Hardware reviews publish hundreds of benchmark scores for every product. Understanding which scores are meaningful for your use case — and which are marketing-adjacent numbers — is as important as knowing the scores themselves.
Gaming Benchmarks
Gaming benchmarks run specific titles at specific settings and record average fps, 1% lows (the bottom 1% of frame times), and 0.1% lows. The 1% and 0.1% lows are more important for perceived smoothness than average fps. A GPU with higher average fps but worse 1% lows feels worse to play than a slightly lower-average-fps GPU with better consistency.
Look for benchmarks that test your specific games, not just popular benchmarking titles that may not represent your library.
Synthetic CPU Benchmarks
Cinebench R24 and Blender benchmarks measure multi-threaded rendering performance. Passmark CPU Mark measures a blend of single and multi-thread operations. These are useful for comparing within categories but overstate real-world differences for general use — a CPU with 20% better Cinebench score does not make a browser 20% faster.
Storage Benchmarks
Sequential read/write speeds (tested by CrystalDiskMark) measure large file transfers. These numbers are what SSD manufacturers advertise. The more relevant benchmark for most use cases: random 4K read/write speeds, which measure how quickly the drive handles the small file operations typical of OS loading, game loading, and application launching.
The Limits of Benchmarks
Benchmarks test specific scenarios in controlled conditions. Real-world performance includes driver quality, thermal throttling, memory subsystem behaviour, and software optimisation — factors that benchmarks standardise away. A GPU that scores 10% higher in a synthetic benchmark may not consistently outperform in real games, especially at different settings than the benchmark used.