Steam Hardware Survey

suppressPackageStartupMessages({
library(tidyverse)
library(cowplot)
library(ggrepel)
library(scales)
})
theme_set(theme_cowplot())
options(repr.plot.width=15, repr.plot.height=9)

df <- read_csv('shs.csv', col_types='')
head(df,3)

df_plat = read_csv('shs_platform.csv', col_types='')
head(df_plat,3)

filter(df_plat, grepl('Proc', category)) |>
filter(platform %in% c('pc','linux')) |>
filter(grepl('GenuineIntel|AuthenticAMD', name)) |>
mutate(
    label=ifelse(date==max(date, na.rm=T), sprintf("%s | %.4g%%",name,percentage*100), NA)
) |>
ggplot(aes(x=date, y=percentage, color=name)) +
geom_hline(yintercept=0.5, color='gray') +
geom_line(alpha=0.3) +
geom_smooth(se=F, span=0.5, linewidth=1) +
scale_y_continuous(labels=percent_format()) +
geom_label_repel(aes(label=label), hjust='left', direction = 'y', nudge_x=2.5e3, segment.color='gray') +
guides(color='none') +
scale_x_date(date_breaks='2 years', date_labels = '%Y') +
labs(title='Processor vendor') +
facet_wrap(~platform, ncol=2)

filter(df_plat, grepl('Proc', category)) |>
filter(platform=='pc', date>ymd('2018-03-01')) |>
filter(grepl('GenuineIntel|AuthenticAMD', name)) |>
ggplot(aes(x=date, y=percentage, color=name)) +
geom_rect(aes(ymin=0, ymax=1, xmin=ymd('2026-03-01'), xmax=ymd('2030-01-01')), fill='#FAFAD2',alpha=1, color=NA) +
annotate(geom='text', x=ymd('2027-01-01'), y=0.25, label='Prediction') +
geom_line(alpha=0.5) +
geom_smooth(method='lm', fullrange=TRUE, linewidth=1) +
labs(title='Processor vendor', color='') +
scale_y_continuous(labels=percent_format()) +
scale_x_date(limits=c(ymd('2018-03-01'), ymd('2030-01-01')), date_breaks = '1 years', date_labels = '%Y')

filter(df_plat, grepl('Proc', category), platform=='mac') |>
ggplot(aes(x=date, y=percentage, color=name)) +
geom_line() +
labs(title='Processor vendor on mac platform')

filter(df_plat, grepl('Proc', category)) |>
slice_max(date, n=1) |>
arrange(platform, -percentage)

map_df(c(2, 4, 6, 8, 12, 16, 24, 32), function(target_p) {
    filter(df, grepl('Physical CPUs', category)) |>
    mutate(n_cpus = as.numeric(str_extract(name, '(\\d+)', group=1))) |>
    filter(!is.na(n_cpus), n_cpus >= target_p) |>
    distinct(date, percentage) |>
    mutate(label=paste(target_p, 'cpus or more')) 
}) |>
summarize(
    percentage=sum(percentage),
    .by = c(date, label)
) |>
mutate(
    label2=ifelse(date==max(date, na.rm=T), sprintf("%s | %.4g%%",label,percentage*100), NA)
) |>
ggplot(aes(x=date, y=percentage, color=label)) +
geom_line(alpha=0.3) +
geom_smooth(se=F, span=0.5, linewidth=1) +
scale_y_continuous(labels=percent_format()) +
scale_x_date(date_breaks='2 years', date_labels = '%Y') +
geom_label_repel(aes(label=label2)) +
labs(title='Number of cpus') +
guides(color='none')

