Intel Xeon 6 Chips: Revolutionizing Data Centers for Enterprise Artificial Intelligence

Intel introduced its Xeon 6 chip to assist knowledge facilities deal with the workloads they want as extra corporations start to deploy synthetic intelligence functions and fashions. The processor comes with two microarchitectures: an environment friendly core (E-core) and a efficiency core (P-core) and can be out there on the 6700 and 6900 platforms. Nonetheless, solely the 6700 E-core model will launch on June 4. The 6900 P-core model will debut within the third quarter of 2024, whereas the others can be out there within the first quarter of 2025.

“It is all about empowering business prospects to drive actual enterprise outcomes,” Intel vice chairman and basic supervisor Matt Langman mentioned on the press convention. “We see each firm changing into an AI firm, whether or not these corporations wish to be extra environment friendly of their operations, extra environment friendly in product growth, extra environment friendly—or simply extra environment friendly—in interacting with prospects.”

Along with the Xeon 6 processor, Intel is revealing for the primary time the costs of its accelerator chips Gaudi 2 and Gaudi 3. The usual AI package with Gaudi 2, launched in 2022, will value $65,000, whereas Gaudi 3, which was launched in April, will value $125,000.

Xeon 6: What it’s worthwhile to know

Intel calls the Xeon 6 a “strong computing platform” that “exceeds each efficiency and effectivity,” two areas it says are vital to “assembly ever-increasing knowledge middle calls for.” The processor helps a variety of use circumstances, from intensive synthetic intelligence and high-performance computing to conventional enterprise and high-density energy-efficient functions.

Corporations are set to improve their legacy knowledge middle techniques with Xeon 6 chips, which guarantees to extend value financial savings, assist meet sustainability objectives, optimize bodily ground and rack area, and create new digital capabilities.

The Xeon 6 P-core and E-core, previously codenamed Granite Rapids and Sierra Forest, share a typical {hardware} platform and software program stack. They’re outfitted to assist elevated core counts, have excessive reminiscence bandwidth with DDR5, have multiplexed DIMMs of mixed rank, elevated inter-socket bandwidth with UPI 2.0, Compute Specific Hyperlink 2.0, Frequent OS and firmware and extra.

Citing the completely different necessities that knowledge facilities might have, Intel presents two completely different microarchitectures for its Xeon 6. It comes all the way down to workloads and capabilities. The P core is prone to be extra appropriate for these working with high-performance computing and intensive synthetic intelligence. Nonetheless, it is also helpful for “the widest vary of enterprise functions.” Langman explains, “It is actually a few of these latency-sensitive workloads that the advantages of P-core will assist obtain or exceed these excessive single-threaded, excessive efficiency per-core capabilities.” Nonetheless, workloads that embody cloud Web, microservices, and digital providers might profit from the Xeon 6 E-core, “profiting from greater density and improved efficiency per watt.”

So what are the variations between the Xeon 6 6700 Collection and the 6900 Collection?

Xeon 6 processors (6700 collection)

• As much as 144 environment friendly cores / 86 efficiency cores
• Socket assist: Help 1S/2S and 4S/8S (P-core solely).
• Max. TDP: As much as 350 watts per processor
• Mem Channels: 8-channel reminiscence, DDR5 reminiscence as much as 6400 MT/s, MCR DIMM reminiscence 8000 MT/s (P-core)
• PCIe/CXL: As much as 88 PCIe 5.0/CXL 2.0 lanes
• UPI Hyperlinks: 4 UPI 2.0 hyperlinks, as much as 24 GT/s

Xeon 6 processors (6900 collection)

• As much as 288 environment friendly cores / 128 efficiency cores
• Socket assist: Help 1S/2S
• Max. TDP: As much as 500 watts per processor
• Mem Channels: 12-channel reminiscence, DDR5 reminiscence as much as 6400 MT/s, MCR DIMM reminiscence 8800 MT/s (P-core)
• PCIe/CXL: As much as 96 PCIe 5.0/CXL 2.0 lanes
• UPI Hyperlinks: 6 UPI 2.0 channels, as much as 24 GT/s

What a couple of phased launch?

