Intel formally announced the Xeon 6 line of server chips featuring two types of cores and a variety of configurations supporting multiple use cases. But the main emphasis is on AI.
In a first for the Xeon line, the processor comes with two microarchitectures: an efficient core (E-core) and a performance core (P-core). Performance cores are larger, more high-performance, and more complex, making them appropriate for high-end compute. Efficiency cores are smaller, with fewer instructions and slower clock speed for tasks that don’t require performance-level horsepower. The different processors will mix and match P cores and E cores, with one E-core design sporting up to 288 cores.
The different cores will apply to different use cases, but the overarching scenario is, of course, AI. “We do see every company becoming an AI company,” Matt Langman, Intel vice president and general manager, said on a conference call. “Whether these companies are looking to be more efficient in their operations, more efficient in their product development, more efficient with their customer engagements, or just more effective with their customer engagements.”
For now, Intel is kicking off the launch of the Xeon 6 with a pair of processor families: the Xeon 6 6700 Series and the 6900 Series. The 6700 Series features up to 144 Efficient-cores or 86 Performance-cores supporting 8 channel memory, up to 6400 MT/s DDR5 memory, 8000 MT/s MCR DIMM memory, and up to 88 lanes of PCIe 5.0/CXL 2.0 and 4 UPI 2.0 links. It has a power draw of 350 W per chip.
The 6900 Series sports up to 288 Efficient-cores or 128 Performance-cores, 12 memory channels of 6400 MT/s DDR5 memory, 8800 MT/s MCR DIMM memory, up to 96 lanes PCIe 5.0/CXL 2.0 and six UPI 2.0 links. It runs at a toasty 500 W.
Intel’s strategy behind selling the E-core is to promote server consolidation; the company is hyping the potential of 3-to-1 rack-level consolidation, up to 4.2x rack-level performance gains, and up to 2.6x performance per watt compared to its second-generation Intel Xeon processors on media transcode workloads.
“Let’s say you take approximately 200 racks, which would be a typical mid-sized data center deployment, looking at about 15 kilowatt racks and 22U servers … You look at the advancements that we’re delivering an Intel Xeon 6 with the perf per watt and overall perf performance improvement, and you get that three to one rack consolidation down to 66 racks. Huge, amazing savings,” said Langman.
He went on to predict and energy usage savings of upwards of 84,000-megawatt hours over a period of four years.
Gaudi pricing disclosed
In addition to the Xeon 6 news, Intel announced pricing for two of its latest Gaudi AI accelerators, which are designed to compete against Nvidia’s H100 and AMD’s Instinct MI300X in processing large language models.
Intel announced the pricing in terms of a cluster of eight accelerators, which AMD and Nvidia also do. Customers using a standard AI kit with eight Intel Gaudi 2 accelerators with a universal baseboard will pay $65,000, while a kit with eight Intel Gaudi 3 accelerators will be $125,000.
Gaudi is on track to ship in Q3 2024. Intel has numerous OEM partners lined up, including Dell, Hewlett-Packard Enterprise, Lenovo, Supermicro and new partners Asus, Foxconn, Gigabyte, Inventec, Quanta and Wistron.
Source:: Network World