Open a raw file
Decode, GPU upload, demosaic and pyramid, for a 24–26 MP raw
70–250 ms
From the README, 29 Sept 2026.
Performance
How fast Redlamp is, measured
Redlamp is built to answer every slider within a frame. These are its measurements over time, on an Apple M1 Ultra with a Release build: the figures the README recorded as the work landed, and the runs its benchmark harness records now. A metric counts as faster or slower only when it moves by more than 10%, or by more than its runs' own spread.
- Open a raw fileFrom the README, 29 Sept 2026
- 70–250 ms
- Render at Fit while editingFrom the README, 29 Sept 2026
- 0.6–3 ms
- Render the whole photo at 1:1From the README, 29 Sept 2026
- 13 ms
- Full-resolution export renderFrom the README, 29 Sept 2026
- 45 ms
- Slowest 5% of iterations while draggingFrom the README, 30 Sept 2026
- 5 ms
- All 50,000 photos listedFrom the README, 2 Oct 2026, under load
- 209 ms
Against a display's frame
What a slider costs while you drag it, and the whole photo at 1:1, beside one frame of a 120 Hz and a 60 Hz display.
Render at Fit with two gradient masks1.3 ms
Render at Fit while editing1.8 ms
Dragging sharpening's Amount at 1:12.2 ms
Dragging noise reduction's Luminance at 1:14.3 ms
Dragging sharpening's Radius at 1:15.5 ms
Default sharpening for a 1:1 view, first render7.2 ms
Render the whole photo at 1:113 ms
What's moved since it was first measured
Each metric's latest figure against its first, from the same source on the same machine. Some work makes a step slower on purpose, for better results; the change that did it is named beside each.
Getting faster
- Slowest 5% of iterations while dragging97% faster
157 ms on 30 Sept 2026, 5 ms on 30 Sept 2026
1d243f3 Port the histogram and Tone Curve panel to AppKit
- Detail stage GPU memory, 60 MP at 1:169% less
7.5 GB on 4 Oct 2026, 2.3 GB on 4 Oct 2026
5bf4dc9 README and tracker: detail-stage memory bounded (AUD-03, #160)
- Main thread busy while dragging a slider66% faster
100% on 30 Sept 2026, 34% on 30 Sept 2026
1d243f3 Port the histogram and Tone Curve panel to AppKit
- Detail stage GPU memory, 24 MP at 1:158% less
3.1 GB on 4 Oct 2026, 1.3 GB on 4 Oct 2026
5bf4dc9 README and tracker: detail-stage memory bounded (AUD-03, #160)
- Default sharpening for a 1:1 view, first render15% faster
8.5 ms on 30 Sept 2026, 7.2 ms on 30 Sept 2026
0bf07c0 Faster Radius drags: precomputed blur weights, half-float deconvolution
Getting slower
Every metric over time
- As recorded in the README
- A harness run
- Measured under load (drawn, never compared)
- A range: the README's, or the fastest to slowest sample file
Opening
Editing
Render at Fit while editing
Every adjustment in one fused pass, at a 2560 × 1600 view
0.6–3 ms
From the README, 29 Sept 2026.
Render at Fit with two gradient masks
A linear and a radial mask, each with local adjustments
1.3 ms
From the README, 29 Sept 2026.
Render the whole photo at 1:1
The full frame at full resolution, as when zoomed in
13 ms
From the README, 29 Sept 2026.
Export
Full-resolution export render
Developed at full resolution, read back for encoding
45 ms
From the README, 29 Sept 2026.
Full-resolution export with noise reduction
Luminance 50, rendered in tiles
100 ms
From the README, 30 Sept 2026.
Detail
Detail stage at 1:1: noise reduction
GPU time on a 1:1 region of about 10 MP of pyramid texels
5 ms
One measurement so far.
From the README, 30 Sept 2026.
Detail stage at 1:1: Texture and Clarity
GPU time on the same region
1 ms
One measurement so far.
From the README, 30 Sept 2026.
