Access Official Technical PDF

Share Summary:LinkedInEmail

For infrastructure teams, data center hardware trends affect more than technology roadmaps: they shape server procurement, storage architecture and the cost of scaling compute. A 2018 market snapshot pointed to a period of transition. Semiconductor demand and component prices were rising, while improvements in price-performance were slowing in several core areas. These observations are historical, not a forecast of today’s market, but they show why engineering decisions need to account for the full system rather than processor specifications alone.

Data center hardware trends: progress meets constraints

The semiconductor market was growing rapidly in 2017, with much of the increase linked to higher RAM prices. At the same time, manufacturing advances were becoming more difficult and expensive. Process-node labels also did not provide a simple basis for comparing manufacturers: the reported 10 nm Intel process was described as comparable to a 7 nm process from Samsung or TSMC. The implication for buyers is that a nominal node size alone does not explain the capabilities or economics of a device.

Demand for personal computers and tablets was declining, while smartphone sales were broadly flat. Longer device lifetimes and a replacement-focused market constrained growth in some consumer segments. Meanwhile, processor development was increasingly directed toward machine learning, including specialized accelerators and chips designed to handle workloads locally rather than sending all data to cloud systems.

That shift matters to industrial automation because computing requirements vary by application. Local processing can be relevant where systems need to analyze data near its source, but the 2018 material does not establish that a particular processor or architecture suits every industrial workload. Selection still depends on the task, system design and operating requirements.

Server architecture and procurement economics

In the reported fourth quarter of 2017, the server market generated $20.7 billion in revenue, with 2.8 million servers shipped. The presentation attributed the revenue increase to several factors, including higher component prices, demand for high-end systems, and investment in high-performance computing and artificial intelligence. It also described a market concentrated around Intel: x86 processors accounted for 99.3% of server units, while Intel held 85% of server revenue.

Alternative architectures existed, but their market positions varied. IBM Power9 was aimed at high-end and HPC or AI systems, often in combination with Nvidia GPUs. AMD EPYC and Arm-based designs were presented as potential alternatives, though the source characterized their adoption or prospects as uncertain at the time. This distinction is useful for technical decision-makers: an available alternative is not automatically a proven replacement across an installed server estate.

Procurement comparisons also require care. The source estimated future CPU server price-performance improvement at 15% per year and disk-server price-per-capacity improvement at 20% per year, based on CERN procurement experience. It also noted substantial variation between hard-drive models and between street and purchase prices. These figures should be read as assumptions from that context, not universal benchmarks. For a robust comparison, teams need to assess the complete configuration and actual acquisition costs.

Memory and storage: different technologies, different trade-offs

DRAM and NAND

DRAM scaling was becoming more challenging as smaller cell sizes required increasingly demanding capacitor structures and higher fabrication costs. The presentation reported a roughly 120% DRAM price increase over the previous 18 months. It also described newer memory technologies—including resistive RAM, magnetic RAM, phase-change memory and ferroelectric RAM—as contenders, but said no cost-effective option was then ready to replace both DRAM and NAND.

NAND production had moved from two-dimensional to three-dimensional designs, with layer counts continuing to rise. Yet additional layers brought manufacturing challenges, including cell uniformity and current density, alongside substantial investment requirements. For infrastructure planners, the broader point is that memory and flash supply economics can influence system costs independently of CPU pricing.

HDD, SSD and tape

Hard disk drives remained important for high-capacity nearline storage, including cloud and research environments. Desktop, mobile and enterprise uses were increasingly being replaced by SSDs, but the source expected the capacity-drive price gap between SSDs and HDDs to remain significant. It also reported a slowdown in annual price improvements across storage categories.

Manufacturers were pursuing higher HDD capacity through approaches such as heat-assisted and microwave-assisted magnetic recording, while multi-actuator designs were discussed as a way to address input/output performance per terabyte. These developments did not remove the central planning question: whether a workload prioritizes capacity, performance or cost.

Tape offered another capacity-oriented option. The presentation identified LTO-8 at 12 TB and IBM TS1155 at 15 TB as current-generation products in 2018, and noted that tape density had room to improve. However, tape media shipments and drive sales had been declining, with a small supplier base. Organizations considering tape therefore needed to weigh its storage role against supplier and ecosystem considerations, rather than treating capacity alone as the deciding factor.

What infrastructure teams should take from the 2018 picture

The 2018 market snapshot showed technology progress continuing alongside manufacturing constraints, volatile component prices and concentrated suppliers. It also highlighted why storage tiers and server configurations should be matched to workload needs, not selected from headline specifications alone. For engineering and operations teams, practical planning means:

  • Comparing complete server configurations and procurement costs, not just processor models.
  • Separating compute, memory and storage requirements when evaluating price-performance.
  • Choosing HDD, SSD or tape according to workload priorities and capacity needs.
  • Tracking processor and memory supply conditions when budgeting infrastructure changes.

These historical findings are not a substitute for current market data. Their enduring practical lesson is that infrastructure decisions depend on the interaction between technology, supply, workload and cost. A disciplined evaluation of those factors helps plant and IT decision-makers build systems that fit operational requirements rather than chase a single market trend.

Access Official Technical PDF

About The Author