Preparations underway in China for AI-enhanced RTX 5090s: Industrial manufacturing transforms gaming GPUs and memory onto server-compatible PCBs with powerful cooling systems
In an intriguing development, RTX 5090 gaming cards are being extensively refurbished in China to create enterprise and AI-focused "Turbo" blower-style models. This transformation is driven by the growing demand for AI and the falling prices of gaming GPUs, as well as geopolitical factors such as US export controls on higher-end AI-focused GPUs.
The process begins with boxed retail RTX 5090 cards from brands like Zotac, Palit, and Inno3D, predominantly the "D" variants available in China due to US export restrictions. Each card undergoes functional testing of the GPU die and memory to ensure stability, with any defective boards discarded. Only the best-performing GPU dies, representing high-quality silicon, are selected for reuse.
The original large, triple-fan coolers designed for gaming heat dissipation are removed. The “full-blood” GB202 chips and memory are then transplanted to a new server-ready PCB with blower-style cooling, better suited to enterprise racks and AI computational loads. This blower-style design is favored in server environments for its efficient, linear airflow for cooling hot components in dense racks.
The reassembled cards undergo 24-hour burn-in tests and repeated power cycles. The RTX 5090 Turbo models, produced by a Shenzhen-based company called "CT", are being used in AI training clusters, as up to eight cards can fit into a single chassis, making them ideal for AI training clusters.
This trend is a reflection of how the GPU market is shifting under the weight of AI demand. What was once a symbol of gaming excess, the RTX 5090 Turbo, has been reshaped for data centers. The efficiency of CT's operation is a glimpse at how quickly consumer hardware can be repurposed for enterprise needs.
However, whether NVIDIA condones this or turns a blind eye is uncertain, as the company has not pursued any investigation into these illicit operations before. The RTX 5090, a gaming GPU designed for enthusiasts with a power consumption of 600W, is being funneled into AI markets, echoing the shortages seen during the crypto mining boom.
In addition, these repurposed GPUs often emphasize stability and long-duration workloads rather than just raw clock speeds optimized for gaming. There is also mention of more exotic modifications, such as adding a second set of chips via clamshell conversions or custom BIOS/PCB modding for AI performance enhancements.
This news underscores the dynamic nature of the technology industry, where hardware designed for one purpose can be repurposed to meet the demands of another, in this case, AI and enterprise computing. As AI continues to grow and evolve, we can expect to see more innovations like this in the future.
- The process of creating enterprise and AI-focused "Turbo" blower-style RTX 5090 models involves removing original coolers, transplanting the "full-blood" GB202 chips and memory to a new server-ready PCB, and utilizing the efficient, linear airflow of a blower-style design more suitable for AI computational loads and server environments.
- The repurposing of the RTX 5090 GPUs for AI markets is a reflection of the dynamic nature of the technology industry, where hardware designed for gaming can be reimagined and revitalized to meet the demands of AI and enterprise computing.