Noise reduction for a 1:1 view
GPU time, Luminance 50, a 2560 × 1600 view of a 24 MP photo
3.6 ms
From the README, 2 Oct 2026; 33% slower than the record before.
Default sharpening for a 1:1 view, first render
GPU time, the same view
7.2 ms
From the README, 30 Sept 2026; 15% faster than the record before.
Dragging sharpening's Radius at 1:1
GPU time per frame, the same view
5.5 ms
One measurement so far.
From the README, 30 Sept 2026.
Dragging sharpening's Amount at 1:1
GPU time per frame, with the analysis cached
2.2 ms
From the README, 30 Sept 2026.
Dragging noise reduction's Luminance at 1:1
GPU time per frame, the same view
4.3 ms
From the README, 2 Oct 2026; 54% slower than the record before.
Focus stacking
Focus stack of 25 Canon CR3 frames
17 MP frames: decode, align, depth map and fusion
8 s
One measurement so far.
From the README, 30 Sept 2026.
Focus stack of 109 JPEG frames
4 MP frames
15 s
One measurement so far.
From the README, 30 Sept 2026.
Reopen a merged focus stack
From its cache
0.1 s
One measurement so far.
From the README, 30 Sept 2026.
Memory
Detail stage GPU memory, 24 MP at 1:1
With noise reduction and sharpening on
1.3 GB
From the README, 4 Oct 2026; 58% less than the record before.
Detail stage GPU memory, 60 MP at 1:1
With noise reduction and sharpening on
2.3 GB
From the README, 4 Oct 2026; 69% less than the record before.
Peak memory browsing 50,000 photos
The app's footprint while listing, decoding and warming thumbnails
317 MB
From the README, 2 Oct 2026, under load.
Folders
First photos in the filmstrip
50,000 photos in 500 folders, Show Photos in Subfolders on
14 ms
One measurement so far.
From the README, 2 Oct 2026, under load.
All 50,000 photos listed
500 folders, in parallel, in order
209 ms
One measurement so far.
From the README, 2 Oct 2026, under load.
Visible thumbnails from the raw files
The first screenful, decoded from the files
33 ms
From the README, 2 Oct 2026, under load.
Thumbnails warmed from the raw files
In the background, while you're not waiting
705 a second
From the README, 2 Oct 2026, under load.
Thumbnails from the disk cache
Read back from the per-folder cache
6,210 a second
One measurement so far.
From the README, 2 Oct 2026, under load.
Responsiveness
Main thread while listing and warming (p99)
The slowest 1% of run-loop iterations
0.15 ms
One measurement so far.
From the README, 2 Oct 2026, under load.
Main thread scrolling the filmstrip (p99)
End to end at 120 Hz
1.4 ms
One measurement so far.
From the README, 2 Oct 2026, under load.
Main thread busy while dragging a slider
120 events a second, every panel open
34%
From the README, 30 Sept 2026.
Typical run-loop iteration while dragging
The median
0.17 ms
From the README, 30 Sept 2026.
Slowest 5% of iterations while dragging
The 95th percentile
5 ms
From the README, 30 Sept 2026.
Slowest 1% of iterations while dragging
The 99th percentile
8 ms
From the README, 30 Sept 2026.
How it's measured
scripts/perf-record.shbuilds Release and runsredlamp benchon the CC0 sample raws from raw.pixls.us (each measurement repeated after a warm-up; a metric is the median across the files), then drags a slider at 120 events a second with every panel open, and opens 50,000 photos in 500 folders.- Runs recorded while the Mac was busy (a load average above 8) are kept and drawn but never compared, and the figures the README recorded (19 versions from 29 September) are compared only with each other: they were measured by hand, some as ranges.
- The latest harness run: 4 October 2026, commit a010ea1, Apple M1 Ultra with 128 GB, macOS 26.6.2, 7 runs per measurement, load average 27.17 before and 49.94 after, so it's recorded as noisy.
- No Lightroom figures: Redlamp hasn't measured Lightroom on the same photos.
Read from docs/performance/history.jsonl at commit 95a532f.