Geekbench

Geekbench, by Primate Labs, is a popular cross-platform benchmark that runs on Windows, macOS, Linux, Android, and iOS. Its portability and quick runtime made it the most cited consumer CPU comparison, while its GPU compute module benchmarks accelerator performance. Results are normalized against a reference score so numbers are comparable across very different devices, from a phone to a server, which is a large part of its popularity in reviews and leaks.

What It Measures

The CPU test reports single-core and multi-core scores derived from workloads modeled on real tasks: file compression, HTML parsing, PDF rendering, image processing, machine-learning inference, ray tracing, navigation, and more. The Compute test measures GPU performance for parallel workloads through OpenCL, Metal, Vulkan, or CUDA. Scores are unit-less and pegged to a baseline reference machine, so a higher number is proportionally faster. The split between single-core and multi-core scores helps separate per-thread strength from parallel scaling.

Methodology

Geekbench runs a battery of short subtests, each timed and weighted, then aggregates them into geometric-mean scores. The single-core test uses one thread; the multi-core test scales across all available cores. The tool records device, OS, compiler, and clock details with each result and uploads them to an online browser for comparison and sharing. Because runs are brief, lasting only minutes, thermal and power limits strongly influence sustained-device results, and a device may post a higher score on the first run than on repeated back-to-back runs once it heats up.

How to Interpret Results

Use single-core scores to judge latency-sensitive and lightly threaded performance, and multi-core scores for parallel workloads. Higher is better. Be cautious comparing across major versions: Geekbench rebaselines its reference between versions (for example v5 to v6), so scores are not interchangeable across versions and a higher version number does not mean a faster device. On phones and thin laptops, short test duration can report peak rather than sustained performance, overstating real-world throughput under thermal throttling, so pair it with sustained-load testing where heat matters.

Limitations

Geekbench's brevity makes it sensitive to boost clocks and thermal headroom, so it can flatter devices that cannot sustain those speeds in longer workloads. Its consumer-oriented subtests do not reflect server, database, or HPC workloads well, and the multi-core score's scaling assumptions may not match heavily threaded server software. Version rebaselining breaks long-term comparisons, and leaked single-run scores can be unreliable. For rigorous server evaluation prefer SPEC CPU; treat Geekbench as a fast, broad, cross-platform indicator rather than a definitive measure.