Lisa Su, the famous CEO of AMD, has shared with the world that a server part based on Zen architecture is scheduled for Q2 2017 and it looks like the company will meet the goal.
Naples attacks the core of Intel’s success, the data center network. Many have heard the term cloud computing and we might disappoint some of you when say that a "cloud" is just another server computer sitting at some data center around the world.
Whenever you open any photo from the cloud - for example Facebook or Google Photos - you are requesting stuff from one of the data center machines. This is where Naples fits in, as this is the CPU that targets the server and data center markets.
Naples has 32 cores and 64 Hyper thread support and it is the first of many processors to launch in the coming quarters. The support for 8-channels memory per Naples devices makes this data center / server SoC rather unique. The dual socket server supports up to 32 DIMMS of DDR4 on 16 memory channels delivering up to four terabytes of total memory capacity. This will play an important role in some memory intensive systems.
A two socket Naples system supports 16-channels memory, effectively doubling the memory channels.
AMD is using the very fast Infinity Fabric coherent interconnect that should increase the performance of Naples in dual socket configurations. The processor is a complete fully integrated SoC and comes with high-speed I/O supporting 128 lanes of PCIe 3.
AMD's Senior Vice President, Forrest Norrod, said that the scalability will increase with AN Infinity Fabric connection. When you add a second CPU, you can expect a 2X increment in performance. Naples, it is important to mention, has the chipset inside of the SoC, making it a server SoC rather than just a CPU.
Some might ask why would you need a 128 lines of PCIe 3 and the simple answer is that the system will be able to have faster access to your superfast network cards of Radeon Instinct cards.
Naples also comes with a highly-optimized cache structure for high-performance, energy efficient compute as well as dedicated security hardware.
It comes as a surprise that AMD didn’t talk in detail about the Zeppelin clusters that are the key element of any Naples Zen based server core. Zeppelin comes with eight Zen cores, each having 512 KB cache (4 MB for eight cores) and an additional 16 MB of L3 cache where four Zen processors get access to 8MB. PCIe Gen 3 is integrated on the Zeppelin cluster that also integrates the SATA 3 and 10GbE card. AMD secure processor, DDR4 memory controlleras well as server control hub are the part of Zeppelin cluster.
You need four Zeppelin clusters to get the 32 core based Naples. Norrod said to expect Naples generation of Zen based server and data center products to ship in the second quarter of this year. Experience has taught us to bet on the latter part of the quarter, with late May / early June - that's Computex time - as a possible launch timeframe.
AMD made a short Naples introduction video, below.