Chinese GPU manufacturer Moore Threads confirmed at its 2026 semi-annual results briefing on September 3rd that the “Lushan” GPU chip, built on the new fifth-generation “Huagang” architecture, is expected to officially launch in the fourth quarter of this year. As China’s first domestic GPU to support hardware real-time ray tracing, “Lushan” carries high expectations yet has also sparked controversy: some Chinese media outlets use headlines like “surpassing the RTX 5090” to grab attention, but clicking through reveals they mean surpassing it in memory capacity, not actual gaming performance.
Moore Threads Huagang Architecture Second Generation: Lushan GPU Technical Specifications
Moore Threads CEO Zhang Jianzhong revealed during the earnings call that “Lushan” is the second main chip under the fifth-generation “Huagang” architecture, positioned for the high-performance graphics rendering market. The architecture adopts a new-generation instruction set design, delivering 50% higher compute density and 10x energy efficiency compared to the previous generation, with full support for precision computing ranging from FP4 to FP64.
In terms of core technology, “LuShan” integrates a first-generation AI Generative Rendering architecture (AGR) and a second-generation hardware ray tracing acceleration engine, fully compatible with the DirectX 12 Ultimate standard. Moore Threads had already unveiled these two core rendering technologies at its inaugural MUSA Developer Conference in December 2025. Among them, the hardware ray tracing engine delivers complete support for DirectX Raytracing (DXR), making real-time, realistic lighting effects possible for the first time on a domestic GPU; AGR, meanwhile, drives a paradigm shift in rendering technology from traditional “computation” to “generation.”
Performance Leap: 15x Boost in AAA Games, 50x Surge in Ray Tracing
Compared to the previous generation S80, the performance improvements Moore Threads has disclosed are quite striking: 15x faster rendering performance in 3A games, 50x surge in ray tracing performance, 16x improvement in geometry processing, and 8x improvement in atomic memory access. AI compute performance has reached a 64x improvement, with a unified task engine (UNITE) that centrally schedules all computing units. Combined with the MUSA software stack’s full coverage of mainstream inference frameworks such as vLLM, SGLang, and Ollama, this card also demonstrates solid capability in local large-model inference.
Memory capacity can reach up to 64GB, a key figure that later sparked controversy over headlines. On the financial side, Moore Threads reported first-half 2026 revenue of RMB 1.736 billion (approximately NT$7.8 billion), up 147.42% year-over-year, already surpassing its full-year 2025 level. R&D investment reached RMB 769 million, up 38.16% year-over-year. The number of developers in the MUSA ecosystem surpassed 800,000, with 2,167 cumulative patent applications.
Headline Controversy: Wordplay That Goes Beyond the RTX 5090
“Lushan” does have noteworthy breakthroughs in technical specifications, but the practice of some Chinese media outlets using “surpassing the RTX 5090” as news headlines has sparked widespread discontent among netizens.
The actual article text is marked in red: “VRAM capacity can reach up to 64GB, surpassing the RTX 5090’s 32GB,” and it adds that “the industry generally believes its performance is on par with the RTX 50 series,” rather than that its overall graphics computing power exceeds NVIDIA’s flagship card. (The article’s headline was so heavily criticized that it was changed to “Historic Breakthrough for Domestic Graphics Cards.”)
Many netizens flooded the comment section after reading the report to express their displeasure: “Can’t you survive without hyping things up?” “Bragging every single day, have you no shame! Who can you possibly fool? The stock price is about to hit limit-down, that’s what you get for bragging!” “The headline says it surpasses the 5090, but when you click in, it turns out it’s just the VRAM capacity that surpasses it.” This kind of clickbait tactic, swapping “VRAM capacity” for the concept of “performance surpass,” has instead cast a shadow over the actual technological breakthroughs of the Lushan GPU. (Moore Threads kind of got caught in the crossfire.)
Historical Context and Market Positioning
Moore Threads was founded in 2020 by Zhang Jianzhong, former NVIDIA Vice President of Worldwide Sales and General Manager of Greater China, and is one of the few Chinese companies with independent GPU design capabilities (and one that produces general-purpose graphics cards capable of gaming). Its first product, the S80, launched in 2022, targeting the entry-level gaming and workstation markets, but due to the lack of hardware ray-tracing support, it lagged significantly behind mainstream NVIDIA and AMD products in AAA gaming experiences. The hardware ray-tracing support in “Lushan” is seen as a critical step toward closing that gap. Compared with other competitors also investing in China’s domestic GPU space, manufacturers such as Jingjia Micro and XinDong Technology have yet to commercialize consumer-grade hardware ray tracing, giving Moore Threads a relatively leading position in this area. In terms of market positioning, Lushan’s target customers are primarily domestic Chinese gamers and the government procurement market. With US export controls continuing to tighten, the importance of independent R&D in China’s domestic GPU sector has grown even more apparent. In fact, Chinese manufacturers have achieved some notable results in recent years—beyond Moore Threads, Lucid Technology recently released a graphics card with performance close to an RTX 4060. If “Lushan” truly delivers RTX 50-series-level performance, coupled with 64GB of VRAM, it could be a sweet-spot GPU for AI applications (provided the ecosystem is good enough).
However, from the perspective of actual performance positioning, the industry generally believes that “Lushan”‘s overall computing capability is comparable to mid-range RTX 50 series products (such as the RTX 5060 or 5070), rather than the top-tier RTX 5090. Furthermore, Moore Threads has yet to release any third-party independent gaming frame rate test data—all performance figures are officially self-reported, lacking cross-verification from other media outlets or testing institutions.
Outlook: Q4 listing still faces uncertainties
The “Lushan” is expected to officially launch in the fourth quarter of this year, but Moore Threads’ official stance remains relatively cautious, only saying it is “expected to be on the market within the year,” without giving a specific release date. This also reflects the long-standing difficulties facing domestic GPUs. As industry insiders often say: there is no shortage of spec numbers at launch events; what is lacking are products that consumers can actually buy and that can stably run ray-traced games. Looking back at the performance of the Moore Threads S80 after its release, although the official claims put its performance on par with the GTX 1060, actual game compatibility and driver stability still lag behind international giants. If Lushan can launch as scheduled, it will mark the first time a Chinese domestic GPU has planted its flag in the hardware ray tracing field, but whether it gains market recognition will require independent reviews after the actual product is available.