Ghostty is crazy slow, am I missing something?

[nazdridoy]

Ghostty was supposed to be as fast as other fast terminals, or "the fastest", but in every benchmark I tried, it was extremely slow compared to Alacritty and Kitty. Am I missing something?

Test A

kitten __benchmark__ 

Kitty

Ghostty

Test B

DOOM-fire-zig

kitty

Ghostty

Additional Info

OS: Archlinux

❯ ghostty --version
Ghostty 1.0.1-arch

Version
  - version: 1.0.1-arch
  - channel: tip
Build Config
  - Zig version: 0.13.0
  - build mode : builtin.OptimizeMode.ReleaseFast
  - app runtime: apprt.Runtime.gtk
  - font engine: font.main.Backend.fontconfig_freetype
  - renderer   : renderer.OpenGL
  - libxev     : main.Backend.io_uring
  - GTK version:
    build      : 4.16.12
    runtime    : 4.16.12
  - libadwaita : enabled
    build      : 1.6.3
    runtime    : 1.6.3
  - libX11     : enabled

[nazdridoy]

no response or update whatsoever, this is sad....

[rhodes-b]

There are many benchmarks cherry picking a few results is generally not super represenative
you are free to not use ghostty if the "results" aren't up to your standard

I ran the results of the kitty benchmark and only saw a major difference in the long escape codes and images.
Its also known thing that ghostty parsing kitty images is much slower than kitty, otherwise it was slightly behind kitty on my machine. Unsure on long escape codes probably work to be done there

It also seems you have old hardware on my newer laptop (not the most expensive out there ~1k usd at time of purchase) my ghostty results are 2x faster than your kitty results (except in the 2 categories listed above).

If you are using a prepackaged ghostty as well your results will also suffer a bit because of the zig baseline flag since zig tries to aggressively optimize to the current hardware platform. I believe most of ghosttys testing is generally not run with the baseline flag as they came in the non public release period where it wasnt packaged anywhere. That shouldnt have too much of a significant impact especially if your hardware is older. One that will have a impact is ghostty does detect SIMD stuff at runtime in which a older platform would suffer. (all my results are also on a prepackaged ghostty so no influence there either)

I also ran a DOOM-fire-zig with the Doptimize=ReleaseFastwhich had a HUGE difference in performance on both platforms

Terminal	zig build	zig build -Doptimize=ReleastFast
Ghostty	250fps	500fps
Kitty	450fps	600fps

suddenly ghostty is not too far behind kitty now (max fps is a diminishing curve so they are even closer in this case than the results look) and had a 2x difference when you actually tell zig to build in release mode

Most of the published benchmarks are from MacOS and not Linux afaik, where the MacOS side is definitely faster than the Linux side. Still shows that ghostty is "fast" like advertised (note im also on linux so no these are not the "faster" mac comparisons)

[mitchellh]

Ghostty was supposed to be as fast as other fast terminals, or "the fastest",

From my personal writing on the official Ghostty website:

In all categories, I am not trying to claim that Ghostty is the best (i.e. the fastest, most feature-rich, or most native). [...] I wanted to create a terminal that was competitive in all three categories and I believe Ghostty achieves that goal.

Ghostty is absolutely not the fastest. It's meant to be competitive and broadly speaking from benchmarks around the internet (not just mine), it is. Outliers will happen, and I'd love to learn more in those cases!

Your benchmarks are not indicative of normal results. @rhodes-b benchmark results look more indicative of normal, especially the Kitty benchmark results in particular. Our image throughput is definitely slower (and I've said that publicly many times!) and our long escape code throughput is slower, though that hasn't come up in practice much yet.

I'd say there's something interesting about your machine making the results different. I'd be interested in learning more but of course only you have the machine and I'm not sure if you're interested in doing the work to find out more. That's not a dig at it, just genuinely I don't know if you care and that's totally okay. 😄

If you find specific hotspots I'd be happy to address.

[nazdridoy]

@mitchellh Thanks for your thoughtful response. I totally understand Ghostty isn’t aiming to be “the fastest” terminal, and honestly, it doesn’t need to be. I’ve really been enjoying trying it out. It’s clear you’ve put a lot of thought and care into the project, and I appreciate it!

