More CPU cores always means a faster computer

What we know

Modern CPUs (central processing units) are commonly marketed based partly on their core count, individual processing units within a single chip capable of executing instructions, with more cores widely assumed by consumers to mean straightforwardly better or faster performance across the board. This assumption does not match how computing performance actually works for most everyday tasks, because taking advantage of multiple cores requires software specifically designed to divide its workload into parallel, simultaneously executable pieces, a design requirement not automatically satisfied just because more cores are physically available in the hardware.

Many common computing tasks are substantially or entirely "single-threaded," meaning the software executes as a sequential series of steps that cannot be meaningfully divided across multiple cores, either because the task's logic requires steps to happen in order (each step depending on the previous one's result) or because the software was written without parallel execution in mind. For these workloads, a CPU with fewer cores but a higher per-core clock speed and more efficient per-core architecture can outperform a chip with many more cores but weaker individual core performance, a well documented pattern in CPU benchmarking, including extensive testing conducted by hardware review outlets like AnandTech and Tom's Hardware, which consistently show single-threaded application benchmarks (such as many older games, certain productivity software functions, and specific compression algorithms) scaling poorly or not at all with additional cores beyond a certain point.

This relationship is formally described by Amdahl's Law, a foundational principle in computer science articulated by Gene Amdahl in 1967, which mathematically demonstrates that the overall speedup achievable by parallelizing a task is fundamentally limited by the proportion of that task which cannot be parallelized; even a task that is 90 percent parallelizable has a hard mathematical ceiling on total speedup regardless of how many additional cores are thrown at the remaining, unavoidably sequential 10 percent, a principle computer architecture researchers and processor manufacturers explicitly account for in chip design and software optimization work.

Certain workloads do scale well with additional cores, and this is where the "more cores" marketing claim holds up: video editing and rendering, 3D rendering, scientific simulations, certain server and database workloads handling many simultaneous independent requests, and software specifically written using parallel programming techniques can see substantial, sometimes near-linear, performance improvements from additional cores, because these tasks can genuinely be broken into many independent pieces processed simultaneously. Modern operating systems and many contemporary applications have also increasingly been redesigned to take better advantage of multiple cores for background tasks and multitasking between different applications running simultaneously, meaning more cores can meaningfully improve the experience of running many programs at once even if any single program does not itself benefit dramatically. The evidence-based summary is that more cores provide a real, meaningful performance benefit for specific, genuinely parallelizable workloads and for general multitasking across many simultaneous applications, but the blanket claim that more cores make a computer faster for all tasks misunderstands the fundamental software and architectural requirements for parallel execution, and many common single-threaded tasks see limited or no benefit from cores beyond what a given piece of software can actually use.

Common claims

• A 16-core CPU will always outperform an 8-core CPU.False. For single-threaded tasks, per-core speed matters more than core count.
• More cores are always better for gaming.False. Most games are limited by single-core performance and GPU speed.
• Professionals doing video editing benefit from more cores.True. Video encoding and rendering are highly parallelizable workloads.

Evidence hierarchy

All sources

• Validity of the Single Processor Approach to Achieving Large Scale Computing CapabilitiesGene Amdahl, AFIPS Conference Proceedings · 1967
• CPU architecture and multi-core performance benchmarkingAnandTech · 2023
• Multi-core processors explainedIntel · 2023
• Parallel computing fundamentalsTom's Hardware · 2022

Related terms