docs: republish the Linux benchmarks with ghostty's release and felis's display-rate painting #383

Merged

natsukium merged 2 commits from docs/republish-linux-benchmarks into main 2026-09-24 23:04:10 +09:00

Conversation 0 Commits 2 Files changed 22 +223 -171

natsukium commented 2026-09-24 21:41:56 +09:00

Owner

Copy link

Stacked on #382 (the macOS republish); only the last commit belongs to this PR. Merge #382 first.

The published Linux run (75d08630) predates two things that decide how its figures read: felis painted about 15 frames per second under DOOM-fire, so its frame count carried a "not comparable" caveat, and the field had no ghostty release, so ghostty-tip's ~1 s keystroke latency could not be shown to be tip-only. This republishes the Linux section from a fresh three-round run at 9d1d842f (v0.1.0-rc.6 plus the daemon's flood fixes, #375 and #380).

• ghostty 1.3.1 beside the tip. 1.3.1 measures 7.0 ms mean latency against the tip's 1,064 ms, so the regression is the tip's alone.
• DOOM-fire is ranked. The measured build paints at the display rate, so the non-comparability note is gone; felis leads at 2,632 fps (foot 2,324).
• Sources follow the bench shell: kitty 0.49.0 from nix develop .#bench (#381), ghostty-tip at the commit locked in dev/flake.lock (7c40388), which also retires the "daily build" limit.

Kept honest on the page:

• The CI runner and the Windows VM stayed up, so the load average started at 3.6 (1.3 last time); ghostty-tip's vtebench legs raised it to 8–12 and foot's following legs ran under that.
• On the non-fullscreen scrolling rows felis varies 25–50% between rounds; the order is unresolved on three rows, and ghostty-tip is resolved ahead on the two top-region rows.
• ghostty 1.3.1 caps scrollback by bytes only, so its flood-memory row is not held to the 10,000-line depth.
• GPU terminals' PSS is 54–64 MB lower than in the morning run while foot's barely moved (shared driver pages); the page already says to compare within one run.

Not investigated: felis's time to first window is 236 ms cold / 174 ms warm, against 197 / 154 in the previous run, while the other terminals moved by 0–8 ms. Load explains part of it at most; the page publishes the figures as measured.

Verified: every leg completed in all three rounds (ghostty-tip's round-1 latency leg passed on the harness's automatic retry); just prose-check clean; treefmt applied.

Stacked on #382 (the macOS republish); only the last commit belongs to this PR. Merge #382 first. The published Linux run (`75d08630`) predates two things that decide how its figures read: felis painted about 15 frames per second under DOOM-fire, so its frame count carried a "not comparable" caveat, and the field had no ghostty release, so ghostty-tip's ~1 s keystroke latency could not be shown to be tip-only. This republishes the Linux section from a fresh three-round run at `9d1d842f` (v0.1.0-rc.6 plus the daemon's flood fixes, #375 and #380). - **ghostty 1.3.1 beside the tip.** 1.3.1 measures 7.0 ms mean latency against the tip's 1,064 ms, so the regression is the tip's alone. - **DOOM-fire is ranked.** The measured build paints at the display rate, so the non-comparability note is gone; felis leads at 2,632 fps (foot 2,324). - **Sources follow the bench shell**: kitty 0.49.0 from `nix develop .#bench` (#381), ghostty-tip at the commit locked in `dev/flake.lock` (`7c40388`), which also retires the "daily build" limit. Kept honest on the page: - The CI runner and the Windows VM stayed up, so the load average started at 3.6 (1.3 last time); ghostty-tip's vtebench legs raised it to 8–12 and foot's following legs ran under that. - On the non-fullscreen scrolling rows felis varies 25–50% between rounds; the order is unresolved on three rows, and ghostty-tip is resolved ahead on the two top-region rows. - ghostty 1.3.1 caps scrollback by bytes only, so its flood-memory row is not held to the 10,000-line depth. - GPU terminals' PSS is 54–64 MB lower than in the morning run while foot's barely moved (shared driver pages); the page already says to compare within one run. Not investigated: felis's time to first window is 236 ms cold / 174 ms warm, against 197 / 154 in the previous run, while the other terminals moved by 0–8 ms. Load explains part of it at most; the page publishes the figures as measured. Verified: every leg completed in all three rounds (ghostty-tip's round-1 latency leg passed on the harness's automatic retry); `just prose-check` clean; treefmt applied.

