AMD Aims To Shake Up Enterprise Server Market With New Zen-based Naples Platform
Posted on March 7, 2017 12:00 PM by Rob Williams
Leading up to the launch of AMD’s first Ryzen CPUs, the hype was pretty mind-blowing. I’ve gone through and dealt with a slew of hyped launches, but I’m pretty sure Ryzen’s takes the cake. One of the big reasons for that hype is the fact that people have been craving for something new – and if it’s not game-changing, it at least should be a shaker-upper. Ryzen is certainly that.
That sentiment was conveyed by the SVP and GM of AMD’s Enterprise, Embedded and Semi-Custom Business Group, Forrest Norrod, during a presentation to press and analysts a couple of weeks ago. In one of his first slides, Norrod quoted the late Steve Jobs: “What’s the point of focusing on making the product even better when the only company you can take business from is yourself?”
Pow! Right in the kisser.
To help out the enterprise market, which is “starved of innovation”, AMD is bringing its biggest, baddest Ryzen to market later this year. If you think the 1800X’s 8 cores are impressive, dig this: Naples will launch with 32-core parts. That’s 64 threads. And with it comes some other impressive capabilities.
With an eight-channel memory controller, Naples can push unbelievable bandwidth, and in extreme workloads, AMD says that its platform can survive a true torture test easier. In a seismic benchmark, AMD limited its 32-core chips to 22-core (in a 2 socket system) to match the 44 cores on the Intel platform (2699 v4). Ultimately, AMD won, and as seen in the shot below, when the 32-core chip (and its partner in crime) battle Intel’s high-end duo, it can dominate.
This particular benchmark is very memory intensive, so one part of this performance gain is thanks to the general architecture and number of cores, while the other part is the improved memory bandwidth. For a brutalizing test, AMD ramped-up the complexity of the problem, and when in that situation, Intel’s platform failed to produce a result.
It should be stressed (no pun) that this is just one benchmark, and an awfully specific one, but the fact that AMD’s chip could survive a test Intel’s couldn’t is noteworthy. Naples is built to both take a beating, and deliver a beating.
Here’s how AMD positions itself against one of Intel’s latest top-end Xeons, the E5-2699A V4:
Total PCIe Gen3
8×16 = 128 Lanes
2×40 = 80 Lanes
While the top Naples chip will have 128 PCIe lanes, half of those will be dedicated to the “Infinity Fabric” AMD’s created that improves communications between the system hardware. Unlike Ryzen’s first desktop chips, which leave most of the peripheral support on the motherboard, Naples is different. These are in effect SoCs, even complete with Gigabit Ethernet. Even though half of the lanes go missing due to the Infinity Fabric, 64 free per CPU means there’s plenty of bandwidth for compute cards, ultra-fast PCIe storage, et cetera.
Ultimately, AMD says that Naples improves the core count over Intel’s top-dog by 45%, improves bandwidth by 122%, and offers 60% better I/O. We’ve seen on the desktop that Ryzen can tear through workloads that take advantage of many cores, so we’d expect to see the same happen on Naples.
If you happen to be at the Open Compute Summit, you’ll be able to learn more about Naples first-hand. For the rest of us, we’ll learn more as the first half of the year trucks on, as AMD expects its first Naples chips to be released sometime before 2H.