map_df(c(1.4, 2.0, 2.3, 2.7, 3, 3.3, 3.7), function(target_f) {
    filter(df, category == 'Intel CPU Speeds') |>
    mutate(
        f0 = case_when(
            grepl('^\\S+ Ghz to', name) ~ str_extract(name, '^(\\S+) Ghz to', group=1),
            grepl('\\S+ Ghz and above', name) ~ str_extract(name, '(\\S+) Ghz', group=1),
            TRUE ~ NA
        )
    ) |>
    filter(f0 >= target_f) |>
    distinct(date, percentage) |>
    mutate(freq=target_f)
}) |>
summarize(
    percentage=sum(percentage),
    .by =c(date, freq)
) |>
mutate(label=ifelse(date==max(date, na.rm=T), sprintf('%.1f Ghz or more | %.2g%%', freq, percentage*100), NA)) |>
ggplot(aes(x=date, y=percentage, color=freq, group=freq)) +
geom_line(alpha=0.3) +
geom_smooth(se=F, span=0.5, linewidth=1) +
geom_label_repel(aes(label=label)) +
scale_y_continuous(labels=percent_format()) +
labs(title='Intel CPU Speeds') +
scale_color_viridis_c(option='turbo') +
scale_x_date(date_breaks='2 years', date_labels ='%Y')

map_df(c(4, 8, 12, 16, 24, 32, 48, 64), function(target_p) {
    filter(df, category=='System RAM') |>
    mutate(
        val=as.numeric(str_extract(name, '(\\S+) [MG]B', group=1)),
        unit=str_extract(name, '([MG]B)', group=1),
    ) |>
    filter(val >= target_p, unit == 'GB') |>
    summarize(
        percentage=sum(percentage),
        .by=date
    ) |>
    mutate(label=paste(target_p,'GB or more'))
}) |>
mutate(label2=ifelse(date==max(date, na.rm=T), sprintf('%s | %2.2g%%',label,100*percentage), NA), .by=label) |>
ggplot(aes(x=date, y=percentage, color=label)) +
geom_line(alpha=0.3) +
geom_smooth(se=F,span=0.5, linewidth=1) +
geom_label_repel(aes(label=label2)) +
guides(color='none') +
labs(title='System RAM') +
scale_x_date(date_breaks = '2 years', date_labels = '%Y') +
scale_y_continuous(label=percent_format())

map_df(c(4, 8, 12, 16, 24, 32), function(target_p) {
    filter(df, category=='VRAM') |>
    mutate(
        val=as.numeric(str_extract(name, '(\\S+) [MG]B', group=1)),
        unit=str_extract(name, '([MG]B)', group=1),
    ) |>
    filter(val >= target_p, unit == 'GB') |>
    summarize(
        percentage=sum(percentage),
        .by=date
    ) |>
    mutate(label=paste(target_p,'GB or more'))
}) |>
mutate(label2=ifelse(date==max(date, na.rm=T), sprintf('%s | %2.2g%%',label,100*percentage), NA), .by=label) |>
ggplot(aes(x=date, y=percentage, color=label)) +
geom_line(alpha=0.3) +
geom_smooth(se=F,span=0.5, linewidth=1) +
geom_label_repel(aes(label=label2)) +
guides(color='none') +
labs(title='VRAM') +
scale_x_date(date_breaks = '1 years', date_labels = '%Y') +
scale_y_continuous(label=percent_format()) -> p
suppressWarnings(print(p))

filter(df, grepl('Video Card Desc', category)) |>
mutate(
    vendor=case_when(
        grepl('AMD|Radeon|ATI', name) ~ 'AMD',
        grepl('Intel|Haswell', name) ~ 'Intel',
        grepl('NVIDIA', name) ~ 'NVIDIA',
        TRUE ~ 'Other'
    ),
    vendor = factor(vendor, levels=c('AMD','NVIDIA','Intel','Other'))
) |>
summarize(
    percentage=sum(percentage),
    .by = c(date, vendor)
) |>
mutate(label=ifelse(date==max(date, na.rm=T), sprintf('%s | %.2g%%', vendor, percentage*100), NA)) |>
ggplot(aes(x=date, y=percentage, color=vendor)) +
geom_line(alpha=0.3) +
geom_smooth(se=F, span=1, linewidth=1) +
guides(color='none') +
labs(title='Video Card Vendor') +
scale_y_continuous(label=percent_format()) +
geom_label_repel(aes(label=label), max.overlaps = Inf) 

