Final Analysis
The Xeon Scalable Processor launch is indicative of Intel’s larger agenda. The company is doubling down on its storage, networking, and FPGA efforts. Not surprisingly, then, it builds several (complementary) technologies into the new Scalable Processors and Lewisburg chipset.
The consolidated platform fuels an almost entirely converged solution for enterprise-oriented customers. Chipset-integrated 10 GbE networking and on-package Omni-Path connectivity could eliminate the need for add-in NICs altogether. QuickAssist technology, which provides a boost to some storage and networking workloads, may replace dedicated accelerators. And we might see Intel's vROC feature render dedicated HBA and/or RAID controllers obsolete for large NVMe-based SSD deployments. We’re told to expect integrated FPGAs in the future, but Intel already has Altera-derived add-in options available.
Intel also has Optane DIMMs in development, which would widen its reach even more. We expected those modules to make their debut with the Purley platform, but they've clearly been pushed back. Regardless, the company is serious about storage. Optane DC P4800X SSDs are on offer for the best possible performance in storage-bound workloads, along with Intel's Memory Drive technology that merges Optane into the memory hierarchy.
Our performance metrics revealed sizeable gains in multi-threaded workloads compared to the Broadwell-EP generation. Higher Turbo Boost clock rates helped improve single-threaded results in applications poorly optimized for parallelism, too. The Platinum 8176’s enhanced memory controller improves throughput, really benefiting memory-bound tasks. As you saw, our benchmarks showed phenomenal bandwidth figures. The addition of AVX-512 boosts per-core compute power in properly-optimized software, while changes to the cache hierarchy and mesh topology more generally improve the CPU's performance profile. It is worth noting that Intel's architectural tweaks don't benefit all applications; it's possible that optimization could change this in the future, though.
Although we saw higher power consumption numbers from the Platinum 8176-based machine than our comparison systems with fewer cores, a bit of math shows Intel's latest achieving much better per-core efficiency than any generation prior.
Most data center and enterprise customers are locked into three- and five-year refresh cycles due to maintenance contracts. These folks typically upgrade to new servers as those contracts expire. Anyone replacing hardware that old with Xeon Scalable Processors is going to realize a massive performance and efficiency improvement.
Intel’s data center dominance is about to be challenged by AMD’s EPYC processors. Fortunately for the incumbent, it's going to take some time for the EPYC ecosystem to mature and prove itself trustworthy. Still, considering AMD’s solid line-up of launch partners, the company is off to a good start. Until we’ve seen concrete data from AMD’s latest, though, it’s hard to speculate on how warmly businesses embrace the underdog. Much of AMD’s success is likely to hinge on cost more than performance. The Scalable Processor family features Intel’s famous segmentation and high cost structure, and there’s no doubt enterprise buyers look forward to a competitive AMD to keep Intel on its toes.
Intel’s advances, taken as a whole, are impressive. The company is also transitioning to a Data Center-First strategy that will find the newest architectures debuting on server platforms. IT professionals, rejoice!
MORE: Best CPUs
MORE: Intel & AMD Processor Hierarchy
MORE: All CPU Content