JHVEPhoto
South Korea on-line platform Naver’s announcement that it has changed synthetic intelligence accelerators from Nvidia (NASDAQ:NVDA) with these from Intel (NASDAQ:INTC) is “doable,” however solely in “very slim use circumstances,” funding agency Bernstein stated on Monday.
A gaggle of analysts from the funding agency, led by Bokyung Suh, stated that Naver, which is akin to Google, changing Nvidia’s (NVDA) H100 chips with Intel’s (INTC) Sapphire Rapids CPU and AI accelerators is feasible, nevertheless it’s not an “ideally suited” state of affairs, given how how expensive AI duties like coaching massive language fashions are.
“Even if NVIDIA H100 is way more costly on chip foundation, the full price for coaching and inference continues to be decrease no matter its manufacturing and provide subject,” the analysts wrote.
“In conclusion, we expect the collaboration between Naver and Intel is doable for very slim use circumstances, or inference perform solely, or low sufficient parameters; however undoubtedly not the sport changer but.”
The analysts added that regardless of the sensational information, Naver’s generative AI LLM mannequin has roughly 200B parameters, nevertheless it’s estimated that the Naver Place LLM parameter is a “tiny portion,” or lower than 5% of the general mannequin. As such, it is not for coaching or tuning, which is the place nearly all of AI prices are incurred.
“Which means, Inference for Gen AI and different fashions can run on CPUs; and, Intel’s Sapphire Rapids CPUs might be utilized with AI accelerators for inference,” they defined.
Naver introduced final week that it’s going to use Intel’s (INTC) Sapphire Rapids CPU and AI accelerators for its Naver Place AI, a suggestion and retailer and restaurant search service for inference.
Nvidia’s (NVDA) H100 GPU was unveiled in March 2022 and is extra highly effective than its earlier AI accelerator, the A100.
Nvidia (NVDA) shares rose round 1.5% in pre-market buying and selling on Monday, whereas Intel (INTC) noticed fractional losses.