filter(df, grepl('Video Card Desc', category)) |>
filter(date==max(date)) |>
mutate(
    name = ifelse(is.na(name), 'Other', name),
    vendor=case_when(
        grepl('AMD|Radeon|ATI', name) ~ 'AMD',
        grepl('Intel|Haswell', name) ~ 'Intel',
        grepl('NVIDIA', name) ~ 'NVIDIA',
        TRUE ~ 'Other'
    ),
    vendor = factor(vendor, levels=c('AMD','NVIDIA','Intel','Other'))
) |>
summarize(
    percentage=sum(percentage),
    .by = c(vendor, name)
) |>
filter(vendor != 'Other') |>
top_n(percentage, n=50) |>
ggplot(aes(y=fct_reorder(name, percentage), x=percentage, fill=vendor)) +
geom_col() +
theme(legend.position = 'inside', legend.position.inside = c(0.3, 0.1)) +
coord_cartesian(expand=0) +
labs(title='Video Card Models',y=NULL) +
scale_x_continuous(labels=percent_format())

filter(df, grepl('Video Card Desc', category)) |>
filter(grepl('Mobile|Laptop', name)) |>
summarize(
    percentage=sum(percentage),
    .by = c(date)
) |>
mutate(label=ifelse(date==max(date), sprintf('%.2g%%', percentage*100), NA)) |>
ggplot(aes(x=date, y=percentage)) +
geom_line(alpha=0.3) +
geom_smooth(se=F, span=0.5, linewidth=1, color='black') +
geom_label_repel(aes(label=label)) +
labs(title='Mobile/Laptop GPUs') +
scale_y_continuous(label=percent_format())

linux_v_list <- filter(df_plat, category=='Linux Version', name != 'Other') |> with(unique(name))
os_df <- 
    filter(df, category=='OS Version') |>
    mutate(
        OS = case_when(
            grepl('Windows', name) ~ 'Windows',
            grepl('MacOS', name) ~ 'MacOS',
            name %in% linux_v_list ~ 'Linux',
            TRUE ~ 'Other'
    ))

filter(df_plat, grepl("Version", category)) |>
rename(percentage_plat=percentage) |>
right_join(os_df, by=c('date','name')) |>
mutate(change.x=NULL,change.y=NULL) |>
mutate(
    corr_fac = 1/sum(percentage_plat),
    .by=c('OS','date')
) |>
mutate(
    corr_fac=replace_na(corr_fac, 1),
    p2 = percentage * corr_fac
) |>
summarize(
    percentage=sum(p2),
    .by=c(date, OS)
) |>
filter(date>ymd('2015-01-01')) |>
ggplot(aes(x=date, y=percentage, color=OS)) +
geom_line(alpha=0.5) +
geom_smooth(se=F, span=0.5, linewidth=1) +
facet_grid(OS~., scales='free_y') +
scale_x_date(date_breaks = '1 years', date_labels = '%Y') +
scale_y_continuous(labels=scales::percent_format()) +
labs(title='Operating System') +
guides(color='none')