natsukium added 2 commits 2026-09-24 21:41:57 +09:00

docs: republish the macOS benchmarks with ghostty's release and resource use ... 7a1b054143

The published macOS results measured only ghostty's tip, whose throughput
is not what the 1.3.1 release ships, and predated the harness that records
CPU time and peak memory per workload. A new three-round run on the same
machine, with felis carrying the daemon's flood fixes, now reports both
ghostty builds, time to first window for every terminal (ghostty-tip
included, which the old exit-timed startup suite could not measure), and
the resource tables the Linux section already has.

Three findings keep the page honest. felis and alacritty share the lowest
mean latency and their ranges overlap, so no winner is claimed. felis
(cold) reaches its first window before felis (warm) in every round, which
this run does not explain. And every terminal's peak footprint while
drawing is far above its idle figure (felis's client alone holds 430 to
460 MB under load), so the idle table is not a working-set figure.

Assisted-by: Claude Code:Opus 5.5

docs: republish the Linux benchmarks with ghostty's release and felis's display-rate painting ... da80202f68

The published Linux run predates two changes that decide how its figures
read. felis painted about 15 frames per second under DOOM-fire, so its
frame count was marked not comparable; the build measured here includes
the fix that paints at the display rate, so that caveat is gone and the
figure is ranked like the others. And the field had no ghostty release,
so ghostty-tip's 1 s keystroke latency could not be shown to be a tip
regression; the new ghostty column measures 1.3.1 at 7 ms beside it.

kitty now comes from the bench shell, and ghostty-tip is the commit
pinned in dev/flake.lock rather than a daily build, so the page names
them that way and drops the daily-build limit.

The CI runner and the Windows VM stayed up this time, so the load
average started at 3.6 rather than 1.3; the limits section states it.
PSS for every GPU terminal fell by 54 to 64 MB against the morning run
while foot's barely moved, the same shared-driver-page effect the page
already warns about.

Assisted-by: Claude Code

natsukium merged commit 956d8378e4 into main 2026-09-24 23:04:10 +09:00

natsukium deleted branch docs/republish-linux-benchmarks 2026-09-24 23:04:10 +09:00

natsukium referenced this pull request from a commit 2026-09-24 23:04:12 +09:00

docs: republish the Linux benchmarks with ghostty's release and felis's display-rate painting (#383)

Commenting is not possible because the repository is archived.

Reviewers

No reviewers

Labels

Clear labels   

priority/P0

Must be resolved before the v0.1.0 tag

  

priority/P1

Must be settled before the first public contract freezes

  

priority/P2

Explicitly deferred until after v0.1.0

  

release/v0.1.0

Work tracked for the first tagged release

  

status/blocked

Cannot start now: waits on an upstream release, another issue, or a recorded revisit trigger

  

status/planned

Triaged with a source-grounded plan and a passed pi review; ready to implement

  

type/bug

Shipped behavior that contradicts the docs or the code's own intent; needs a product code change

  

type/design

An open scope or trade-off question; needs a recorded decision in the owning explanation doc before code

  

type/test-gap

Documented or reviewed behavior that no test pins; usually split from a review finding, needs no product code change

  

type/tracker

Parent issue that tracks child work

No labels

priority/P0

priority/P1

priority/P2

release/v0.1.0

status/blocked

status/planned

type/bug

type/design

type/test-gap

type/tracker

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

1 participant

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference

natsukium/felis!383

No description provided.