Tech | Visa | Scholarship/School | Info Place

Intel Vision 2024 brings a new look to Gaudi 3 artificial intelligence chip

After first announcing the existence of the Gaudi 3 AI accelerator last year, Intel is preparing to deliver the chip to OEMs in the second quarter of 2024. Intel also announced this and other news, including new Xeon 6 branding and an open Ethernet standard for AI. The Intel Vision Conference will be held April 8-9 in Phoenix, Arizona, following the April 1 The workload was introduced at a pre-launch event.

Gaudi 3 AI accelerator will be available to Dell, Hewlett Packard Enterprise, Lenovo and AMD

Gaudi 3 will be launched with Dell, Hewlett Packard Enterprise, Lenovo and Supermicro as OEM partners.

Suppliers will offer Intel Gaudi 3 in three form factors: mezzanine card, universal substrate or PCle CEM. Intel said that compared to NVIDIA’s H100 AI chip, Gaudi 3 trains large language models 40% faster, and LLM’s inference speed is 50% faster compared to NVIDIA H100.

Gaudi 3 may go head-to-head with NVIDIA’s newly released AI acceleration chip Blackwell. Jeff McVeigh, vice president and general manager of Intel’s Software Engineering Group, said Gaudi 3 is “very competitive.” McVeigh noted that these two products are not yet ready for practical testing.

New Xeon 6 brand to launch in Q2

Xeon 6 processors are available in Performance Core and Efficiency Core versions and will be shipping soon. E-core processors will ship in the second quarter of 2024, with P-core processors to follow.

Both variants of the Xeon 6 processor share the same platform foundation and software stack. Performance cores are optimized for compute-intensive and artificial intelligence workloads, while efficiency cores are optimized for efficiency for the same workloads. Compared with its predecessor, the Intel Xeon 6 processor with E-core delivers 2.4 times better performance per watt and 2.7 times better performance per rack.

Xeon 6 processors demonstrate significant energy savings compared to 2nd generation Intel Xeon processors by requiring fewer server racks and reducing power consumption by up to 1 megawatt.

Network interface card supports open internet standards for artificial intelligence workloads

As part of Intel’s efforts to provide broad artificial intelligence infrastructure, the company announced the availability of artificial intelligence network interface cards for Intel Ethernet Adapters and Intel IPUs. Intel said the AI ​​network interface cards already used by Google Cloud will provide a secure way to offload functions such as storage, network and container management and manage AI infrastructure. The aim is to be able to train and run inference on increasingly larger generative AI models, which Intel predicts organizations will want to deploy across Ethereum.

Intel is working with the Ultra Ethernet Alliance to create open standards for artificial intelligence networking over Ethernet.

The AI ​​network interface card is expected to be available in 2026.

Broad scalable systems strategy aims to boost AI adoption

To prepare for the future of artificial intelligence the company predicts, Intel plans to launch an enterprise scalable systems strategy.

“We want it to be open and allow enterprises to have choices in hardware, software and applications,” Sachin Katti, senior vice president and general manager of Intel’s Network and Edge Group, said at a pre-release event.

To this end, the Scalable Systems Strategy provides Intel products for all areas of AI within the enterprise: hardware, software, frameworks and tools. Intel is working with a variety of partners to make this strategy a reality, including:

  • Google Cloud.
  • Thales.
  • Cohesion.
  • Nahuel.
  • Bosch.
  • Ola/Kutrim.
  • Nielsen IQ.
  • seeker.
  • Identification of friend or foe.
  • CtrlS group.
  • Login to Artificial Intelligence.
  • Robot flow.

Intel predicts the future of AI agents and AI capabilities

Carty said in a pre-meeting briefing that companies are in the era of artificial intelligence co-pilots. Next may come the age of AI agents, which can coordinate other AIs to perform tasks autonomously, and then the age of AI functions. Sachin said the rise of AI capabilities could mean a group of agents taking on the work of an entire department.

See: Articul8, maker of a generative artificial intelligence software platform, Spun off from Intel in January. (Technology Republic)

Intel’s competitors

Intel is trying to differentiate itself from its competitors by focusing on interoperability in an open ecosystem. Intel’s competition in the field of artificial intelligence chips includes:

  • NVIDIA released the next generation Blackwell chip in March 2024.
  • AMD announced in February 2024 the launch of a new AI inference architecture solution based on AMD Ryzen embedded processors.

Intel competes with TSMC, Samsung, IBM, Micron Technology, Qualcomm and other companies for dominance in chip manufacturing.

TechRepublic is covering Intel Vision remotely.

#Intel #Vision #brings #Gaudi #artificial #intelligence #chip

Leave a Reply

Your email address will not be published. Required fields are marked *