filter(df_plat, platform=='linux', category=='Linux Version') |>
mutate(
    distro=case_when(
        grepl('Ubuntu', name) ~ 'Ubuntu',
        grepl('Arch Linux', name) ~ 'Arch',
        grepl('SteamOS', name) ~ 'SteamOS',
        grepl('Pop!_OS', name) ~ 'Pop!_OS',
        grepl('Linux Mint', name) ~ 'Mint',
        grepl('Manjaro', name) ~ 'Manjaro',
        grepl('EndeavourOS', name) ~ 'EndeavourOS',
        grepl('CachyOS', name) ~ 'CachyOS',
        grepl('Bazzite', name) ~ 'Bazzite',
        grepl('Freedesktop', name) ~ 'Freedesktop SDK',
        grepl('Debian', name) ~ 'Debian',
        grepl('Nobara', name) ~ 'Nobara',
        grepl('Fedora', name) ~ 'Fedora',
        TRUE ~ 'Other'
    )
) -> df_distro
summarize(df_distro,
    percentage=sum(percentage),
    .by=c(date, distro)
) |>
mutate(label=ifelse(date==max(date, na.rm=T), sprintf("%s | %.4g%%",distro,percentage*100), NA), .by=distro) |>
ggplot(aes(x=date, y=percentage, color=distro)) +
geom_line(alpha=0.3) +
geom_smooth(se=F, span=0.5, linewidth=1) +
geom_label_repel(aes(label=label), max.overlaps = Inf) +
guides(color='none') +
scale_y_continuous(labels=percent_format()) +
labs(title='Linux distros') +
scale_x_date(date_breaks='1 years', date_labels = '%Y') -> p
suppressWarnings(print(p))

filter(df_distro, distro=='Other', date==max(date)) |>
summarize(
    percentage=sum(percentage),
    .by=name
)

filter(df_distro, grepl('Freedesktop', distro)) |>
slice_max(date, n=1, by=name)

filter(df_plat, platform=='mac', category=='OSX Version') |>
mutate(
    ver=str_extract(name, '(MacOS \\d+)', group=1),
    ver=replace_na(ver, 'Other')
) |>
summarize(
    percentage=sum(percentage),
    .by=c(date, ver)
) |>
filter(date > ymd('2020-01-01')) |>
mutate(label=ifelse(date==max(date, na.rm=T), sprintf("%s | %.4g%%",ver,percentage*100), NA), .by=ver) |>
ggplot(aes(x=date, y=percentage, color=ver)) +
geom_line(alpha=0.3) +
geom_smooth(se=F, span=0.5, linewidth=1) +
geom_label_repel(aes(label=label), max.overlaps = Inf) +
labs(title='MacOS version') +
scale_y_continuous(labels=percent_format()) +
guides(color='none') -> p
suppressWarnings(print(p))

filter(df_plat, platform=='pc', grepl('Windows Version',category)) |>
filter(grepl('Windows', name)) |>
mutate(
    ver=gsub(' (64|32) bit','', name)
) |>
summarize(
    percentage=sum(percentage),
    .by = c(date, ver)
) |>
mutate(label=ifelse(date==max(date, na.rm=T), sprintf("%s | %.4g%%",ver,percentage*100), NA), .by=ver) |>
ggplot(aes(x=date, y=percentage, color=ver)) +
geom_line(alpha=0.3) +
geom_smooth(se=F, span=0.5, linewidth=1) +
geom_label_repel(aes(label=label), max.overlaps = Inf) +
labs(title='Windows version') +
guides(color='none') +
scale_y_continuous(labels=percent_format()) -> p
suppressWarnings(print(p))

filter(df, category=='Primary Display Resolution') |>
mutate(
    disp = fct_lump_prop(name, w = percentage, prop = 0.005)
) |>
summarize(
    percentage=sum(percentage),
    .by=c(date, disp)
) |>
mutate(label=ifelse(date==max(date, na.rm=T), sprintf('%s | %2.2g%%',disp,100*percentage), NA), .by=disp) |>
ggplot(aes(x=date, y=percentage, color=disp)) +
geom_line(alpha=0.3) +
geom_smooth(se=F,span=0.5, linewidth=1) +
geom_label_repel(aes(label=label), max.overlaps = Inf) +
scale_y_continuous(labels=scales::percent_format()) +
labs(title='Primary Display Resolution') +
guides(color='none')

