Chinese chipmaker, Moore Threads, has unveiled its two brand new GPUs, Lushan for gaming & Huashan for AI, promising big performance boosts.
Moore Threads Claims Its Upcoming Gaming GPU, Lushan, Delivers 15 Times Higher Gaming Performance With 50x Boost In RT, 64x Boost In AI, and Quadruple The Memory Capacities
Well its official, Moore Threads has announced its brand new GPUs based on its next-generation Flower Harbor (Huagang) architecture at the MUSA 2025 summit. The architecture will power the Lushan and Huashan GPU products, aimed at Gaming and AI workloads, respectively.
As per Moore Threads, the Flower Harbor architecture comes with an improved compute unit, which heightens the compute density of the chip by 50%, and energy efficiency also sees a 10% improvement.
The architecture is fully compliant with FP4 and up to FP64 compute formats, and Moore Threads also added exclusive formats such as MTFP6 and MTFP4 for mixed low precision. The architecture also packs an asynchronous programming model and an ultra-large-scale interconnect, which will be supported by the AI-ready Huashan GPUs. This allows the company to scale more than 100,000 GPUs within a cluster using the MTLink high-speed interconnect technology.
The following are the full compute formats supported by the new architecture:
FP64
FP32
TF32
FP16
BF16
FP8
FP6
FP4
INT8
MTFP8
MTFP6
MTFP4
Starting with the Lushan, this is a gaming and content creation chip that is designed as the successor to the Moore Threads MTT consumer family. The company isn’t disclosing any product details right now, but they have detailed what users can expect from the upcoming products when they launch. Cards based on this chip will be replacing the older MTT S80 and MTT S90 products, which have been out for a couple of years.
Coming to the details, the Lushan Gaming GPU from Moore Threads is expected to offer a 15x boost in AAA gaming performance, 64x boost in AI compute, 16x the geometry processing performance, 4x the texture fill rate performance, 8x atomic memory access performance, and a massive 50x boost in RT (ray tracing) capabilities.
Next-Gen Moore Threads Gaming GPU Improvements:
64x AI Compute Boost
50x Ray Tracing Boost
15x Gaming Performance Boost
16x Geometry Processing Boost
4x Texture Fill Rate Boost
8x Atomic Access Performance
4x Memory Capacity
One major upgrade on Moore Threads’ part is that this new architecture is fully compliant with the latest APIs, including DirectX 12 Ultimate. The previous MTT consumer lineups severely lacked here, and it looks like the company is trying to make up for the shortcomings in the upcoming architecture. They have also highlighted a new AI generative rendering design under the UniTE unified rendering architecture, while the brand new hardware ray-tracing engine is expected to deliver some big gains, paving the path ahead for Neural Rendering and Path Tracing.
As for the Huashan GPU, the GPU seems to be made up of two chiplets and packs 8 HBM sites. Moore Threads compares the capabilities of its Huashan AI GPU against NVIDIA’s Hopper and Blackwell GPUs. The Floating-Point compute seems to be close to Blackwell B200, while the total bandwidth matches that of the B200, and in memory access capacity, the Huashan AI GPU seems to excel beyond the Blackwell chip.
Besides this, the new GPUs are expected to pack 4x the memory capacity. The MTT S80/S90 packs 16 GB GDDR6 memory, so we can expect up to 64 GB memory in the upcoming solutions.
Once again, Moore Threads is expected to launch the first graphics cards in the Lushan gaming lineup in 2026, while the AI products are expected around a similar timeframe.
Lastly, Moore Threads also shared some performance figures of its MTT S5000 GPU which is said to tackle the NVIDIA Hopper lineup of AI GPUs, achieving 1000 Tokens/s in Decode and 4000 Tokens/s in Prefill on DeepSeek V3. The MTT S5000 is expected to be part of the MTT C256 supernode server which is expected next year.
Follow Wccftech on Google to get more of our news coverage in your feeds.