Taking on NVIDIA and AMD head-on! Intel launches the new generation AI chip Xeon 6

Author: Zhao Yuhe

Source: Hard AI

The AI chip battle is in full swing. After NVIDIA and AMD successively launched new generation chips at COMPUTEX Taipei International Computer Show in Taiwan, China, Intel, unwilling to fall behind, also released a new artificial intelligence chip Xeon 6 for data centers on Tuesday to challenge its competitors NVIDIA and AMD.

Intel CEO Pat Gelsinger stated that the Xeon 6 processor will provide better performance and energy efficiency in high-intensity data center workloads compared to its predecessors, allowing operators to reduce the required space for a given task to one-third of the previous generation hardware.

NVIDIA and AMD had previously launched new AI chips on Sunday and Monday respectively, vying for the leading position in the thriving AI industry. Intel had released its 5th generation Intel Xeon processor for data center workloads six months ago and Gaudi 3 processor for AI model training and deployment two months ago.

Intel also revealed on Tuesday that the prices of Gaudi 2 and Gaudi 3 AI accelerators are lower than those of competitors' chips. Gelsinger stated:

"Customers are looking for high-performance, cost-effective general AI training and inference solutions. They are starting to turn to alternatives like Gaudi. They want more choices, open software and hardware solutions, and solutions that can bring products to market quickly with significantly lower TCO."

Gelsinger mentioned that Intel's Gaudi system will compile its chips into multi-processor suites specifically for handling generative AI training, which will be provided by partners such as Dell and Inventec. A suite with eight Intel Gaudi 2 accelerators is priced at $65,000. A more powerful suite with eight Intel Gaudi 3 accelerators is priced at $125,000. The company estimates that the prices of these two products are more affordable than those of competitors.

Each Gaudi 3 cluster consists of 8192 accelerators. Intel estimates that compared to an equivalent-sized NVIDIA H100 GPU cluster, Gaudi 3 can speed up AI model training by 40%. Intel also stated that when performing AI inference tasks, the speed of Gaudi 3 will be twice that of NVIDIA H100. However, analysts point out that these advantages may not be enough to displace NVIDIA from its leading position in the data center AI processing field At the same time, Intel also revealed details of its upcoming Lunar Lake processor architecture, aiming to "continue to drive the growth of the AI PC category." The Lunar Lake chip, expected to ship in the third quarter, will compete with NVIDIA and AMD chips designed specifically for AI PCs.

Analysts believe that in the AI frenzy, Intel has been on the edge while NVIDIA and AMD are significantly ahead. Tech giants like Meta, Microsoft, and Google have also purchased NVIDIA chips as much as possible.

It is reported that NVIDIA and AMD each year develop new data center chip roadmaps. NVIDIA released a new "Rubin" chip on Sunday to replace the "Blackwell" model announced in March. AMD detailed the schedule for releasing new Instinct accelerators annually until 2026.

Unlike chip designers Intel and AMD, NVIDIA not only designs chips but also manufactures them. However, its foundry business has been struggling, with a $7 billion increase in operating losses in 2023 compared to the previous year.

In terms of chip manufacturing, NVIDIA has lost its advantage to competitors like TSMC, but the nearly $20 billion in funding provided by the Biden administration's CHIPS and Science Act may help Intel advance its semiconductor manufacturing and research.

Although Intel's revenue decline has stopped, analysts are not predicting a rapid rebound, with the company expecting revenue in 2024 to be $20 billion less than in 2021. Meanwhile, NVIDIA's sales are expected to double, and AMD's growth rate will exceed 10%, as these companies are better leveraging the wave of AI computing hardware spending.

To counter NVIDIA, Gelsinger directly criticized NVIDIA CEO Jensen Huang's previous statement that "traditional processors like Intel's have gradually lost momentum in the era of artificial intelligence." Huang had previously stated that the era of traditional processors is over and Moore's Law is dead.

Gelsinger emphasized that Intel will play a crucial role as a major supplier of PC chips in the proliferation of artificial intelligence:

"Contrary to what Jensen Huang would have you believe, Moore's Law is still alive and well. I think it's like the internet 25 years ago, it's huge. We believe this will drive the semiconductor industry to reach $1 trillion by 2030."

However, netizens largely disagree with Gelsinger's statement. Some netizens said that Huang's statement is correct,

"Even if Moore's Law is not dead, it is on its last legs."

There are also netizens who said, "Intel once tried to make its competitors go bankrupt through illegal means, but now they have become irrelevant. Goodbye, no need to see them off."

Some people also said,

"It can only be said that some things are still alive for Intel, but for NVIDIA, it is already dead."

After Intel announced its new generation of chips, its stock price fell by 0.46% to $30.15 in Tuesday's intraday trading.

This article is from WeChat Official Account "Hard AI". For more cutting-edge AI information, please click here