Section by Ian Cutress
The launch of Intel’s Ice Lake Xeon Scalable processors has been within the wings for quite a lot of years. The delays to Intel’s 10nm manufacturing course of have given quite a lot of setbacks to all of Intel’s proposed 10nm product traces, particularly the excessive efficiency Xeon household: attempting to craft 660 mm2 of silicon on a course of is tough at the perfect of instances. But Intel has 10nm in a spot the place it’s economically viable to begin retailing massive Xeon processors, and the official launch right this moment of Intel’s 3rd Generation Xeon Scalable is on the again of over 200,000+ models shipped to main clients to this point. The new flagship, the Xeon Platinum 8380, has 40 cores, provides PCIe 4.0, and takes benefit of the IPC achieve in Intel’s Sunny Cove processor core. We’re testing it towards the perfect available in the market.
Intel’s 3rd Generation Xeon Scalable: 10nm Goes Enterprise
Today Intel is launching the total stack of processors below the threerd Generation Xeon Scalable Ice Lake branding, constructed upon its 10nm course of. These processors, as much as 40 cores per socket, are designed solely for single socket and twin socket methods, competing in a market with different x86 and Arm choices accessible. With this new era, Intel’s providing is aimed to be two-fold: first, the generational uplift in comparison with 2nd Gen, but additionally the narrative round promoting an answer fairly than merely promoting a processor.
Intel’s messaging with its new Ice Lake Xeon Scalable (ICX or ICL-SP) steers away from easy single core or multicore efficiency, and as a substitute is that the distinctive characteristic set, corresponding to AVX-512, DLBoost, cryptography acceleration, and safety, together with acceptable software program optimizations or paired with specialist Intel household merchandise, corresponding to Optane DC Persistent Memory, Agilex FPGAs/SmartNICs, or 800-series Ethernet, supply higher efficiency and higher metrics for these truly shopping for the methods. This angle, Intel believes, places it in a greater place than its opponents that solely supply a restricted subset of those options, or lack the infrastructure to unite these merchandise below a single easy-to-use model.
An Wafer of 40-core Ice Lake Xeon 10nm Processors
Nonetheless, the launch of a brand new era of merchandise and an expanded portfolio warrants the product to truly be put below take a look at for its uncooked base efficiency claims. This era of Xeon Scalable, Intel’s first on 10nm, makes use of a more moderen structure Sunny Cove core. Benefits of this core, as defined by Intel, begin with an additional 20% uncooked efficiency enhance, enabled via a a lot wider core with an improved entrance finish and a extra execution assets. Outside of the core, reminiscence bandwidth is improved each by rising reminiscence channels from six to eight, but additionally new reminiscence prefetch methods and optimizations that will increase bandwidth as much as 100% with one other +25% effectivity. The mesh interconnect between the cores additionally makes use of up to date algorithms to feed IO to and from the cores, and Intel is selling higher energy administration via unbiased energy administration brokers inside every IP block.
On high of this, Intel is layering on accelerative options, stating that over the uncooked efficiency, software program optimized for these accelerators will see a better-than-generational uplift. This begins with the essential core structure, particularly because it pertains to SIMD instructions corresponding to SSSE, AVX, AVX2, and AVX-512: Intel is enabling higher cryptography assist throughout its ISA, enabling AES, SHA, GFNI, and different directions to run concurrently throughout all vector instruction units. AVX-512 has improved frequencies throughout extra complicated bit operations for ICX with smarter mapping between directions and energy draw, providing an additional 10% frequency for all 256-bit directions. On…
