Lisuan G100: China’s 6nm GPU should chase RTX 4060 – but fails due to the GTX 660 Ti

When a project is announced in a big way, with national pathos, technical superlatives and geopolitical thrust, then caution is advised. And if a Geekbench test then only spits out the performance level of an old GTX 660 Ti instead of a “4060 killer” – well, it’s time to bring the media hype back down to earth. Welcome to the first Chinese 6nm GPU, the Lisuan G100, and welcome back to 2012, at least in terms of performance.

Technology with pride, but no bite

Lisuan had big plans: a 6nm GPU, manufactured in-house, that would play on a par with NVIDIA’s RTX 4060. That sounds ambitious – and it was. In practice, the G100 delivered a meagre 15,524 points in Geekbench (OpenCL). This is not only below what even older mid-range GPUs from AMD and NVIDIA can manage, but it also scratches the surface of the “retro gaming with a lap of honor” category. Mind you, the benchmark result is below a GTX 1050 – yes, the “without-X” version. In other words: more “Graphics Capable” than “Graphics Card”.

The geopolitical context

We must not forget the circumstances under which this GPU was created. US embargoes, technology restrictions and trade wars – all of which are forcing China to become technologically independent. The strategy is clear: own chips, own software, own know-how. And all as quickly as possible. Lisuan presumably uses SMIC’s 6nm process – the same production node that breathed new life into Huawei’s Kirin processors. This is undoubtedly a considerable milestone. But an efficient node alone does not make a competitive GPU. Especially not if the software is based on drivers that are more reminiscent of “beta” than “ready for the mass market”.

Drivers – the Chinese bottleneck

Moore Threads, Birentech, Zhaoxin – they all share one curse: miserable drivers. Lisuan is not alone with this problem, but it is not in a good position either. Because without optimized software, even the best hardware will fizzle out like a rocket without control. OpenCL is just one measuring point, yes. But a significant one. And as long as even basic compatibility or memory management are bumpy, even the best architecture won’t help. By the way: 256 MB VRAM in a 6nm chip? Sounds like an engineering error or a placebo module for prototyping.

A glimmer of hope with limitations

The current performance is embarrassing – no question. But: It is most likely an early sample, perhaps a developer board with reduced specifications. The final chip could perform significantly better. Perhaps. Hopefully. So don’t write off the Lisuan G100 too soon. But it is also not (yet) the technological liberation that Beijing would like to market on its state channels.

Patriotism beats physics (not yet)

China’s path to technological independence is a marathon, not a sprint. The Lisuan G100 is a visible step – but more symbolic than substantial. The hardware, and above all the software, is still years behind. And if you think you can build an RTX 4060 from scratch with export bans and stubbornness, you should know that even 6nm won’t help if the rest is trial and error.

Source: Tom’s Hardware