Chinese tech companies are now repurposing high-end NVIDIA gaming chips for AI

Chinese tech companies are now repurposing high-end NVIDIA RTX 4090 gaming GPUs as AI chips. This is in response to Washington’s restrictions on the export of high-performance processors. However, experts believes this is a stop-gap solution at best

In response to Washington’s restrictions on the export of high-performance processors, Chinese tech firms are turning to unconventional methods to acquire chips for their artificial intelligence (AI) projects.

NVIDIA gaming graphics cards, designed for PC gaming, are being repurposed as makeshift AI processors, providing a workaround to the scarcity of high-end chips in China.

Factory managers and chip buyers reveal that thousands of NVIDIA gaming graphics cards are being dismantled each month in Chinese factories and workshops. The stripped components are then integrated into new circuit boards, presenting a creative yet desperate solution to the challenges imposed by export controls.

Despite the raw computing power of NVIDIA’s gaming-focused products, industry experts note that they may not match the high-precision calculations required for training certain large language models with extensive datasets.

The limitations of interconnection speeds between chips further complicate efforts to address this issue by forming computing clusters.

Analyst Charlie Chai from research group 86Research describes the move as “a desperate move by Chinese companies under the export controls,” comparing it to using a kitchen knife for artwork—doable, but suboptimal.

The Biden administration’s tightened export controls on cutting-edge AI chips in October have made it more difficult for chip companies like NVIDIA to sell high-performance semiconductors to China. This has led to a surge in demand for the repurposed graphics processing units sourced from NVIDIA gaming cards.

While the repurposed components find buyers among public enterprises and small AI labs, concerns are raised about potential violations of NVIDIA’s intellectual property rights. Moreover, the gaming cards themselves could face bans from being sold to China at any time.

NVIDIA’s response to the export controls includes the release of a slower version of its banned cards, the GeForce RTX 4090 D, complying with the latest regulations. However, industry insiders warn that the modified versions may not be powerful enough for extensive language model training.

In an effort to meet China’s growing demand for AI systems while adhering to export controls, NVIDIA has developed three chips tailored for the Chinese market. Nevertheless, the performance of these chips is reportedly weaker than previous models, and they are not expected to be widely available until March.

Chinese customers have expressed dissatisfaction with the prices set for these inferior processors, prompting some companies to explore alternatives within China’s developing chip ecosystem. In the meantime, others are turning to NVIDIA’s less expensive gaming chips, hoping for successful albeit temporary solutions.

Buyers remain cautiously optimistic, acknowledging the uncertainty of this reinvention but expressing a collective hope that these adapted machines will prove usable, at least in the short term.

(With inputs from agencies)

Chinese tech companies are now repurposing high-end NVIDIA RTX 4090 gaming GPUs as AI chips. This is in response to Washington’s restrictions on the export of high-performance processors. However, experts believes this is a stop-gap solution at best

In response to Washington’s restrictions on the export of high-performance processors, Chinese tech firms are turning to unconventional methods to acquire chips for their artificial intelligence (AI) projects.

NVIDIA gaming graphics cards, designed for PC gaming, are being repurposed as makeshift AI processors, providing a workaround to the scarcity of high-end chips in China.

Factory managers and chip buyers reveal that thousands of NVIDIA gaming graphics cards are being dismantled each month in Chinese factories and workshops. The stripped components are then integrated into new circuit boards, presenting a creative yet desperate solution to the challenges imposed by export controls.

Despite the raw computing power of NVIDIA’s gaming-focused products, industry experts note that they may not match the high-precision calculations required for training certain large language models with extensive datasets.

The limitations of interconnection speeds between chips further complicate efforts to address this issue by forming computing clusters.

Analyst Charlie Chai from research group 86Research describes the move as “a desperate move by Chinese companies under the export controls,” comparing it to using a kitchen knife for artwork—doable, but suboptimal.

The Biden administration’s tightened export controls on cutting-edge AI chips in October have made it more difficult for chip companies like NVIDIA to sell high-performance semiconductors to China. This has led to a surge in demand for the repurposed graphics processing units sourced from NVIDIA gaming cards.

While the repurposed components find buyers among public enterprises and small AI labs, concerns are raised about potential violations of NVIDIA’s intellectual property rights. Moreover, the gaming cards themselves could face bans from being sold to China at any time.

NVIDIA’s response to the export controls includes the release of a slower version of its banned cards, the GeForce RTX 4090 D, complying with the latest regulations. However, industry insiders warn that the modified versions may not be powerful enough for extensive language model training.

In an effort to meet China’s growing demand for AI systems while adhering to export controls, NVIDIA has developed three chips tailored for the Chinese market. Nevertheless, the performance of these chips is reportedly weaker than previous models, and they are not expected to be widely available until March.

Chinese customers have expressed dissatisfaction with the prices set for these inferior processors, prompting some companies to explore alternatives within China’s developing chip ecosystem. In the meantime, others are turning to NVIDIA’s less expensive gaming chips, hoping for successful albeit temporary solutions.

Buyers remain cautiously optimistic, acknowledging the uncertainty of this reinvention but expressing a collective hope that these adapted machines will prove usable, at least in the short term.

(With inputs from agencies)