AMD Naples Shows Off Scalability
AMD HAS MADE PLENTY OF WAVES with its Zen architecture, which has some really interesting concepts, along with a price that has attracted a lot of attention. One of the key design elements of Zen is that it’s supposed to be highly scalable, from ultra-low
To make everything work properly, AMD reworked many of the building blocks for Zen, including the Infinity Fabric. Zen features a four-core/eight-thread module as its primary building block, combining 512KB of L2 cache per core with a shared 8MB L3 cache, and Ryzen uses two of these modules for the eight-core/16-thread parts. Naples then takes four of those, and puts them in a single package, yielding a 32-core/64-thread part. And in the process, all of the major interfaces are doubled.
That gives Naples a massive eight-channel DDR4 memory controller, supporting up to 16 DIMMs, with 128 PCIe Gen3 lanes of external connectivity. The whole package also comes with an integrated SoC (System on a Chip), so there’s no need for external chipsets. One Naples CPU is all that’s required to support memory, IO, graphics, and more.
AMD showed off a dual-socket 2U server, running 64-cores/128-threads. The sockets use the Infinity Fabric to talk to each other—in this case, 64 PCIe lanes are used for the 63GB/s interlink between sockets. That leaves 128 total PCIe lanes for external devices, and doubles the memory channels to 16. And AMD rates Naples at up to DDR4-2400 memory speeds with all DIMM sockets populated.
Many server workloads become heavily bound by memory bandwidth and performance, and AMD’s Naples delivers more than twice the memory bandwidth of Intel’s top dual-socket server. For high-end workloads like HPC, transactional databases, and virtualization, this could give AMD a major advantage.
To drive home this point, AMD put together a head-to-head battle between its Naples server and another server equipped with two of Intel’s fastest Xeon part, the E5-2699A v4. Running a seismic analysis workload that involves computationally intensive 3D wave equations, which taxes the entire system, Naples destroyed Intel’s Xeon server. With identical core counts and memory speed, Naples was twice as fast, and with a fully armed battlestation sporting 64 cores and DDR4-2400 memory, Naples was 2.5 times faster.
How realistic is this comparison? That’s difficult to say, but in the high performance computing scenarios where AMD wants Naples to compete, running software that’s optimized for the specific machine architecture is common practice. So, Naples out of the gate should be very impressive, and with plenty of extra cores and memory bandwidth, you’d expect AMD to come out ahead. Zen represents a hard reboot of AMD’s processors, with much greater performance and scalability. We may not want to run 32-core/64-thread processors on our desktops, but servers can easily make use of such capabilities.
Naples is slated to launch in Q2 of 2017, though AMD hasn’t revealed pricing or the various models yet. It will be partnering with other companies to bring the servers to market, and Naples should garner a lot of interest. Intel will also have new Xeon parts later this year, based on Skylake/Kaby Lake, which may close the gap. I wouldn’t expect to see a rapid overhaul of the existing IT infrastructure, as companies tend to be a lot more cautious, but it will be good to have a competitive AMD server solution again.
Naples delivers more than twice the memory bandwidth of Intel’s top dual-socket server