By | Zhou Luyao
Perhaps, we will soon be able to plug a Loongson GPU into a Windows PC!
On November 17, 2025, Loongson Technology (688047.SH) stated in its latest investor relations activity record that the first GPGPU chip, the 9A1000, has been taped out at the end of September, and they are currently working on developing Windows drivers for the 9A1000 to make it compatible with Windows PCs. Furthermore, the Loongson 9A1000 integrates graphics and AI computing power, making it suitable for AIPC.
In terms of performance, according to previous information disclosed by Loongson Technology, the graphics performance of the Loongson 9A1000 exceeds that of the integrated graphics in CPUs, positioning it as an entry-level discrete graphics card, primarily competing with AMD’s Radeon RX 550 discrete graphics card released in 2017. In terms of AI computing, the Loongson 9A1000 can provide 32-40 TOPS of AI computing power, slightly lower than the performance of the XDNA 2 NPU (up to 50 TOPS) integrated in AMD’s Ryzen AI Max+ (codenamed Strix Halo) processor. Functionally, the Loongson 9A1000 supports OpenGL 4.0 and OpenCL ES 3.2 application programming interfaces.
In fact, the Loongson 9A1000 is not the only GPU chip from Loongson Technology; the company is also planning the 9A2000 and 9A3000.
It is reported that, unlike the Loongson 9A1000, the Loongson 9A2000 is a high-performance GPU chip aimed at mid-to-high-end graphics cards, applicable to desktops and servers, designed to be optimized and strengthened.
In terms of specific performance, the GPU core of the Loongson 9A2000 has been upgraded to a third-generation architecture, further enhancing computing power per unit area, supporting graphics API OpenGL 4.6, adding virtualization support, and tensor units supporting more data types. The GPU scale is x4 (vs 9A1000), with single-precision floating-point computing power of 5 Tflops, INT8 AI computing power of 160 TOPS, memory bandwidth of 256 GB/s, supporting dual-chip interconnection, and overall performance doubling, reaching an internationally advanced level for the same process generation.
It is worth noting that during this investor relations activity, in addition to the GPU, Loongson Technology also revealed its progress in advanced processes.
Loongson Technology emphasized that the Xnm advanced process 32-core and above server chip 3D7000 is a key R&D focus for Loongson from 2025 to 2027. Depending on the progress of the Xnm process, it is also possible to first develop a 1Xnm process 16-core server chip 3C6600. The company has already begun IP design work for advanced X-nanometer processes, including phase-locked loops, multi-port register files, DDR5-PHY, PCIe5-PHY, etc., as the independent development of these IPs generally starts one and a half to two years before the actual chip development begins.
If you have any comments on the article, please email [email protected] to indicate any inappropriate content. Upon receiving the email, we will quickly initiate an internal processing mechanism.
