You are not logged in.
I recently set up two Precision 7740s with identical specs, one with Arch (Linux 7.1.3-arch1-3) and one with Fedora (kernel-7.1.3-201.fc44). I noticed a non-trivial GIMP performance discrepancy between them.
Exporting a 5555x5555px PNG containing noise, the Fedora one is ~300% faster.
Fedora: 8.23s
Arch: 31.8s
Sysbench, Geekbench, hdparm -Tt and Phoronix Blender results are near-identical between the systems.
Has anyone else seen this as a general tendency?
Does your system bench closer to the 8s or the 32s?
Could there be specific tweaks in the Fedora-issued kernel optimizing for GIMP or more general file operations to account for this massive gap?
These two bad boys happen to be identically specced, but I run this export as part of a tiny benchmarking suite and I've seen the same ballpark gap between countless systems (even ARM ones).
SPECS:
CPU: Intel i9-9880H
RAM: 64GB @ 2666MHz
SSD: Micron 2200S NVMe 1024GB
GPU1: Intel UHD 630
GPU2: Quadro RTX 3000 Max-Q, Driver 580.173.02, Prime Render Offload
WM: Ly v1.4.0, Hyprland v0.56.2
Compositor: Wayland
CPU Governor: Performance
Power management: TLP v1.10.1 (set to AC mode)
GIMP version: 3.2.4-2
GIMP settings are identical (Tile Cache 31GB, OpenCL off)
TO REPRODUCE:
- Make sure Tile Cache in GIMP is set to high (1GB under system RAM if lower than 32GB - it seems to not matter much unless it's really low)
- OpenCL should default to off, turn off if not (you know if you changed this, ignore if not)
- Make a new 5555x5555 document
- Do Filters -> Render -> Noise -> Cell Noise (default settings)
- Export as .png with default settings (start timer after second settings dialog)
- Stop timer at write complete
Any feedback greatly appreciated. I apologize for this being a cross-community question. I figured our community might find it interesteing.
Last edited by spinus777 (2026-08-14 18:34:03)
for (int x = 0; x < SAMPLE_RATE; x++) { output[x] = sin(2 * M_PI * frequency * x / SAMPLE_RATE); }
Offline
fedora seems to unconditionally use https://archlinux.org/packages/extra/x8 … ng-compat/
Offline
Indeed it does.
After installing it (replacing zlib-1:1.3.2-3 with zlib-ng-compat-2.3.3-1) and restarting GIMP, the export times are aligned. No linking needed. Thank you for that expert tidbit!
Are there any drawbacks to deploying those optimizations? Or is it just that older hardware can't use them, so it makes for a better vanilla configuration?
for (int x = 0; x < SAMPLE_RATE; x++) { output[x] = sin(2 * M_PI * frequency * x / SAMPLE_RATE); }
Offline
Afaict it even SSE2 is runtime-optional, it's not "the" zlib and there's always a chance to run into odd compatibility issues if the client panders to quirks in the original zlib.
A lot of users will use it, though, because git hard-depends on it.
Please always remember to mark resolved threads by editing your initial posts subject - so others will know that there's no task left, but maybe a solution to find.
Thanks.
Offline
@spinus777 some Arch history on this topic:
• https://archlinux.org/todo/zlib-ng-migration/
• https://lists.archlinux.org/archives/li … GVEHPSTKDZ
Offline