After archetypal announcing the beingness of the Gaudi 3 AI accelerator past year, Intel is acceptable to enactment the spot successful the hands of OEMs successful Q2 2024. Intel announced this and different news, including a caller Xeon 6 marque and an unfastened Ethernet modular for AI workloads, astatine a pre-briefing held connected April 1 up of the Intel Vision conference, which is being held April 8-9 successful Phoenix, Arizona.
Gaudi 3 AI accelerator volition vessel to Dell, Hewlett Packard Enterprise, Lenovo and Supermicro
The Gaudi 3 volition motorboat with Dell, Hewlett Packard Enterprise, Lenovo and Supermicro arsenic OEM partners.
Intel Gaudi 3 volition beryllium disposable from vendors successful 3 signifier factors: mezzanine card, cosmopolitan baseboard oregon PCle CEM. Gaudi 3 has 40% faster time-to-train ample connection models versus NVIDIA’s H100 AI spot and is 50% faster inferencing connected LLMs versus the NVIDIA H100, Intel said.
Gaudi 3 whitethorn spell head-to-head with NVIDIA’s recently announced AI accelerator chip, Blackwell. Gaudi 3 is “highly competitive,” said Jeff McVeigh, firm vice president & wide manager of the Software Engineering Group astatine Intel. McVeigh noted that real-world investigating for the 2 products hasn’t been imaginable yet.
New Xeon 6 marque coming successful Q2
Xeon 6 processors, which travel successful the 2 variants of Performance-core and Efficient-core, volition vessel soon. E-core processors volition vessel successful Q2 2024, with P-core processors pursuing soon after.
The 2 variants of Xeon 6 processors stock the aforesaid level instauration and bundle stack. The Performance halfway is optimized for compute-intensive and AI workloads, portion the Efficient-core is optimized for ratio successful the aforesaid workloads. The Intel Xeon 6 processor with the E-core shows 2.4 times show per watt betterment compared to anterior generations and 2.7 times show per rack betterment compared to anterior generations.
The Xeon 6 processor shows marked vigor savings compared to the 2nd Gen Intel Xeon Processor owed to needing less server racks, for up to 1 megawatt of powerfulness reduction.
Network interface paper supports unfastened net modular for AI workloads
As portion of Intel’s effort to supply a wide scope of AI infrastructure, the institution announced an AI web interface paper for Intel Ethernet Network Adapters and Intel IPUs. The AI web interface cards, which are already being utilized by Google Cloud, volition supply a unafraid mode to offload functionality similar storage, networking and instrumentality absorption and negociate AI infrastructure, Intel said. The intent is to beryllium capable to bid and tally inference connected the progressively larger generative AI models Intel predicts organizations volition privation to deploy each implicit Ethernet.
Intel is moving with the Ultra Ethernet Consortium to make an unfastened modular for AI networking implicit Ethernet.
The AI web interface cards are expected to beryllium disposable successful 2026.
Wide-reaching scalable systems strategy aims to creaseless retired AI adoption
In bid to hole for what the institution predicts volition beryllium the aboriginal of AI, Intel plans to rotation retired a scalable systems strategy for enterprise.
“We privation it to beryllium unfastened and for enterprises to person prime successful hardware, bundle and applications,” said Intel Senior Vice President and General Manager of Network and Edge Group Sachin Katti astatine the pre-briefing.
In bid to bash so, the scalable systems strategy provides Intel products for each segments of AI wrong the enterprise: hardware, software, frameworks and tools. Intel is moving with a assortment of partners to marque this strategy a reality, including:
- Google Cloud.
- Thales.
- Cohesity.
- NAVER.
- Bosch.
- Ola/Kutrim.
- NielsenIQ.
- Seekr.
- IFF.
- CtrlS Group.
- Landing AI.
- Roboflow.
Intel predicts a aboriginal of AI agents and AI functions
Katti said successful the pre-brief that endeavor is successful an property of AI copilots. Next mightiness travel an property of AI agents, which tin coordinate different AI to execute tasks autonomously, followed by an property of AI functions. The emergence of AI functions could mean groups of agents taking connected the enactment of an full department, Sachin said.
SEE: Articul8, makers of a generative AI bundle platform, spun retired of Intel successful January. (TechRepublic)
Competitors to Intel
Intel is trying to acceptable itself isolated from competitors by focusing connected interoperability successful the unfastened ecosystem. Intel competes successful the AI spot abstraction with:
- NVIDIA, which announced the next-generation Blackwell spot successful March 2024.
- AMD, which successful February 2024 announced a caller architectural solution for AI inferencing based connected AMD Ryzen Embedded processors.
Intel competes for spot manufacturing dominance with Taiwan Semiconductor Manufacturing Co., Samsung, IBM, Micron Technologies, Qualcomm and others.
TechRepublic is covering Intel Vision remotely.
English (US) ·