What can your Mac do?
Real times for local logging.
ClipLogger's local engine runs entirely on your Mac, so your chip and memory set the speed, and nothing uploads while it works. This table shows how long one clip takes on the Mac you already own, and when it makes sense to hand a shoot to Rush instead.
Your Mac, per clip
Estimated time to fully log one busy clip with the default on-device model. Pick your chip and memory to highlight your row.
| Mac | Chip | Unified memory | Bandwidth | Per clip, est. | A 200-clip shoot |
|---|---|---|---|---|---|
| Mac Studiodesktop | M3 Ultra | 96–512 GB | 819 GB/s | 2–4 min | 8 hrs |
| Mac Pro · Mac Studiodesktop | M2 Ultra | 64–192 GB | 800 GB/s | 2–4 min | 8 hrs |
| MacBook Pro · Studiolaptop / desktop | M4 Max | 36–128 GB | 410 GB/s | 3–6 min | 13 hrs |
| MacBook Prolaptop | M3 Max | 36–128 GB | 400 GB/s | 4–7 min | 16 hrs |
| MacBook Pro · Studiolaptop / desktop | M1 / M2 Max | 32–96 GB | 400 GB/s | 5–9 min | 20 hrs |
| MacBook Pro · Mac minilaptop / desktop | M4 Pro | 24–64 GB † | 273 GB/s | 6–11 min | 25 hrs |
| MacBook Pro · Mac minilaptop / desktop | M2 / M3 Pro | 18–36 GB † | 150–200 GB/s | 8–14 min | 33 hrs |
| MacBook Air · iMac · minieveryday Macs | M1–M4 (base) | 16–32 GB † | 100–120 GB/s | 14–25 min | 2 days |
When a deadline lands, send it to Rush.
A 200-clip shoot runs from an afternoon to overnight on your own Mac. Rush splits the same job across many GPUs in parallel and logs 500 clips in about 5 minutes, so the local engine covers the quiet nights and Rush covers the deadline. Rush frames are retained after processing; there is no fixed deletion window. See our Privacy Policy and Terms for retention details and deletion requests.
The math behind one clip
Where those minutes go when the model reads one clip.
The default is a 30-billion-parameter vision model quantized to 4-bit, which takes about 21 GB of memory, and the number of frames it reads per clip keeps going up as the models improve, so read these numbers as today's floor.
Where the time goes
Reading the frames
Pushing about 40,000 image tokens through the model takes most of the time, and it scales with your GPU cores.
Writing the log
Generating under 400 tokens of tags, names and notes is quick, and it scales with memory bandwidth.
So GPU cores and memory bandwidth both matter, and more memory is what lets the full model load at all.
These are modeled estimates, not lab benchmarks, and real times move with clip resolution, how many frames the model samples, the quantization level and thermal headroom, since laptops throttle under a long batch and desktops do not. They are anchored to published Apple silicon inference numbers and a Qwen-VL-72B run on an M3 Ultra, which took about 3.5 minutes for one full-resolution image, so read them as order of magnitude.