Archive — CN — technode
Moore Threads unveils new Huagang GPU architecture at MUSA Developer Conference
Moore Threads announced its Huagang GPU architecture at the MUSA Developer Conference, aiming to enhance its role in AI and high-performance computing, highlighting increased compute density by 50% and energy efficiency gains. The architecture supports full-precision computing and enables scaling for intelligent computing clusters of up to 100,000 GPUs, alongside the introduction of two new chips, Huashan and Lushan.
Key points
- Moore Threads unveiled Huagang GPU architecture at MUSA Developer Conference.
- Huagang architecture offers 50% increase in compute density.
- Energy efficiency improved by tenfold with Huagang.
- Full-precision computing supported from FP4 to FP64.
- MTLink high-speed interconnect technology powers Huagang, allowing scalability of intelligent computing clusters up to 100,000 GPUs.
- Two new chips launched: Huashan for AI training/inference and Lushan for graphics rendering.
Why it matters
- The announcement aligns with global trends toward increasing computational power for AI and deep learning as AI applications continue to surge and require more powerful hardware.
- 2025 IT Trends indicate a rise in demand for efficient energy use in computing, addressing climate concerns and operational costs.
- Moore Threads' advancements mirror a competitive landscape where companies push boundaries in GPU technology to capture segments in AI and gaming.
The launch of the Huagang architecture by Moore Threads positions the company as a considerable player in the evolving landscape of AI and high-performance computing, reflecting broader industry trends focused on efficiency, scalability, and advanced capabilities.