Not all Xeon 6 variants can be out there at the moment. Intel says it is intentionally slowing down numerous CPU SKUs as a result of that is apparently what prospects need. “What made essentially the most sense for the assorted prospects we serve with Xeon 6 is the staggered order that you just see. And what’s actually thrilling is as we get nearer to the launch of the primary one, it is in nice form and we’re getting loads of buzz within the market round it, and it is thrilling to look at that occur,” Ryan Tabra, vice chairman of Intel. and basic supervisor of the E-core product line, clarifies.

Xeon 6 E-core has bragging rights

Because the Xeon 6 E-core is the primary debut—and the primary model of the Xeon with an E-core—Intel has supplied some benchmarks, saying it permits 3-to-1 rack consolidation, delivering rack-level efficiency. a rise of as much as 4.2x and gives a efficiency achieve of as much as 2.6 per watt in comparison with second technology Intel Xeon processors for media transcoding workloads.

“If we take a look at at the moment’s knowledge middle, and for instance you are taking about 200 racks, which might be a typical mid-sized knowledge middle deployment, taking a look at about 15 kilowatt racks and 22 servers, and also you take a look at how they ship media streams per second on second-generation Intel Xeon. You take a look at the progress we’re making on Intel Xeon 6 with efficiency per watt and general efficiency enhancements, and also you get this 3-to-1 rack consolidation to 66 racks—an enormous, wonderful financial savings in rack area and capabilities. ” Langman says.

“However… it is not nearly saving racks, it is also about saving vitality and carbon footprint. And what we’re seeing, with the park’s vitality use, is a financial savings of over 84,000 megawatt hours over 4 years — a major vitality financial savings — coupled with a 34,000 metric ton discount in carbon emissions … over the identical time interval. On this means, we’re reaping the advantages of shelf consolidation and reaching the business’s sustainability objectives.”

Gaudi 3: costs and new system suppliers

Along with the Xeon 6 information, Intel has launched pricing for its two newest Gaudi merchandise. Designed to compete with Nvidia’s H100, these accelerator chips assist practice and infer massive language fashions, however at a decrease general working value.

Corporations utilizing the usual eight-accelerator Intel Gaudi 2 AI package with a common mainboard can pay $65,000, which Intel estimates is one-third the price of comparable alternate options. An analogous package with eight Intel Gaudi 3 accelerators will value $125,000, or about two-thirds the worth of Intel’s rivals.

Whereas Intel did reveal pricing, it did not present any extra particulars on Gaudi 3 availability, aside from that the chip is slated for launch within the third quarter of 2024.

To assist with its go-to-market technique, the corporate is working with no less than ten system distributors, together with Dell, Hewlett-Packard Enterprise, Lenovo, Supermicro, and new additions Asus, Foxconn, Gigabyte, Inventec, Quanta, and Wistron.

Synthetic intelligence utilization examples for Xeon and Gaudi

When requested how Xeon and Gaudi will work collectively, Tabrah describes the connection as complementary. “It frees up prospects’ knowledge facilities,” he says. “They’re going through these energy bottlenecks. The information middle itself nearly turns into a bottleneck. And if you happen to can untangle that knowledge middle for you, simply by shortly and simply shifting to a really environment friendly shared computing structure and never touching something, then you’ll be able to simply innovate the remainder of the infrastructure to do extra AI; it is simply great.”

Anil Nanduri, one other of Intel’s vice presidents, factors out that the entire level is to assist the enterprise unlock its knowledge. With the introduction of generative AI, organizations are searching for higher methods to make use of massive language fashions to construct options which are proper for the workloads they want and finest match for the compute prices they’re snug investing in. “You may see the place accelerators will nonetheless run these [LLMs], however a use case just like RAG the place I can maintain my datasets extra updated and get higher outcomes for shoppers doing data discovery or different use circumstances, retailer their vector plugins on the Xeon after which plug it into the mannequin is engaged on an accelerator,” he shares. “It is an excellent use case the place we’ll anticipate loads of good compatibility, efficiency and worth that Xeon and Gaudi can deliver.”

“Intel is without doubt one of the few corporations on this planet that’s innovating throughout the total spectrum of AI market alternatives, from semiconductor manufacturing to PCs, networking, edge techniques and knowledge facilities,” Intel CEO Pat Gelsinger mentioned in an announcement. “Our newest Xeon, Gaudi and Core Extremely platforms, mixed with the facility of our {hardware} and software program ecosystem, ship the versatile, safe, sustainable and cost-effective options our prospects want to maximise the large alternatives forward.”