Intel’s CEO raised the curtain on a new “Sierra Forrest” Xeon Scalable processor, the name of which has yet to be determined. The “Emerald Rapids” part is a fifth-generation model with 288 efficient or E-cores and 288 threads, a clear attempt to take advantage of the cloud-native workloads that hyperscalers are so fond of.
The company’s press release states that the component will deliver 2.5 times better rack density and 2.5 times higher performance per watt than the fourth-generation Intel Xeon. And that includes the 288-core model.
AMD (128 cores), Amazon (64 cores), Ampere (192 cores) and others are also focusing on single-core and single-thread products but have lower core counts than Intel’s new product.
Paul Alcorn from Tomshardware estimates that Intel is using two chiplets, each containing 144 E-cores, but speculates that Intel could add a third chiplet bringing the total core count to a staggering 432 cores. By comparison, the Xeon Phi 7295, Intel’s latest attempt at manycore products, topped out at 72 cores and 288 threads.
Such a large amount of cores tightly packed together generates a number of problems: power consumption and dissipation, memory bandwidth, cache coherence, clock speed etc. What we do know is that this new Xeon is built using the new Intel 3 manufacturing process alongside that of Intel 7.
The cloud computing market is moving towards a range of products that offer a combination of more powerful and less powerful cores. Arm brought this paradigm by coining the big.LITTLE concept more than a decade ago, and only recently has this paradigm moved into data centers and desktops due to growing concerns about power consumption and high-density computing.
AMD and Intel have used different ways to target the same audience. Intel opted for two completely different cores, while AMD developed a slightly different version of the existing Zen-4-based core, differentiated only by minor changes (cache, I/O). The launch, yesterday, of his Siena EPYC 8004 processor points in that direction.
Leave a Reply