I use terminals a lot. On most machines, I have anywhere from five to ten of them open across one to four desktops (there are 12 terminals open on two desktops on the machine I'm writing on). Having started using Linux in 1996, I used xterm because it was the baseline. I found rxvt around 1998 and used that for a few years. I'm not sure how it went from there, but about three years ago I was mostly using lxterminal but switched to alacritty because it did most things right and worked in both X11 and Wayland (I'm stuck between those two worlds). However, a little less than a year ago I discovered kitty which was very appealing because it ticked the same boxes as alacritty but also supported pixel-perfect rendering of photos IN THE TERMINAL (using timg - https://github.com/hzeller/timg) ... which is very handy when you're SSHed to a remote web server. I switched most of my machines to kitty.

About six weeks ago, I noticed that one of my machines was incredibly sluggish because it was drowning in swap. It's a relatively old machine with an i7 processor and a spinning disk ... but it had 16G of memory and I wasn't doing anything memory-intensive like video editing or running a large database. This provided a learning opportunity, and I discovered smem (https://www.selenic.com/smem/ ... also available in both the Fedora and Debian repositories). It reports the memory usage of processes (including USS, PSS - which smem claims is a better way of looking at memory usage, RSS, and Swap). And what I immediately found was that ten instances of kitty had chewed through all my swap. 50G of swap. (Why I had that much on the machine I don't know - probably some failed or forgotten experiment.) I killed all instances of kitty and replaced them with ten instances of xterm and the 50G of swap that had been full was completely empty.

xterm isn't particularly beautiful, but I don't use the stuff other people would miss like tabs and menus. I miss transparency a bit, but it's not a big deal. I'll keep kitty around because of timg, although that's a relatively rare use-case.

I started switching most of my machines back to alacritty, and wondered what I should use in the long term. This inspired me to devise a test: I would run every terminal I was considering all at once - better yet, one set on Wayland and one set on X11 - and then run the same set of commands in every terminal. And measure the memory usage.

Openbox / X11

Before the Commands Were Run

      PID Command                         Swap      USS      PSS      RSS
    115733 st                                 0     2592     3008    12788
    115892 xterm                              0     6808     7414    17732
    115207 lxterminal                         0     6528     9891    41380
    114856 /usr/libexec/gnome-terminal        0    10252    14984    52836
    115034 /usr/bin/kitten __atexit__         0    19772    19783    21168
    114615 alacritty                          0    34224    43478    98396
    131279 konsole                            0    35108    50308   101220
    115005 kitty                              0    56376    66584   129180
    115381 ptyxis                             0    54760    67603   144640
    

After the Commands Were Run

      PID Command                         Swap      USS      PSS      RSS
    115733 st                                 0     3396     3834    13780
    115207 lxterminal                         0     7800    11159    42656
    115892 xterm                              0    11640    12266    22748
    114856 /usr/libexec/gnome-terminal        0    10692    15416    53340
    115034 /usr/bin/kitten __atexit__         0    21972    21983    23368
    114615 alacritty                          0    40188    49444   104408
    131279 konsole                            0    38528    54193   106256
    115381 ptyxis                             0    58252    71107   148316
    115005 kitty                              0    70716    80917   143568
    

KDE / Wayland

Before the Commands Were Run

      PID Command                         Swap      USS      PSS      RSS
    30975 st                                 0     2924     3374    12832
    31112 xterm                              0     6460     6691    13204
     7905 foot                               0     5096     7110    17296
    30850 rxvt                               0     8936     9440    19248
    29538 lxterminal                         0     9200    14622    47240
    29259 /usr/libexec/gnome-terminal        0    12168    19005    53420
    29417 /usr/bin/kitten __atexit__         0    22852    22864    24464
    29098 alacritty                          0    25468    28837    73416
    43371 /usr/bin/konsole                   0    27348    39068   130252
    29401 kitty                              0    46468    51520   105816
    29690 ptyxis                             0    73220    79204   141012
    

After the Commands Were Run

      PID Command                         Swap      USS      PSS      RSS
    30975 st                                 0     3204     3820    13836
    31112 xterm                              0     9888    10119    16632
     7905 foot                               0     8764    11114    21704
    30850 rxvt                               0    12872    13736    27096
    29538 lxterminal                         0     9824    15343    48804
    29259 /usr/libexec/gnome-terminal        0    12660    19313    54056
    29417 /usr/bin/kitten __atexit__         0    22884    22896    24496
    29098 alacritty                          0    34188    37470    82580
    43371 /usr/bin/konsole                   0    33244    48094   145496
    29401 kitty                              0    53048    58032   112936
    29690 ptyxis                             0    75060    81130   143284
    

You may notice some anomalies: rxvt wasn't measured on one machine. The controls were poor all around: terminals were different sizes, had different scrollback buffers, etc. But since the results above show a spread with the memory usage of the smallest being less than one tenth that of the largest, we're not measuring subtle differences here. So I'm not too concerned.

The results are really interesting:

• I didn't think I would find a terminal with less memory use than xterm - but st beats it by a noticeable margin
• I didn't think I would find a terminal that was more of memory hog than kitty (although if you add accessory app kitten's memory usage, maybe it still "wins")
• I expected gnome-terminal's memory usage to be in line with konsole (KDE's default terminal), but gnome-terminal shows remarkably well in this test
• the environment (Fedora vs Debian, Wayland vs X11) doesn't seem to have made much difference at all
• I was surprised to find that konsole supports pixel-perfect images from timg. timg was the original reason I switched to kitty and the list of terminals that support timg at pixel level is short. The others that I'm aware of are foot (tested under Wayland only as it doesn't work in X11) and ghostty (not tested because I'm sticking to terminals in the Debian repos). foot also seems to prove that you can have reasonable memory usage and support timg - good to know.

I think I'm going to try out st some more. Not that my machines are memory-constrained, but if it meets my other needs, why not use the least memory-hungry terminal? Since I commonly run ten terminals, it adds up. If st doesn't work for me, I'm going to try out gnome-terminal.

Update(s)

2026-05-08: st turns out to have NO scrollback. I'm now leaning to lxterminal for X11, and foot for Wayland.

I should have noted: terminals compiled for X11 run in Wayland just fine because there's an X11 compatibility layer. There's no such thing going the other way, so Wayland-native terminals (like foot) don't run under X11.