In the x86 server CPU market, Intel is projected to capture 99% of the total annual shipments for this year. By contrast, AMD’s shipment share is estimated at just 1%.

DRAMeXchange analyst Mark Liu pointed out that although the x86 architecture has helped lower the average manufacturing cost of servers by becoming the prevailing market standard, the overall average utilization rate of x86 CPU cores has been stuck around 50%. “Software optimization can raise utilization rates of x86 CPUs to the range of 70% to 80%,” said Liu. “However, data centers for high-performance computing applications require significantly higher utilization rates from their server processors.

Major server ODMs are now working closely with CPU makers to improve processor design and hardware integration. Their efforts have led to immediate and incremental differences in the utilization rates for the latest solutions. Some of the improvements being offered include the addition of an embedded FPGA and deploying GPUs as accelerators.

In the competition within the mainstream server market, the x86 architecture remains in the dominant and advantageous position due to having a wider range of products as well as enjoying a greater level of hard- and software support. Conversely, the ARM architecture is at a disadvantage because its system-on-chip (SoC) products are used by a smaller and specific group of customers.

Furthermore, the competitiveness of ARM-based solutions depends on their ability to integrate with the whole server system. Looking ahead, x86-based solutions with Intel leading their development will continue to be the market mainstream in 2018. The global shipment share of x86 server processors for 2018 is expected to be maintained at above the 90% level.

As the demand in the server market increasingly focuses on high-end servers with more powerful computing capability, the role of GPUs in helping the optimization of CPU cores also becomes more significant. When the frontend servers in a major data center are dealing with multiple tasks that consume huge amounts of computing power, they often rely on server GPUs, each of which contains thousands of smaller and more efficient processing cores. Compared with CPUs, GPUs are more effective at doing parallel computing.

Majority of shipments of discrete GPUs used in mainstream servers come from NVIDIA and AMD, according to DRAMeXchange’s market tracking during the first half of 2017. In terms of the global shipment share for this half-year period, NVIDIA managed to take nearly 70%. NVIDIA’s main offerings for high-end servers are still based on the Pascal architecture. Solutions derived from this platform are mainly designed for large-scale Internet data centers.

As data centers evolve to include high-density computing zones, DRAMeXchange anticipates that more high-performance computing (HPC) servers will adopt general-purpose computing on graphic processing units (GPGPU) in 2018. The penetration rate of GPGPU in the global HPC server market is forecast to go up from 3% in 2017 to 5% in 2018.