filter(df, category=='Primary Display Resolution') |>
mutate(
    w = as.numeric(str_extract(name, '^(\\d+)', group=1)),
    h = as.numeric(str_extract(name, 'x (\\d+)', group=1)),
    ratio = w/h,
    ratio_intervals = cut(ratio, breaks=quantile(ratio, na.rm=TRUE), include.lowest = TRUE),
    ratio_intervals = fct_na_value_to_level(ratio_intervals, level = 'Other')
) |>
summarize(
    percentage=sum(percentage),
    .by = c(date, ratio_intervals)
) |>
mutate(label=ifelse(date==max(date, na.rm=T), sprintf('%s | %2.2g%%',ratio_intervals,100*percentage), NA), .by=ratio_intervals) |>
ggplot(aes(x=date, y=percentage, color=ratio_intervals)) +
geom_line(alpha=0.3) +
geom_smooth(se=F,span=0.5, linewidth=1) +
geom_label_repel(aes(label=label)) +
scale_y_continuous(labels=scales::percent_format()) +
labs(title='Primary Display aspect ratios')

filter(df, category=='Multi-Monitor Desktop Resolution') |>
filter(name != 'Other') |>
mutate(
    disp = fct_lump_prop(name, w = percentage, prop = 0.01)
) |>
summarize(
    percentage=sum(percentage),
    .by=c(date, disp)
) |>
mutate(label=ifelse(date==max(date, na.rm=T), sprintf('%s | %2.2g%%',disp,100*percentage), NA), .by=disp) |>
ggplot(aes(x=date, y=percentage, color=disp)) +
geom_line(alpha=0.3) +
geom_smooth(se=F,span=0.5, linewidth=1) +
geom_label_repel(aes(label=label), max.overlaps = Inf) +
scale_y_continuous(labels=scales::percent_format()) +
labs(title='Multi-Monitor Desktop Resolution') +
guides(color='none')

filter(df, category=='Multi-Monitor Desktop Resolution') |>
filter(name != 'Other') |>
mutate(
    w = as.numeric(str_extract(name, '^(\\d+)', group=1)),
    h = as.numeric(str_extract(name, 'x (\\d+)', group=1)),
    ratio = w/h,
    ratio_intervals = cut(ratio, breaks=quantile(ratio, na.rm=TRUE), include.lowest = TRUE),
    ratio_intervals = fct_na_value_to_level(ratio_intervals, level = 'Other')
) |>
summarize(
    percentage=sum(percentage),
    .by = c(date, ratio_intervals)
) |>
mutate(label=ifelse(date==max(date, na.rm=T), sprintf('%s | %2.2g%%',ratio_intervals,100*percentage), NA), .by=ratio_intervals) |>
ggplot(aes(x=date, y=percentage, color=ratio_intervals)) +
geom_line(alpha=0.3) +
geom_smooth(se=F,span=0.5, linewidth=1) +
scale_y_continuous(labels=scales::percent_format()) +
geom_label_repel(aes(label=label)) +
labs(title='Multi-Monitor aspect ratios')

filter(df, category=='Language') |>
filter(!is.na(percentage), percentage>0) |>
mutate(label=ifelse(date==max(date, na.rm=T), sprintf('%s | %2.2g%%',name,100*percentage), NA), .by=name) |>
ggplot(aes(x=date, y=percentage, color=name)) +
geom_line(alpha=0.2) +
geom_smooth(se=F,span=0.5, linewidth=1) +
geom_label_repel(aes(label=label), max.overlaps = Inf) +
scale_y_continuous(trans='log10', labels=scales::percent_format()) +
labs(title='Languages') +
guides(color='none') -> p
suppressWarnings(print(p))

filter(df, category=='Language') |>
filter(date==max(date)) |>
filter(percentage>0) |>
ggplot(aes(y=fct_reorder(name, percentage), x=percentage, fill=name)) +
geom_col() +
geom_text(aes(label=sprintf("%.2g%%",percentage*100)), hjust=-.1) +
scale_x_continuous(labels=percent_format()) +
labs(title='Languages', y=NULL) +
guides(fill='none')

