What Is Computing Power? From ENIAC to the AI‑Driven Cloud Era

What Is Computing Power

Computing power, literally "computing ability", refers to the capability to process information data and produce desired results.

Humans have always possessed this ability; our brains act as a powerful computing engine, performing mental calculations continuously.

Historically, simple tools like counting rods and abacuses evolved into mechanical devices, and in 1946 the first electronic computer ENIAC marked the start of the digital era.

The advent of semiconductor technology introduced chips as the main carrier of computing power, leading to the development of integrated circuits in 1958.

During the 1970s‑80s, Moore's Law drove rapid chip improvements, enabling the creation of personal computers (PCs) in 1981, which democratized computing power beyond large enterprises.

In the 21st century, cloud computing emerged as a transformative technology, aggregating dispersed resources into a virtual, scalable "computing resource pool" that users can access on demand.

Cloud computing consolidates CPUs, memory, storage, and GPUs, offering higher reliability, performance, and lower cost compared to self‑owned hardware.

Classification of Computing Power

Computing power is broadly divided into two categories: general-purpose and specialized.

General-purpose chips (e.g., x86 CPUs) handle diverse tasks but consume more power, while specialized chips include FPGAs and ASICs designed for specific workloads.

Examples include Bitcoin mining, which progressed from PC CPUs to GPUs, then to FPGA and ASIC clusters for efficiency.

In data centers, workloads are split into basic general computing and high‑performance computing (HPC), the latter further divided into scientific, engineering, and intelligent (AI) computing.

Scientific and engineering computing generate massive data (e.g., oil‑gas exploration exceeding 100 TB per project) and demand extreme performance.

AI computing relies heavily on GPUs and dedicated accelerators (TPU, NPU, DPU) because of its intensive matrix operations.

Measuring Computing Power

Common metrics include FLOPS (floating‑point operations per second) and its multiples (MFLOPS, GFLOPS, TFLOPS, PFLOPS), as well as MIPS, DMIPS, OPS, etc.

Different precision formats (FP16, FP32, FP64) affect performance and energy consumption.

Forecasts suggest that by 2030 general‑purpose FP32 performance will increase tenfold to 3.3 ZFLOPS, while AI‑oriented FP16 performance could grow 500× to 105 ZFLOPS.

Current State and Future Outlook

Computing power has become a public utility akin to water and electricity, underpinning digital economies and societal intelligence.

Globally, higher national computing capacity correlates with greater GDP; China ranks second in total capacity (135 EFLOPS in 2020) but lags in per‑capita figures.

Future demand will surge with autonomous driving, smart factories, and pervasive IoT, potentially increasing AI compute needs by hundreds of times by 2035.

Predictions indicate global computing capacity could reach 6.8 ZFLOPS by 2025, a 30× rise from 2020.

Conclusion

Despite its importance, computing power utilization remains inefficient, with low usage rates and uneven distribution.

Addressing these challenges requires better resource scheduling, leveraging network technologies, and continued innovation in chip design and cloud infrastructure.