Chinese processor manufacturer Loongson Technology announced on September 15 via an interactive platform that the development of its first GPGPU chip, the 9A1000, is nearly complete, with tape-out scheduled for the third quarter. The success of the chip will depend on the test results after the tape-out.
The 9A1000 is Loongson’s first GPGPU chip officially disclosed in June this year, positioned as a low-cost GPGPU, and it is also the company’s first dedicated GPGPU chip aimed at graphics cards/AI acceleration cards. The graphics performance is comparable to the AMD RX550 and supports OpenGL 4.0; in terms of terminal AI computing performance, the INT8 performance exceeds 32 TOPS. Additionally, the 9A1000 will integrate a video processing module that supports H.264/H.265 encoding and decoding.
Loongson’s Chief Engineer and Head of General GPU Processor Development Su Menghao further pointed out that the 9A1000 offers more than a fivefold performance improvement compared to the second-generation GPGPU core LG200 integrated in the 2K3000, with AI computing power reaching 40 TOPS. This is mainly due to a comprehensive upgrade of the GPU core in the 9A1000, which has doubled the graphics pipeline, increased the main frequency by 25%, and reduced the area of the stream processors by 20%, leading to a 70% reduction in light-load power consumption. Furthermore, the GPU scale of the 9A1000 has increased to four times that of the LG200. This chip is expected to be launched in 2026.
In addition to the 9A1000, Loongson also disclosed the independent graphics card chip 9A2000 aimed at the mid-to-high-end market, which integrates four 9A1000 chips together, upgrades the graphics API to OpenGL 4.6, and adds support for virtualization. Besides supporting INT8 and INT4, the tensor units also increase support for floating-point data types such as BF16 and FP8. In terms of computing power, the single-precision floating-point performance reaches 5 TFLOPS, and the INT8 AI computing power reaches 160 TOPS, with a memory bandwidth of 256 GB/s. Additionally, the 9A2000 supports dual-chip interconnection, allowing for an overall performance increase of up to double, reaching an internationally advanced level under the same process. This chip is expected to be launched in 2027.
Loongson Chairman Hu Weiwu also revealed that the upcoming 9A3000 will be a higher-performance GPGPU, which could achieve a 3-5 times performance improvement if advanced Xnm technology is adopted.
Editor: Chip Intelligence – Wandering Sword
Previous Exciting ArticlesArm Executive Responds to Xiaomi’s Self-Developed Chips!Regarding Chips, the Ministry of Commerce Takes Action! Two Major Associations Speak Out!Rumor: Micron DRAM Suspends Pricing, Expected to Increase by 30%!Domestic Substitution Accelerates! China Launches Anti-Dumping Investigation Against US Analog Chips!Rumor: Oracle China Layoffs: Compensation N+6!Breaking News! The US Includes 23 Chinese Entities, Including Fudan Microelectronics, on the “Entity List”Zhaoxin KX-7000 Empowers Seewo PC, Assisting in Educational Digital TransformationWei Shaojun: China Should Abandon Using Nvidia GPUs for AI DevelopmentArm Lumex CSS Released: Detailed Explanation of the New C1 CPU and G1-Ultra GPUPerformance Below Expectations, Synopsys Cuts 10% of Workforce! Stock Price Plummets 35%Apple iPhone 17 Series Released: Introduced Three Self-Developed Chips!ASICs Decline, Broadcom Hits Back, Nvidia Claims Its GPUs Offer Better Cost-Performance Ratio!Dell China Layoffs, Compensation N+3!Hu Yuhua Appointed Senior Vice President of Sales and Marketing for NXP Greater ChinaAMD: Compared to x86, Arm Architecture Has No Advantages!
For industry communication and cooperation, please add WeChat: icsmart01
Chip Intelligence Official Group: 221807116