map_df(c(100, 200, 500, 1000), function(target_gb) {
    filter(df, category=='Total Hard Drive Space') |>
    mutate(
        lb_gb = case_when(
            grepl('\\d+ GB to', name) ~ as.numeric(str_extract(name, '(\\S+) GB to', group=1)),
            grepl('Above \\d+ TB', name) ~ as.numeric(str_extract(name, '(\\S+) TB', group=1))*1000,
        )
    ) |>
    filter(lb_gb >= target_gb) |>
    distinct(date, percentage) |>
    mutate(
        gb=case_when(
               target_gb >= 1000 ~ sprintf("%.3g TB or more", target_gb/1000),
               TRUE ~ sprintf("%d GB or more", target_gb)
            )
    )
}) |>
summarize(
    percentage=sum(percentage),
    .by = c(date, gb)
) |>
mutate(percentage=pmin(percentage, 1)) |>
mutate(label=ifelse(date==max(date, na.rm=T), sprintf('%s | %.3g%%', gb, percentage*100), NA), .by=gb) |>
ggplot(aes(x=date, y=percentage, color=gb)) +
geom_line(alpha=0.3) +
geom_smooth(se=F, span=0.5, linewidth=1) +
geom_label_repel(aes(label=label)) +
labs(title='Total Hard Drive Space') +
guides(color='none') +
scale_y_continuous(label=percent_format())

map_df(c(100, 250, 500, 1000, 2000, 3000, 4000), function(target_gb) {
    filter(df, category=='Free Hard Drive Space') |>
    mutate(
        lb_gb = case_when(
            grepl('\\d+ GB to', name) ~ as.numeric(str_extract(name, '(\\S+) GB to', group=1)),
            grepl('\\d+ TB to', name) ~ as.numeric(str_extract(name, '(\\S+) TB to', group=1))*1000,
            grepl('Above \\d+ TB', name) ~ as.numeric(str_extract(name, '(\\S+) TB', group=1))*1000,
        )
    ) |>
    filter(lb_gb >= target_gb) |>
    distinct(date, percentage) |>
    mutate(
        gb=case_when(
               target_gb >= 1000 ~ sprintf("%.3g TB or more", target_gb/1000),
               TRUE ~ sprintf("%d GB or more", target_gb)
            )
    )
}) |>
summarize(
    percentage=sum(percentage),
    .by = c(date, gb)
) |>
mutate(percentage=pmin(percentage, 1)) |>
mutate(label=ifelse(date==max(date, na.rm=T), sprintf('%s | %.3g%%', gb, percentage*100), NA), .by=gb) |>
ggplot(aes(x=date, y=percentage, color=gb)) +
geom_line(alpha=0.3) +
geom_smooth(se=F, span=0.5, linewidth=1) +
geom_label_repel(aes(label=label)) +
labs(title='Free Hard Drive Space') +
guides(color='none') +
scale_y_continuous(label=percent_format())

filter(df, grepl('VR ', category)) |>
filter(date > ymd('2020-01-01')) |>
mutate(
    name2=fct_lump_prop(name, w=percentage, prop=0.01)
) |>
summarize(
    percentage=sum(percentage),
    .by=c(date, name2)
) |>
mutate(label=ifelse(date==max(date, na.rm=T), sprintf('%s | %.2g%%', name2, percentage*100), NA), .by=name2) |>
ggplot(aes(x=date, y=percentage, color=name2)) +
geom_line(alpha=0.3) +
geom_smooth(se=F, span=0.5, linewidth=1) +
geom_label_repel(aes(label=label), max.overlaps = Inf) +
labs(title='VR Headsets') +
guides(color='none') +
scale_y_continuous(label=percent_format())

filter(df, grepl('Other', category)) |>
mutate(percentage=pmin(percentage, 1)) |>
mutate(label=ifelse(date==max(date, na.rm=T), sprintf('%s | %.2g%%', name, percentage*100), NA), .by=name) |>
ggplot(aes(x=date, y=percentage, color=name)) +
geom_line() +
guides(color='none') +
geom_label_repel(aes(label=label), max.overlaps = Inf) +
labs(title='Other settings') +
scale_y_continuous(label=percent_format())