Factories in China are converting GeForce RTX 4090 graphics cards into specialized AI solutions – Nvidia

The ban severely limits the sale of high-performance devices used to train models and accelerate artificial intelligence in Chinese territory. The ban also affected all NVIDIA partners that sell GeForce RTX 4090 graphics cards due to their high computing capabilities.

Before the government order on restrictions was implemented, NVIDIA appears to have moved a large inventory of its AD102 GPU as well as GeForce RTX 4090 graphics cards to China, a move that may have contributed significantly to the global shortage of these graphics cards and the significant price spike that exceeded 2000. Euros in many cases.

In an interesting development, many professionals and users on the Chinese Baidu Forum revealed that many specialized factories across China are dismantling GeForce RTX 4090 graphics cards and converting them into artificial intelligence-powered acceleration solutions.

The conversion involves disassembling the graphics cards, removing the cooling systems, and extracting the AD102 graphics processor and GDDR6X memory from the PCBs. The above components are then soldered onto homemade PCBs specifically designed for AI acceleration applications, and then dual-slot blower-type cooling systems (such as those previously used in AMD Radeon graphics cards) suitable for server environments are applied.

Third-party cooling systems suitable for AI-accelerated graphics cards usually occupy three or four slots, while dedicated blower systems occupy only two slots. This allows more of these graphics cards to be placed and run in parallel on an AI accelerator server.

After rigorous testing, the redesigned GeForce RTX 4090 solutions are being made available to Chinese companies performing AI workloads. The retrofit has flooded the Chinese market with GeForce RTX 4090 PCBs and cooling systems, which are now being sold at very low prices since more expensive components like the GPU and memory were removed.

To ensure the good functioning of the graphics cards, they are tested by performing stress tests using different benchmarks and applications, such as Furmark, 3DMark, and various AI applications, while they are also tested with some smaller LLM certification holders to authenticate their computing capabilities.