Intel CEO Gelsinger Publicly Criticises Nvidia’s CUDA Moat, Says It’s Shallow, Narrow

By Paschal Abu

At Friday’s “Let AI Everywhere” event, Intel officially released its new Core Ultra mobile processors and fifth-generation Xeon Scalable processors.

In addition, Intel CEO Pat Gelsinger also publicly criticised Nvidia, saying that its CUDA technology moat is shallow and small, and claimed that inference technology will be more important than training for artificial intelligence.

In the interview, he hinted that Nvidia CUDA’s dominance won’t last forever. “You know, the entire industry is thinking about how to kill the CUDA market,” he said.

He cited examples such as MLIR, Google and OpenAI, saying they are moving to “Pythonic” to make AI training more open.

He also mentioned: “We believe that the CUDA moat is shallow and narrow. Because the entire industry is committed to bringing a broader range of technologies to a wide range of training, innovation, data science, etc.”

Kissinger believes that AI cannot rely solely on training, and reasoning is the truly feasible way. “When you embrace inference, once you train the model you no longer need to rely on CUDA,” he said. “The question is, can you run the model?”

It’s not that Intel won’t compete in training, he said, but “fundamentally, the inference market is where the competition is.”

Gelsinger also took the opportunity to promote OpenVINO, Intel’s standard developed specifically for its AI work, and predicted a world of “hybrid computing,” where some computing exists in the cloud and some exist on-premises. on PC.

READ ALSO | Uncertified Apple Watch Chargers Will Reduce Battery Life – Apple

He also announced, “We will compete for 100% of the data centre AI TAM in three ways,” namely “our leadership CEO, leadership accelerator, and foundry.” “We will pursue every internal opportunity: TPU, inference engine, training chips, etc. We will also pursue every business opportunity, partnering with NVIDIA, AMD, etc. We will become a foundry.”

During the press conference, Intel brought a new generation of Core Ultra mobile processors.

According to officials, Core Ultra is the first processor based on Intel 4 process technology and represents Intel’s most significant architectural change in 40 years.

Intel said that the Intel Core Ultra processor uses Intel’s first on-chip AI accelerator for clients, the “Neural Network Processing Unit (NPU)”, which takes energy-efficient AI acceleration to a new level and brings 2.5 times the performance of the previous generation. Energy efficiency performance.

Intel says Core Ultra processors will bring AI features to more than 230 models from laptop and PC makers around the world next year. By 2028, AI PCs will account for 80% of the PC market and bring new tools to our work, learning, and creation.

Intel’s first Core Ultra series processors include the U and H series, with a base power consumption of 15W and 28W respectively. The U series has 2+8+2 core specifications, with up to 4Xe core display. The H series has up to 6+8+2 core specifications and up to 8Xe core display. In the first quarter of next year, Intel will also release the Ultra 9 flagship model with a basic power consumption of 45W, 6+8+2 core specifications, and an 8Xe core display.