Today, Intel is set to announce a number of future data center products, including new CPUs, networking, and AI products. We are going to cover the announcements live from the show so please excuse typos. We will follow up this live coverage with additional coverage later this week.
As a quick note: I am attending the event as an “Influencer.” The press was not invited, but analysts were. The event is being held about 25 min from the STH studio in Scottsdale, so that is how I got a badge.
Intel Vision 2024 Keynote Live Coverage
Pat Gelsinger is starting the keynote off by saying that every company will be an AI company.
AI is a big theme, and we will have AI PCs everywhere. Intel says we will have over 5 million AI PCs shipped, aiming for 40M this year and over 100M next year.
Lunar Lake is the second-generation AI part and is here today.
Next, we are talking about enterprise and edge AI. 50% of edge deployments are expected to use AI by 2026.
Intel is focused on Retrieval Augmented Generation or RAG to take data companies have today and combine it with real-time data and LLM AI models.
Intel had a RAG demo using Intel Xeon 6 and Gaudi 2 accelerators where Llama2 70b did not have access to real-time data, and the other did with RAG. Note, Intel is saying “Xeon 6” during the demo. Intel is also talking about an Open Platform for Enterprise AI that it will be rolling out next month.
Intel is working with Accenture. Big consulting firms are going to be key to getting AI into enterprises. Companies like Accenture have large IT outsourcing arms. I was on projects when I did management consulting at PwC where Accenture was brought in to do the hands-on ERP implementation and so forth. NVIDIA also knows that these types of firms are imperative to get AI adoption in enterprises, especially those outside of the tech space.
Now we are getting to Intel Xeon 6 based on Intel 3 process. Intel Xeon 6 is the new branding for Intel’s processors. In this, apparently Xeon 1 would have been Skylake or 1st Gen Intel Xeon Scalable.
Sierra Forest is the first volume part on Intel 3.
Granite Rapids will follow Sierra Forest and be the next P-core CPU.
Granite Rapids is on stage. I wonder if Intel has room to take this system back? If not, I have the Cybertruck across the street and it will fit in the bed.
Intel Trust Domain Extensions (Intel TDX) are coming out. Intel said Google Cloud is previewing its next-gen confidential computing instances. Google has an event going on now in Las Vegas. Other service providers also have TDX, but Google was highlighted.
Intel is talking about Gaudi 2 and the Intel Developer Cloud.
A few months ago we had a tour of the Intel Developer Cloud with Gaudi 2.
Intel says that it will support UltraEthernet Consortium in future NICs, future chiplets, hard and soft IP, and more. This is Intel pushing against NVLink and Infiniband using a more open standard for its AI clusters. Intel’s Gaudi line runs on Ethernet, so this makes a lot of sense.
Supermicro is the AI server company. Ray is on stage with Pat, talking about its Gaudi 1 cluster and talking about its Gaudi 2 cluster that is being deployed today.
Supermicro sells a ton of NVIDIA GPUs, Intel Gaudi, and AMD MI300X systems. Supermicro says it has a Gaudi 3 system here, it is real and not just a slide.
Intel Gaudi 3 is set to be announced today. We expect more than just Supermicro announcing systems, but Supermicro and Wiwynn have been at the forefront of Gaudi systems to date.
Intel is showing how it is using LLMs with RAG for its manufacturing via a demo running a few minutes away at the company’s Arizona fabs.
This demo was being conducted on an AI PC. Intel says that using these AI tools in its factories is saving millions already.
Intel has over 5000 engineers from Arizona State University which is a few minutes away from here. The famous analyst, Ian Cutress, went to dinner with my wife and me near ASU this weekend. ASU is training on AI right now.
ASU says that its AI program has helped overcome limits of teaching and is part of the reason it has roughly tripled its engineering graduates over the last few years.
Bosch is a giant 400,000-person firm that is huge in Europe, and they are on stage talking about Intel and AI. The company is talking about how AI is being brought to products and manufacturing including optical inspection.
Andrew Ng’s recorded message goes into why LLMs can be trained and used on business data because the text is similar. On the vision model side, proprietary images may be very different than general images. Landing AI is helping companies do this using the Intel Dev Cloud.
Naver started as a search engine in Korea. It has grown a ton and delivers a number of services and also a cloud. I actually visited a Naver data center in 2019. The company is using Gaudi and Gaudi 2.
Michael Dell is doing a cameo.
We now have an Intel Gaudi 3 announcement for large scale AI computing. Gaudi 3 will start shipping this quarter.
Intel will sell the HL-325L OAM compliant Mezzanine card, a HLB-325 Universal Baseboard (UBB), and a HL-338 as a 600W PCIe CEM double-width add-in card.
We will deep dive into Gaudi 3 soon.
Final Words
Again, we will be covering a few of the announcements in more detail later this week. Also, stay tuned for more reviews of the products that we saw today. We already have a number of reviews planned.
“Intel’s Gaudi line runs on Ethernet, so this makes a lot of sense.”
Take a look at the Networking section in the Gaudi 3 White Paper. Are the RoCE V2 extensions they describe simply implementing what is coming in the Ultra Ethernet NICs?