The server-processor design is no more a closely guarded secret. Unlike few years back, where IBM, Sun Microsystems and Intel had full grip over the server processor market, today ARM 64-bit architecture is soon to make this market just like any other chip in the market. Intel's Xeon which still has a huge market share but is facing powerfull new- comers. The leading fabless chip vendor Qualcomm entered this market by launching a matchable competitor for Intel's Xeon.
The power of processor is not just pack of billions of logic gates but more of the smart architecture to process various kinds of data. Super computers mix and match trditional CPU processors, server processors and graphic processors to achieve high process throughput. Supercomputer also find lot more options now in selecting chips including made-in-China server/super computer processors.
Qualcomm has made available a 10nm Centriq 2400 Family server SoC with up to 48 Custom ARMv8-Compliant Cores for datacenter servers. Qualcomm Centriq 2400 series features the Qualcomm Falkor CPU and target cloud customers seeking new server solutions optimized for total cost of ownership, while meeting performance, efficiency and power demands.
At an event, Qualcomm Datacenter Technologies demonstrated Apache Spark and Hadoop on Linux and Java running on a Qualcomm Centriq 2400 processor. Qualcomm Centriq 2400 processor series is now sampling to key prospective customers and is expected to be commercially available in the second half of 2017. Centriq 2400 uses Qualcomm's new Falkor microarchitecture that is ARMv8-A-compatible and designed exclusively for the Centriq line.
The other popular server processor in the market is, China based Phytium launched 28nm based FT-1500 Series of light weight server Processors, web servers, cloud computing, and data centers. A version of this FT-1500A/16 is called "Earth" server processor chip.
FT-2000 series of processors is based on FTC661 core, a customized core designed by Phytium under architectural license from ARM and is compatible with ARMv8 ISA. Featuring Parallel-System-on-Chip (PSoC) architecture, data-affinitive memory hierarchy and scalable interconnections on chip. A version of this FT-2000/64 is called "MARS" chip. Learn more on this at http://www.phytium.com.cn/Product/detail?language=1&product_id=7
These chips are aimed at fast emerging cloud clients who picked up the trend from cloud experts like Google, Amazon and also targeted at cheap server makers.
Some more server-processors in the market includes Applied Micro's X-Gene family and Cavium's ThunderX 2.
Qualcomm is said to be partnering with Guizhou Province to form a JV called Guizhou Huaxintong Semiconductor Technology to design and sell server chips.
IBM has launched Power9, a processor for the datacenter expected to be available in 2017. Power9 to power super computer such as Summit and Sierra, which expected to be the best in the world beating China's present best super computer Tianhe. NVIDIA Volta GPUs are expected to complement power9s in these super computers. Power9 offered in both single, dual and multi -socket targeting regular clusters, and NUMA servers. Experts say Power9 can connect to virtually any type of accelerator or special-purpose coprocessor supporting PCIe, CAPI or NVLink.
To tell about AMD, AMD to make 32-core Zen server processor. AMD launched 32-core server processor called "Naples" powered by "Zen" x86 processing engine. AMD says "new AMD server processor exceeds today's top competitive offering on critical parameters, with 45% more cores1, 60% more input / output capacity (I/O)2, and 122% more memory bandwidth".
"Naples" features as shared by AMD:
A highly scalable, 32-core System on Chip (SoC) design, with support for two high-performance threads per core.
Industry-leading memory bandwidth, with 8-channels of memory per "Naples" device. In a 2-socket server, support for up to 32 DIMMS of DDR4 on 16 memory channels, delivering up to 4 terabytes of total memory capacity.
The processor is a complete SoC with fully integrated, high-speed I/O supporting 128 lanes of PCIe, negating the need for a separate chip-set.
A highly-optimized cache structure for high-performance, energy efficient compute.
AMD Infinity Fabric coherent interconnect for two "Naples" CPUs in a 2-socket system.
Dedicated security hardware.
Some of server customers designing their own chips, Google has developed TensorFlow Processing Unit (TPU).
As the computing load getting more and more trasferred to cloud from the clients such as phones and notebooks, the global cloud IP traffic is nearly doubling every year. The server processor chips shipments are estimated to grow proportionally.
The use of FPGA in data centers/servers is also a hot trend. FPGA is now important piece of silicon used in data centres for customised computing. Vector processing architecture based accelerators can be implemented on FPGA by rewiring the logic as for the new emerging graphic/image processing loads. If the number of such processors required in thousands of quantities than designing a customised ASIC processor makes business sense.
Nvidia' graphic processors handle well-known standard graphic/video formats helping speeding up vision computing.
All this progress points to ARM based server chip is no more a that lesser performing compared to Intel Xeon server chips. To make the race more interesting, Intel is speeding up its 10nm chips, the 10nm Xeon server are expected to be launched during 2017-18 period.
So for a smart server designer, a mix-and-match of all kind of processor chips make strong sense provided its saves the cost and gets high MIPS computing at lesser power consumption.
All this resulting into cost of computers to run Machine learning and AI coming down significantly.
Note: This article was edited further on 23rd Mar by adding AMD's Naples details.