Last week, Intel unveiled the Core Ultra and Xeon 5 chips, targeting the AI market. During a Q&A session with reporters, Pat Gelsinger mentioned NVIDIA, stating that it doesn’t have a significant advantage over CUDA. He also mentioned that the industry is shifting development to higher-level languages like Python, allowing for easier model porting.
He stated that once models are trained, there are now various options for running them, and these systems often no longer require CUDA. Gelsinger believes that the market for chips that run models will be the main driver in the future.
While Gelsinger’s comments might seem provocative for NVIDIA, the reality is that large tech companies are all producing chips to replace NVIDIA’s. However, they may only use these chips internally or provide them as a service on their own cloud. Intel’s situation is similar to AMD, which is trying to develop the ROCm software as an alternative to CUDA and selling both through the cloud and server markets. Intel also has its own training chip, Gaudi, which recently showcased Gaudi3.
TLDR: Intel recently unveiled the Core Ultra and Xeon 5 chips for the AI market. During a Q&A session, Pat Gelsinger mentioned NVIDIA and emphasized the shift towards higher-level languages like Python for model development and portability. While Gelsinger’s comments may appear challenging for NVIDIA, other large tech companies are developing chips to replace NVIDIA’s, which may be used internally or offered as a cloud service. Intel has its own training chip, Gaudi, which recently showcased Gaudi3.
Leave a Comment