By Max A. Cherney and Stephen Nellis
TAIPEI/SAN FRANCISCO (Reuters) – Intel launched its next generation Xeon server processors on Tuesday in a bid to regain data center market share, and revealed that its Gaudi 3 artificial intelligence accelerator chips would be priced much lower than its rivals’ products.
The sixth generation Xeon chips are crucial for Intel, which has been steadily losing data center market share to Advanced Micro Devices (AMD).
Intel’s share of the data center market for x86 chips has declined 5.6 percentage points over the past year to 76.4%, with AMD now holding 23.6%, according to data from Mercury Research.
Stumbles in Intel’s manufacturing process have allowed AMD to take business, as the latter uses Taiwan Semiconductor Manufacturing Co (TSMC) to fabricate its chips.
The Xeon 6 server processors come in two main flavors, a larger, more powerful version, and an “efficiency” model that Intel pitched as a replacement for older-generation chips.
To achieve the same level of computing power as its second generation chips, they will now require about 67% fewer server racks with the efficiency model, which is designed to serve media, websites and perform database calculations.
“Simply put, performance up, power down,” Intel Chief Executive Pat Gelsinger said at the Computex trade fair in Taipei, where he gave a presentation of the server.
The more powerful performance model is designed to run the computations necessary to generate responses from complex AI models and other tasks that need the increased horsepower.
The “efficiency” Xeon chip is available on Tuesday, with the “performance” model arriving in the third quarter. Intel plans to launch additional variations next year. The sixth generation chip was delayed a year because the company wanted to use a different manufacturing process.
On a briefing call with reporters, Intel said that a Gaudi 3 accelerator kit, which includes eight of the AI chips, sells for about $125,000, and the earlier generation Gaudi 2 has a list price of $65,000.
Speaking in Taipei, Gelsinger said the prices looked “pretty compelling”, especially compared with competitors.
“In other words, it crushes the competition.”
AMD and Nvidia do not discuss pricing of their chips. A comparable HGX server system with eight Nvidia H100 AI chips can cost more than $300,000, according to custom server vendor Thinkmate.
Intel revealed the details of the Gaudi 3 AI chip in April and has positioned it as a considerably cheaper and viable alternative to Nvidia’s H100 chips.
Also on Tuesday, Intel said its next generation laptop chip, called Lunar Lake, uses 40% less power and has more a powerful AI processor in it. Intel said it will ship the chip in the third quarter.
Gelsinger said he was “just thrilled” at seeing so many new PCs using Lunar Lake.
“It will power the largest number of next gen AI PCs in the industry,” he added. “We are committed to the AI PC.”
He also thanked TSMC, which he said was critical to helping with many of the core technologies for Lunar Lake.
Looking ahead at more advanced versions to come, Gelsinger said Arrow Lake would be introduced next year, followed in 2025 by Panther Lake.
( This story has been refiled to add Reuters in the dateline)
(Reporting by Max Cherney in Taipei and Stephen Nellis in San Francisco; Additional reporting by Ben Blanchard; Editing by Sonali Paul and Rashmi Aich)
Source Agencies
