Korean company looks to Intel CPUs to replace Nvidia GPUs

korean company looks to intel cpus to replace nvidia gpus 65453caf40b40 | Dang Ngoc Duy

Korean Economic Daily cited industry sources that the average price of an Nvidia graphics chip model for AI training in Korea has doubled, from 40 million won (29,561 USD) at the beginning of the year to 80 million won (59,193 USD).

But even if you have enough money, buying an AI chip is not easy. Businesses may have to wait 52 weeks, or go to a third market with a much more expensive price. Others are forced to find solutions with Intel CPUs in AI training machines.

Intel Xeon Sapphire Rapids chip for Intel servers. Photo: Intel

Intel Xeon Sapphire Rapids chip for Intel servers. Photo: Intel

For example, Naver Corporation, the company behind Naver – Korea’s leading search portal, currently uses AI models to distinguish false information. Data processing, inference, and verification are performed through a system using Nvidia GPUs.

Typically, a GPU for an AI server processes inference 10 times faster than a CPU. Recently, Intel has significantly improved the processing time of AI inference on its chips, but is still far behind products from Nvidia. However, Naver still chooses Intel in the context of scarce supply of Nvidia chips.

Naver spent a month testing the AI server running on Intel CPUs before putting it into use. In particular, the company mainly upgraded the system with the fourth generation Sapphire Rapids CPU – a data center microprocessor model launched earlier this year and is considered Intel’s most sustainable.

According to analysts, Naver’s move to Intel will affect Nvidia’s market share in Korea. The company is said to be having to reallocate $5 billion worth of GPUs because it is banned from selling to Chinese customers. “It’s a shame it didn’t happen sooner because it could have helped Nvidia keep Naver as its customer,” an expert commented.

Some other analysts assess that the cooperation between Naver and Intel could reduce Nvidia’s influence in the AI processor market, when some other companies also choose similar alternatives. Meanwhile, the company’s global position is difficult to shake and still holds more than 80% of the AI GPU market share.

Naver, Intel and Nvidia have not yet commented.

Bao Lam

Leave a Reply

en_USEN