- Lisuan Tech GPU 7G105 between the professional market with 24 GB of VRAM and major promises
- The architecture of the startup jumps the tracing of the shelves and focuses on the workloads of calculation and virtualization for the general public
- The Lisuan Tech chip supports 8K decoding, four DisplayPort and HDMI jumps for cost reasons
A little -known company from China is preparing to enter the high -end GPU stage with its first generation of graphics cards.
While little outside the region have heard of Lisuan Tech, the startup has ambitions that place it squarely on the way to Nvidia and AMD, especially in the world requiring the workstation and professional IT.
Its newly announced SKU 7G105 has 24 GB of VRAM, a specification which suggests that it does not simply target players, but aimed at establishing credibility in the professional graphics segment.
Positioning beyond games with pro-grade specifications
At the heart of the Lisuan Tech offer is the 7G106 GPU, which is the basis for consumption and professional variants.
This chip is manufactured using the 6NM process of TSMC and is based on the internal Truegpu architecture of the company.
The architecture does not support the tracing of the shelves or DirectX 12 Ultimate, but it manages DirectX 12, Vulkan 1.3 and OpenGL 4.6.
Floating points operations are managed via a SIMD engine with support for FP32 and Int8 formats, reaching up to 24 Teraflops in FP32 performance, which puts it downright with mid -range GPUs of established players.
Although intended for games, the 7G106 has features that pour into the professional field.
The chip includes a video engine capable of decoding 8K AV1 and the HEVC content to 60 frames per second, and in coding in the two formats to 4K or 8K resolutions, depending on the codec.
For the exit, it supports four DisplayPort 1.4 ports with compression of display flows, completely avoiding HDMI, probably due to license costs.
When things start to look like a real workstation product is in virtualization capacity – the 7G106 supports SR -IIV, allowing the GPU to be partitioned in up to 16 virtual containers.
This allows a physical GPU to be shared on several virtual machines, which makes it suitable for business and hyperscalers’ environments that require flexibility and parallel workloads.
However, there are still major unknowns. The company has not yet finalized the GPU clock speeds, memory frequencies or energy consumption.
The first yields show a design using an 8 -pin PCIE connector, while another opts for a 16 -pin version, referring to the variance of the supply targets.
Without these data, performance estimates remain speculative despite the synthetic references such as the 3DMARK fire shot and the geekbench OpenCl showing decent scores.
This attempt is ambitious, but without clear pricing, maturity of the driver or proven reliability, it remains to be seen if Lisuan Tech can translate his technical affirmations into real traction.
Lisuan is not the first Chinese challenger to try to take Amd and Nvidia – Zhaoxin had trouble evolving beyond the basic functions of the GPU, the JMRO JM9 series by Jingjia Micro has never won a traction, and the fantastic range of Innosilicon has not succeeded in attracting real world users despite promising specifications.
Via Techpowerup
