Moore Threads Announces Next-Gen GPUs with Astonishing Performance Claims

Table of Contents

Add a header to begin generating the table of contents

Chinese GPU manufacturer Moore Threads has unveiled its "Huagang" architecture, promising substantial performance leaps for its upcoming "Lushan" gaming and "Huashan" AI GPUs. The company claims these new chips, slated for release next year, will offer dramatic improvements over current offerings, aiming to compete with established players in the GPU market.

Key Takeaways

Moore Threads’ "Huagang" architecture powers new "Lushan" (gaming) and "Huashan" (AI) GPUs.
Claims include a 15x boost in AAA gaming and a 50x increase in ray tracing performance for Lushan.
Huashan AI GPU performance is touted as comparable to NVIDIA’s Hopper and Blackwell.
Full DirectX 12 Ultimate support and a second-generation ray tracing engine are featured.
Memory capacity is expected to increase significantly, with Lushan potentially offering up to 64 GB.

Lushan Gaming GPU

The "Lushan" gaming GPU, built on the new "Huagang" architecture, is positioned as the successor to Moore Threads’ existing MTT S80 and S90 models. The company is making bold claims, including a 15x improvement in "AAA" gaming performance and a remarkable 50x boost in ray tracing capabilities. Additionally, Lushan is expected to see a 64x increase in AI compute performance, 16x in geometry processing, 4x in texture fill rate, and 8x in atomic memory access. A significant upgrade is also anticipated in memory capacity, with Lushan potentially featuring up to 64 GB of VRAM, a substantial jump from the 16 GB found in current models.

Key features for Lushan include full support for modern APIs like DirectX 12 Ultimate, addressing compatibility concerns. It also boasts a dedicated second-generation hardware ray tracing engine and a new AI hardware block integrated into its "UniTE" unified rendering architecture, aiming to bring its rendering pipeline up to par with industry leaders.

Huashan AI GPU

Alongside the gaming GPU, Moore Threads also teased the "Huashan" AI GPU. This chiplet-based design, featuring multiple HBM modules, is claimed to offer performance comparable to NVIDIA’s Hopper and Blackwell GPUs, with memory bandwidth potentially exceeding that of the Blackwell B200. The Huashan GPU supports a wide range of compute formats from FP4 to FP64, along with proprietary low-precision mixed formats like MTFP4, MTFP6, and MTFP8. Moore Threads plans to scale these GPUs for large AI clusters, with interconnect capabilities supporting over 100,000 GPUs via MTLink 4.0 at speeds up to 1314 GB/s.

The "Huagang" architecture itself promises a 50% increase in compute density and a 10x improvement in efficiency. While specific benchmarks for the Lushan and Huashan GPUs were not provided, Moore Threads did demonstrate the MTT S5000 GPU achieving impressive speeds on the DeepSeek V3 model, suggesting competitive AI performance.

Future Outlook

Moore Threads’ ambitious claims signal a strong push to establish a significant presence in both the consumer gaming and enterprise AI markets. The company’s focus on next-generation architectures and performance improvements, coupled with support for the latest APIs, indicates a determined effort to challenge established GPU manufacturers. Further details are expected as the launch dates approach.