Historical PC Build Data

This is an extra post accompanying “Modeling Optimal PC Building Decisions”, which analyzed historical data to construct a model of computer performance over time and optimize budget choices.

Minor details omitted from main post

#

For performance calculations, we used a 70% performance boost in cases where older builds had an SLI GPU configuration.

As mentioned in the main post, the actual improvement depends heavily on the use case, but this is a decent approximation and matches legacy guidance from Logical Increments themselves.

Raw data sets

#

Here, we store the raw data sets that we used.

CPU/GPU benchmarks

#

See cpumark_20240222.csv and gpumark_20240222.csv, which contain complete sets of CPU and GPU benchmarks scraped from PassMark in February 2024.

Logical Increments PC build history

#

See logicalincrements_history_2013_to_2024.csv, containing historical PC builds for the 10 tiers provided by Logical Increments from 2013 to 2024, scraped with help from the Internet Archive.

To be precise, the data ranges from December 29, 2012, to April 1, 2024, with samples taken approximately every three months.

For each build, we include the component names and prices for the GPU, CPU, motherboard, and RAM, as well as the total build cost. Interestingly, some of the older builds had their totals calculated incorrectly in the website, so I’ve corrected those discrepancies here.

Mapping between Logical Increments and PassMark component names

#

Importantly, the Logical Increments data uses shorthands for the component names, so we needed to map them to the more precise names in the PassMark benchmarks. This was overwhelmingly an automated process,¹ but we’ve replicated the final mapping in logicalincrements_passmark_component_mapping.csv.

Note that SLI setups are mapped to their single-GPU equivalents (e.g., both “GTX 770” and “GTX 770 x2” are mapped to “GeForce GTX 770”).

Inflation data

#

For inflation computations, we used the Core CPI, CPILFESL², as available from the Federal Reserve Economic Data (FRED).