Intel released a range of 50 “Purley” processors with the formal name of Intel Xeon Scalable processors based on its 7th generation technology.
It replaces the ageing E5 and E7 CPUs built on its 4th generation Haswell 22nm wafer with four to 18 cores. Xeon Scalable comes in Bronze, Silver, Gold and Platinum specifications from eight to 28 cores in one to eight-socket configurations supporting 1.5 to 12TB of memory.
Intel claims that the processors are capable of up to 5x the performance of an E7 Xeon, 2.2x the performance on AI/deep learning and up to 65% lower total cost of ownership. Of course, these are claims and real-world use in the data centre will be the subject of many more articles.
Xeon Scalable allows for up to 28 cores to run on a single LGA3647 motherboard socket, with support for up to 8 sockets to work together on the same computer. The inter-core connectivity lends itself to computationally intensive tasks, including artificial intelligence, simulations, encryption, and other high-demand applications.
{loadposition ray}
According to Dow Jones Market Data Intel had 99.7% of the global server CPU market at the end of 2016. The minuscule percentage left over was a mix of Nvidia ARM, AMD, and a few other specific CPUs.
By all that has changed with AMD’s stellar resurgence on the back of its Ryzen (consumer) and EPYC (server) CPUs that until today blitzed the older E5 and E7 Xeons.
Intel revealed that it had already sold more than 500,000 of the pre-production chips to 30 major corporations, cloud and communications companies. The Xeon launch was supported by some of its data centre customers – Alphabet/Google and Amazon Web Services. Google said Xeon Scalable CPU had sped up a variety of workloads, bringing performance improvements of 40-100% or more.
Bernstein Research analyst Stacy Rasgon said, “The data centre is the most competitive environment we have seen in a decade. Intel is trying to show that they still have leadership and that they can still have a great business. Part of this is because there is so much hype around the competing products, they want a chance to tell their story.”
Dean McCarron, principal analyst at Mercury Research, said, “The server market moves more slowly than consumer markets, and it will take time before any AMD sales take hold and impact Intel’s dominant server market share.”
“Changes in server happen very slowly at first, the design and qualification cycles are long, so a new product on the market takes a while - often a couple years - to ramp to volume. Without forecasting the share in any way, any changes will be modest simply due to how the market works. Changes made today show up in share results 18 months to two years from now,” he added.
What about AMD’s EPYC assault?
AMD EPYC can have as many as 32 cores per CPU versus Intel Platinum’s 28. Intel says it can get the same or better performance from fewer cores and besides a lot of software licencing is per core so it will lower licence costs.
AMDs official response was that it was pleased to return competition to the data centre and just as Intel had customers at its launch so too did AMD.
It says the ecosystem stands behind EPYC: HPE, Dell, Asus, Gigabyte, Inventec, Lenovo, Sugon, Supermicro, Tyan, and Wistron announced EPYC products. Primary hypervisor and server operating system providers Microsoft, Red Hat, and VMware showcased optimised support for EPYC at launch, and key server hardware ecosystem partners Mellanox, Samsung Electronics, and Xilinx were also there.
Its main product, the EPYC 7000 completely disrupts the existing dynamic of the one-socket and two-socket servers and is specifically designed as a balanced architecture, perfect for scale-out cloud computing, storage, data analytics and the enterprise.