uses a chiplet design with eight HBM memory sites. Moore Threads compared its specs directly to Nvidia's Blackwell, suggesting competitive floating-point compute and bandwidth. They also mentioned scaling over a hundred thousand GPUs in a cluster using their proprietary MTLink tech. Memory capacity across the new lineup is supposed to quadruple, potentially reaching sixty-four gigabytes.
They provided some numbers for their current MTT S5000 AI GPU, claiming it hits a thousand tokens per second on certain tasks. The first products based on the new Lushan and Huashan designs are slated for a 2026 launch.