That said, I’ve noticed Ghostty’s performance on my machine is significantly slower compared to Kitty, and I’d love to help figure out what might be causing this.
My setup:

OS: Arch Linux Linux x86_64
Host: VivoBook_ASUSLaptop X512FJ_X512FJ 1.0
Kernel: 6.12.10-zen1-1-zen
Shell: zsh 5.9
Resolution: 1920x1080
DE: Plasma 6.2.5
WM: kwin
Terminal: ghostty
CPU: Intel i5-8265U (8) @ 3.900GHz
GPU: Intel WhiskeyLake-U GT2 [UHD Graphics 620]
GPU: NVIDIA GeForce MX230  (though the benchmark was run with the Intel iGPU, not the NVIDIA card).

I’m not sure if it’s something specific about my hardware or configuration, but I’m happy to help investigate further. If there are specific benchmarks, debug steps, or tweaks I can test, just let me know. I’d be glad to collect data or experiment to help pinpoint what’s going on.

Thanks again for all the work you’ve put into Ghostty.

[mitchellh]

Thanks @nazdridoy!

I think the best way to pinpoint specific issues is to get profiles/traces of hotspots and see what's going on on your system and if the general shape of your runtime is the same as other systems or if anything stands out on yours.

I know the Kitty benchmarks you can isolate a single benchmark. So just to start to make things easy for example: we should be doing pretty well on the Unicode benchmark for Kitty. On my machine and every other our Unicode processing is actually fastest (I don't want to make the general claim because I don't like the absolutes) but its the one benchmark across multiple benchmarks I've seen Ghostty clearly and easily outdo other terminals. So, it's a good one to focus in on in your case.

If you can run that one while grabbing a CPU profile then it might give me something to work with. (Note, I only need Ghostty's profile, grabbing a profile for Kitty/Alacritty/whatever is not helpful in this case, we're just trying to isolate the point of performance in Ghostty.)

I have to admit that I don't normally profile on Linux because I find the profiling tools harder to use, so I don't have good instructions on how to do this. I'm sorry.

[nazdridoy]

Hi @mitchellh,

I'm sorry for the late reply; I got caught up with a personal project.

As you suggested, I ran the Unicode benchmark from kitten __benchmark__ on Ghostty and captured a CPU profile. Here’s a quick summary of what I did:

1. I used perf to record the performance data while running the Unicode benchmark:

   perf record -g ghostty

   Followed by running the benchmark:

   kitten __benchmark__ unicode

2. I generated both the raw perf.data file and a readable report (perf_report.txt) for analysis.

I’ve attached both files with this message:

• perf.data
• perf_report.txt

Let me know if there’s anything else you’d like me to try or if further profiling is needed. I’m happy to assist!

[rhodes-b]

Our traces look pretty different notably i dont have io-reader or that first large 9% trace for some gtk function
note I ran perf record -g ghostty then kitten __benchmark__ unicode exited then perf report --stdio > perf_data.txt
perf_data.txt

I also saw you are using the zen kernel, I think that could be a factor, not sure how easy it is to swap to test another kernel on arch but if you could try I think it might help.
Looks like it changes the scheduler specifically it seems it has a much lower quantum than the normal kernel. Context switching is a very expensive operation as far as the kernel is concerned in which a terminal will be doing a bunch due to it being very I/O focused it also changes I/O operations but I'm less knowledgeable about what influences that might have)

https://github.com/zen-kernel/zen-kernel/wiki/Detailed-Feature-List

[nazdridoy]

@rhodes-b

ok i've tried with release kernel, and here is the cpu profile.
perf_data.txt

---

systeminfo:
OS: Arch Linux Linux x86_64
Host: VivoBook_ASUSLaptop X512FJ_X512FJ 1.0
Kernel: 6.12.10-arch1-1
Shell: zsh 5.9
Resolution: 1920x1080
DE: Plasma 6.2.5
WM: kwin
Terminal: ghostty
CPU: Intel i5-8265U (8) @ 3.900GHz
GPU: Intel WhiskeyLake-U GT2 [UHD Graphics 620]
GPU: NVIDIA GeForce MX230 (though the benchmark was run with the Intel iGPU, not the NVIDIA card).