Intel Xeon Scalable Debuts: Dual Xeon Platinum 8176 With 112 Threads Tested
Introducing The Intel Xeon Scalable Processor Family
Intel announced its Xeon Scalable processor family based Skylake-SP a few weeks back, but today marks the official launch of the platform. These new processors feature a new microarchitecture versus previous-generation Xeons and Intel has revamped the naming convention and arrangement of the product stack as well. Whereas previous-generation Xeon processors carried version, class, and model number designations – for example, Xeon E5-2697 v4 – the Xeon Scalable processor family is now designated by Platinum, Gold, Silver, and Bronze categories, with a single model number.
The new naming convention will take some getting used to if you’re already familiar with Intel’s previous-generation Xeons, but it’s relatively straightforward in the grand scheme of things. We will explain shortly...
With the Brickley and Grantley-EP platforms, Xeon E7 series processors sat at the top of the stack, followed by Xeon E5s (and Xeon E3s). Moving forward, with this current-generation of Xeon Scalable processors based on the 14nm Skylake-SP microarchitecture and with next-generation Cascade Lake-based Xeons, however, the Platinum, Gold, Silver, and Bronze naming will be used. Xeon Platinum processors obviously sit at the top of the stack, followed by Gold, and so on.
Xeon Platinum 81xx series processors will have the most cores – up to 28 (56 threads) – the highest frequencies, dual FMA units, up to three 10.4 GT/s UPI links, and the most socket flexibility, with support for 2, 4, and 8+ socket configurations. The processors support up to 6 memory channels at speeds up to DDR4-2666, and can support up to 1.5TB of memory. Xeon Platinum series processors also feature up to 48 integrated PCI Express 3.0 lanes. Xeon Platinum processors are the cream of the crop and target mission-critical, scale-up, enterprise applications.
Xeon Gold 61xx and 51xx series processors will feature up to 22 cores (44 threads), dual FMA units, have two (51xx) or three (61xx) UPI links, support DDR4-2400 (51xx) or DDR4-2666 (61xx) memory, and similar PCI Express 3.0 lane configurations. Xeon Gold 61xx series processors also support 2 or 4 socket configurations. Xeon Gold series processors target high-performance, but more mainstream applications than the higher-end Platinum series.
A handful of Xeon Platinum and Gold series processors are also available with integrated Intel Omni-Path Architecture fabric. One of the processors is pictured at the very top of this page that shows the additional Omni-Path connector and we'll outline the full line-up at the end of the piece.
Xeon Silver 41xx and Xeon Bronze 31xx series processors have up to 12 and 8 cores, respectively, and both support single or dual socket configurations and have single FMA units. The peak memory speed on the Bronze 31xx series parts drops down to DDR4-2133 and these processors also feature dual, reduced-speed (9.6GT/s) UPI links. Xeon Silver and Bronze series processors target more entry-level enterprise applications, where power efficiency may trump ultimate performance.
The new line-up of Intel Xeon Scalable processors are more advanced than the previous-gen in a number of ways. They have more cores, and hence support for more threads. The higher-end parts have faster UPI links, they support more / faster memory channels, and have additional PCI Express lanes. These new Xeon Scalable series processors will also fall into somewhat higher power envelopes, however, and their cache configurations are completely different. More on that a little later though.
With each successive generation, Intel’s goal is increase overall throughput. All of the changes mentioned above culminate in a series of processors that can significantly outpace last-year's Broadwell based Xeons. Versus platforms from a decade ago, performance is up roughly 41x, at least according to SPECint (in a 2 socket configuration). The feature set and efficiency of each successive generation is also enhanced, however, and typically improve I/O connectivity, storage performance, virtualization, memory capacity, and a number of other features, over and above CPU performance.