Intel has launched a new AI chip, directly challenging Nvidia.
Pat Gelsinger, CEO of Intel, announced the release of the Core Ultra processor, the fifth-generation Xeon Scalable processor, and the third-generation AI accelerator Gaudi 3 at an event in New York on December 15, as reported by the media.
Gelsinger criticized Nvidia's CUDA (a computing platform developed by Nvidia), describing its technological moat as shallow and narrow. He emphasized the growing importance of inference technology over training in artificial intelligence.
In an interview, Gelsinger suggested that Nvidia's dominance with CUDA won't last forever. He pointed out that the entire industry is motivated to eliminate CUDA's market hold. Gelsinger mentioned MLIR, Google, and OpenAI as examples, hinting at their shift towards a "Python-based programming layer" to make AI training more open.
Gelsinger added that the industry is driven to bring broader technology to a wide range of training, innovation, and data science.
However, Intel's focus isn't solely on training; instead, they believe inference is the key. Gelsinger stated that embracing inference means once a model is trained, there's no longer a dependency on CUDA. While Intel will still compete in the training sector, fundamentally, the inference market is their primary competitive focus, with Gaudi 3 being Intel's product to penetrate the AI inference market.
Gelsinger also took the opportunity to promote OpenVINO, a standard specifically developed by Intel for its AI work. He predicted the emergence of a "hybrid computing" world, where some computations occur in the cloud and others on personal computers.
