As ITHome reported, China is pushing its home GPU efforts into uncharted territory with Lisuan Tech’s first shopper {and professional} graphics playing cards, the 7G106 and 7G105.
Built on TSMC’s 6nm N6 course of, the 7G106 and 7G105 are powered by the corporate’s in-house TrueGPU structure and purpose to compete instantly with mid-range choices from Nvidia and AMD. While the highlight is on gaming efficiency, Lisuan is positioning these chips as multi-purpose accelerators for AI, cloud rendering, and even metaverse functions.
The consumer-focused 7G106 options 12 GB of GDDR6 reminiscence on a 192-bit bus, 192 texture models, 96 ROPs, and an FP32 throughput of as much as 24 TFLOP/s. It options 4 DisplayPort 1.4 outputs with DSC 1.2b compression and helps DirectX 12 (minus ray tracing), Vulkan 1.3, OpenGL 4.6, and OpenCL 3.0. Its single 8-pin PCIe connector additionally suggests a TDP of round 225W.
On the opposite hand, the skilled 7G105 doubles reminiscence to 24 GB with ECC, providing as much as 192 GB/s of pixel fill price, 384 GB/s of texture fill price, and the identical 24 TFLOP/s compute ceiling. Both playing cards embody hardware-accelerated AV1 and HEVC decode as much as 8K60 and encode capabilities at 8K30 for HEVC and 4K30 for AV1.
Lisuan 7G106 and 7G105 Specifications
|
Specification | Lisuan 7G106 (Gaming) | Lisuan 7G105 (Professional/AI) |
|---|---|---|
| GPU Architecture | TrueGPU (G100) | TrueGPU (G100) |
| Process Node | TSMC N6 (6nm DUV) | TSMC N6 (6nm DUV) |
| Compute Performance | Up to 24 TFLOPs (FP32), INT8 assist | Up to 24 TFLOPs (FP32), INT8 assist |
| Texture Units (TMUs) | 192 | 192 |
| Raster Units (ROPs) | 96 | 96 |
| Memory | 12 GB GDDR6 | 24 GB GDDR6 (ECC) |
| Memory Bus | 192-bit | 192-bit |
| Memory Bandwidth | TBD (is determined by remaining clocks) | TBD (is determined by remaining clocks) |
| Video Encode | AV1 (4K @ 30 FPS), HEVC (8K @ 30 FPS) | AV1 (4K @ 30 FPS), HEVC (8K @ 30 FPS) |
| Video Decode | AV1 & HEVC (as much as 8K @ 60 FPS) | AV1 & HEVC (as much as 8K @ 60 FPS) |
| API Support | DirectX 12, Vulkan 1.3, OpenGL 4.6, OpenCL 3.0 | DirectX 12, Vulkan 1.3, OpenGL 4.6, OpenCL 3.0 |
| PCIe Interface | PCIe 4.0 x16 | PCIe 4.0 x16 |
| Display Outputs | 4 × DisplayPort 1.4 (DSC 1.2b, as much as 8K60) | 4 × DisplayPort 1.4 (DSC 1.2b, as much as 8K60) |
| Power Connector | 1 × 8-pin (225 W max implied) | 1 × 8-pin (225 W max implied) |
| vGPU / SR-IOV Support | Up to 16 digital GPUs | Up to 16 digital GPUs |
What units Lisuan other than previous Chinese GPU makes an attempt is its declare of constructing the TrueGPU structure from scratch, together with the instruction set, compute core, and software program stack. The firm touts “intelligent multitasking” with as much as 48 concurrent duties, out-of-order triangle rendering for 50% sooner effectivity in sure scenes, and twin FP32/INT32 instruction emission.
There’s additionally a singular matrix reminiscence structure designed to spice up reminiscence effectivity by 40%, together with dynamic load balancing that distributes rendering and compute duties throughout cores in actual time. Lisuan even claims NRSS, a proprietary rendering high quality optimization system designed to rival Nvidia DLSS and AMD FSR.
Benchmark outcomes paint a blended however promising image. In artificial checks, the 7G106 scored 26,800 factors in 3DMark Fire Strike and a couple of,256 in Steel Nomad, placing it roughly on par with Nvidia’s RTX 4060 in Fire Strike. Geekbench 6 OpenCL noticed it notch 111,290 factors, edging out the RTX 4060 by round 10%.
Gaming demos have been equally noteworthy: Black Myth: Wukong and Wuchang: Fallen Feathers each ran at over 70 FPS in 4K High settings, whereas Shadow of the Tomb Raider topped 80 FPS below comparable circumstances. High-profile titles like these working on a GPU structure in its infancy paint a really optimistic image for the platform’s stability, extra so than uncooked efficiency numbers.
The 7G105 skilled variant leans closely on AI and enterprise markets, supporting SR-IOV virtualization with as much as 16 containers. Lisuan highlights its applicability for cloud gaming, digital twins, digital actuality, and even robotics. The firm claims its eXtreme collection playing cards are able to deal with giant AI fashions comparable to DeepSeek and Wenshengtu, extending their utility past conventional PC gaming.
Lisuan’s announcement marks a brand new section for China’s home GPU ambitions. While previous efforts, comparable to Zhaoxin’s built-in options and Moore Threads’ early discrete GPUs, struggled to interrupt into the mainstream, Lisuan’s 6nm designs characterize a daring try to shut the hole with world gamers. If its claims of architectural independence and efficiency parity maintain up below impartial testing, this could possibly be the primary time a Chinese GPU maker actually competes with AMD and Nvidia within the discrete GPU house.
Mass manufacturing of the Lisuan 7G106 and 7G105 is predicted to start in September 2025, following sampling in August. Pricing and remaining clock speeds are but to be introduced. Still, Lisuan’s domestic-first technique might make these playing cards a viable various for Chinese avid gamers and enterprises going through rising prices and export restrictions.
Follow Tom’s Hardware on Google News to get our up-to-date information, evaluation, and evaluations in your feeds. Make positive to click on the Follow button.