Benchmarks

How the scores work

• Every category score and the total are out of 10,000, which only a CPU that is best on every benchmark involved can reach. All of them use the same calculation; a category uses its own benchmarks (one for single-thread, multi-thread and gaming, the three Blender scenes for rendering). The total uses the four categories plus the manufacturer's recommended price as a fifth benchmark where lower is better (raw = 1 ÷ price).
• The calculation runs over all scored CPUs at once:
  1. pct = raw ÷ max(raw on that benchmark)
  2. Across CPUs: z of a CPU's pct against all CPUs on that benchmark, adjusted = raw × (1 + 0.1 × z), normalized to the best CPU per benchmark and summed per CPU. Dividing by the best sum gives P1.
  3. Within each CPU: the same, but z compares a pct with the CPU's own other benchmarks. This gives P2.
  4. total = (P1 + P2) ÷ 2 × 10,000
  Normalizing every benchmark to its best CPU keeps the category with the biggest numbers from deciding the ranking. The z-score factors reward CPUs that stand out against the field and are strong across their own results.
• Prices are Intel's recommended customer price from Intel ARK (2018 snapshot) or, for other products, the launch price listed on Wikipedia. 1748 of 2125 scored CPUs have one. Where a result is missing (usually the price), it counts as the average result of the CPUs that have one, so it neither helps nor hurts.
• * marks estimated values. 2125 of 2754 CPUs have results; 1132 of them have measured rendering results.

Ranking