Skip to main content

Introduction to Computing Power Units

·550 words·3 mins
Computing Power FLOPS TOPS MIPS DMIPS Hash/S
Table of Contents

Computing power—also known as computing capability—refers to the speed and efficiency with which a system performs computational tasks. Because computing power represents a measurable capability, standardized units are required. Different chip architectures and workloads emphasize different metrics, but the most common computing power units are explained below.

Methods for Measuring Computing Power
#

FLOPS
#

FLOPS (Floating Point Operations Per Second) measures how many floating-point calculations a system can perform per second. It is widely used for evaluating CPUs, GPUs, and supercomputers.

Common FLOPS magnitude units:

  • KFLOPS – 10³ FLOPS
  • MFLOPS – 10⁶ FLOPS
  • GFLOPS – 10⁹ FLOPS
  • TFLOPS – 10¹² FLOPS
  • PFLOPS – 10¹⁵ FLOPS

General FLOPS formula:

$$ [ \text{FLOPS} = \text{Cores} \times \text{Frequency} \times \text{Instructions per Cycle} \times \text{Operations per Instruction} ] $$

A more practical simplified formula for peak theoretical performance in many modern FPUs is: $$ [ \text{TFLOPS} = \text{Cores} \times \text{Frequency (GHz)} \times \text{Vector Width} \times 2 ] $$

Example (based on the original logic):

If a processor has:

  • 4 cores
  • 3.5 GHz clock
  • 4 instructions per cycle
  • 0.5 ns effective execution time per instruction

Then:

$$ [ \text{FLOPS} = 4 \times 3.5 \text{ GHz} \times 4 \times 0.5 \text{ ns} = 56 \text{ GFLOPS} ] $$

TOPS
#

TOPS (Tera Operations Per Second) measures performance in trillion operations per second and is especially important for AI and neural-network accelerators. $$ [ \text{TOPS} = \text{Clock Frequency} \times \text{Instructions per Cycle} \times \text{Ops per Instruction} ] $$ Many AI chips also publish TOPS/W, representing how many trillion operations they achieve per watt—critical for edge AI and mobile devices.

MIPS
#

MIPS (Million Instructions Per Second) measures how many general CPU instructions a processor executes per second.

Calculation:

  1. Count total executed instructions in 1 second.
  2. Divide by 1,000,000.

Example:

If a processor executes 500,000 instructions in one second: $$ [ \text{MIPS} = \frac{500{,}000}{1{,}000{,}000} = 0.5 \text{ MIPS} ] $$

MIPS is simple but often not representative of real-world performance, because instruction complexity varies widely across architectures.

DMIPS
#

DMIPS (Dhrystone MIPS) is based on the Dhrystone benchmark, which uses a fixed synthetic workload to estimate CPU performance in a more standardized way than raw MIPS.

Calculation:

  1. Run the Dhrystone benchmark for a fixed time.
  2. Count executed benchmark instructions.
  3. Divide by 1,000,000.

Example:

If 800,000 benchmark operations run in 1 second:

$$ [ \text{DMIPS} = \frac{800{,}000}{1{,}000{,}000} = 0.8 \text{ DMIPS} ] $$

DMIPS is often used in embedded systems evaluation, such as ARM Cortex-M or R-series processors.

Hash/s
#

Hash/s (hashes per second) measures how fast a device computes cryptographic hash functions, commonly used in cryptocurrency mining.

How it is measured:

  1. Select a hash algorithm (e.g., SHA-256 for Bitcoin).
  2. Run the hash function repeatedly.
  3. Count how many hashes are computed per second.

Example:

If a computer computes 100,000 SHA-256 hashes in one second:

$$ [ 100{,}000 \text{ hash/s} ] $$

Hash rate is crucial for workloads requiring brute-force hashing, such as blockchain mining, password cracking, and integrity verification.

Summary
#

Computing power units vary depending on the workload:

  • FLOPS → floating-point scientific or GPU workloads
  • TOPS → AI accelerators and deep learning inference
  • MIPS/DMIPS → general CPU and embedded processing
  • Hash/s → cryptographic and blockchain workloads

Understanding these units helps evaluate processors more accurately, especially when comparing CPUs, GPUs, NPUs, and specialized accelerators across different application domains.

Related

Understanding TOPS, FLOPS, MIPS, and DMIPS
·473 words·3 mins
Computing Power TOPS FLOPS MIPS DMIPS
Arm to Cancel Qualcomm Architecture Licensing Agreement
·404 words·2 mins
Arm Qualcomm Legal Dispute Semiconductors
Global Rack Server Solutions for the NVIDIA Blackwell Platform
·341 words·2 mins
NVIDIA Blackwell GB200 Microsoft Azure Google Cloud Meta