The "big shots" speech at the Computex in Taipei, China is still ongoing, and today is Intel's "home".
After senior executives from NVIDIA, AMD, and Qualcomm successively launched a series of "big moves", Intel CEO Pat Gelsinger also gave a brilliant and unparalleled speech.
Gelsinger opened by saying, "Artificial intelligence is driving one of the most important innovation eras in the history of the chip industry, and the magic of silicon is once again making exponential progress in computing technology, which will push the boundaries of human potential and provide power for the global economy in the coming years."
Gelsinger also gave a grand introduction to some new products, including the release of the sixth generation Xeon processor, emphasizing the cost-effectiveness of Gaudi AI accelerator, and announcing the details of the new generation Lunar Lake processor architecture. He expects these new products to help Intel regain some market share held by peers such as Nvidia and AMD.
6th generation Xeon processor
At the meeting, Gelsinger showcased his new sixth generation Xeon 6 data center processor, which has a more efficient kernel and significantly reduced size.
There are two main versions of the Xeon 6 processor, one is the more powerful version (P-series); Another version is the "efficient" version (E-series).
The efficient version of the Xeon 6 processor with an E-core (Efficiency Core) is mainly aimed at high-density, scalable loads, etc., and aims to provide services for media, website, and database computing. This version has a huge density advantage, providing customers with a rack level performance gain of up to 4.2 times and a performance gain of up to 2.6 times per watt. The server rack size can be reduced by one-third.
Gelsinger introduced, "Simply put, it means improving performance while reducing power consumption.".
The high-performance version with a Performance Core is designed to run necessary calculations and is equipped with up to 128 performance cores for computationally intensive and artificial intelligence workloads.
The Intel Xeon 6 E core (codenamed Sierra Forest) will be launched this Tuesday, and the Xeon 6 P core (codenamed Granite Rapids) will be launched in the third quarter of this year. In addition, Intel plans to release more versions next year.
Gaudi AI Accelerator with Ultra High Cost Performance
When it comes to the artificial intelligence accelerator Gaudi, Kissinger emphasized that the Gaudi 3 has excellent cost-effectiveness while ensuring efficiency.
Gaudi 3 was first unveiled at Intel's Vision 2024 conference in April this year, specifically for handling generative artificial intelligence training. Each Gaudi 3 cluster consists of 8192 accelerators, and Intel estimates that it trains artificial intelligence models 40% faster than similarly sized Nvidia H100 GPU clusters.
Cost performance is Gaudi's biggest advantage. A kit containing 8 Intel Gaudi 2 accelerators is priced at $65000; The package, which includes 8 Gaudi 3 accelerators, is priced at $125000. In his speech, Gelsinger bluntly stated that its price looks very attractive, "in other words, it has surpassed its competitors."
AMD and Nvidia have not disclosed the pricing of their respective AI chips. However, according to custom server supplier Thinkmate, the cost of a similar HGX server system equipped with 8 Nvidia H100 AI chips may exceed $300000.
Intel also stated that through testing, it has been found that the Gaudi 3 will be twice as fast as Nvidia's H100 in executing artificial intelligence inference tasks.
However, some analysts suggest that these advantages may not be enough to overturn Nvidia's leading position in the field of artificial intelligence processing.
NEXt Curve analyst Leonard Lee pointed out that the performance of a single accelerator is no longer the most important thing. Nvidia's biggest advantage is that it has a cohesive integrated ecosystem and proprietary technology to ensure its computing cluster operates as a whole.
Notebook processor: Lunae Lake
On Tuesday, G ü ersinger also announced Intel's next-generation laptop processor, the Lunar Lake, with a 40% reduction in power consumption and more powerful performance. Intel stated that the chip will officially debut in the third quarter of this year.
Kissinger pointed out that the Lunar Lake laptop processor will be used in the next generation of AI PCs (artificial intelligence personal computers), with significant improvements in both drawing and AI processing, three times higher than the AI performance of the previous generation product.
One of the highlights is that Intel directly integrates two layers of LPDDR5X-8500 memory into the chip packaging, configured as 16GB or 32GB. This design change can reduce the data transmission load by approximately 40%, significantly improving data processing efficiency and speed.
Intel mentioned that for customers with more memory needs, the company will announce the Arrow Lake later this year to meet such customer needs. And it is expected to be upgraded to Panther Lake processors next year.
At the same time, Intel has officially launched an AI PC development kit, equipped with a Lunar Lake processor, with 32GB of LPDDR5 memory and 512GB of storage. The overall design is a square box with dimensions of 165x150x27mm, supporting Wi Fi 7 and Bluetooth 5.1, equipped with 2 x USB-C, 2 x USB-A, HDMI, and wired network ports.
Gelsinger pointed out that the Lunar Lake processor will provide support for over 80 different AI PC designs from 20 original equipment manufacturers. Later this year, a batch of laptops equipped with Lunar Lake will be launched.
In the AI PC race, Intel is in a leading position. While other companies were preparing to enter the AI PC market, Intel had already begun large-scale shipments, and data showed that by the first quarter of 2024, Intel had delivered more AI PC processors than all competitors combined.
Gelsinger emphasized that as a leading supplier of personal computer chips, Intel will play an important role in the AI boom. In addition, he reiterated the importance of partnerships for Intel.