Remotion bakeoff · 2026-07-22
Remotion on Apple Silicon beat small cloud containers on wall-clock.
We rendered identical Remotion 4.0.381 compositions locally on Apple Silicon and on Cloudflare Containers standard-1 (4 GiB / 0.5 vCPU). For Chromium-bound Remotion, a single M2 Max was about 19.5× faster than the measured CF live 3-minute job. M1 Airs stayed competitive on the long kinetic job.
remotion-power-bench · 1080p30 · local Chromium (not Remotion Lambda fan-out).Mac hosts: M2 Max 96 GB research laptop; two M1 Air 16 GB agents (Silver1 / Silver2).
Cloud: CF Containers live worker + Docker local shape matching standard-1 (
--cpus=0.5 --memory=4g).Source:
LONG-WORKLOAD-AND-CLOUD-CONCLUSION-2026-07-22.md and fleet summaries under benchmarks/results/.Short calibration (5–15 s video)
| Composition | Frames | M2 Max | M1 Silver1 | M1 Silver2 | CF live |
|---|---|---|---|---|---|
| LightMotion (5 s) | 150 | 3.47 s | 8.78 s | 8.66 s | 54.12 s |
| MediumMotion (10 s) | 300 | 5.98 s | 9.14 s | 8.54 s | - |
| HeavyMotion (15 s) | 450 | 32.62 s | 42.93 s | 40.87 s | - |
LightMotion alone: M2 Max was ~15.6× faster than CF live (3.47 s vs 54.12 s). HeavyMotion shows M1 Airs within roughly 25–30% of M2 Max - Remotion is Chromium-bound, so an M1 is not “too weak” for this motion-graphics class.
Sellable-length kinetic jobs
| Job | Output | M2 Max | M1 Silver1 | M1 Silver2 | CF live | Docker CF-shape |
|---|---|---|---|---|---|---|
| LongKinetic3m | 3:00 (5400 f) | 111.5 s | 133.7 s | 145.6 s | 2175.3 s | 6593 s PROJECTED |
| LongKinetic20m | 20:00 (36000 f) | 679.5 s | - | - | 14502 s PROJECTED | 43955 s PROJECTED |
Compute dollars (rate-card estimate, not invoice)
Cloudflare Containers Workers Paid published rates used in the research notes: memory $0.0000025 / GiB-s and CPU $0.000020 / vCPU-s. For standard-1 (4 GiB + 0.5 vCPU) that is about $0.00002 per wall-second if CPU is fully utilized for the whole wall time (upper bound).
| Path | Job | Wall used | Est. compute $ |
|---|---|---|---|
| CF live | LightMotion 5 s | 54.12 s | ~$0.00108 |
| CF live | LongKinetic3m | 2175.3 s measured | ~$0.0435 |
| CF live | LongKinetic20m | 14502 s PROJECTED | ~$0.290 PROJECTED |
| Mac package electricity | LongKinetic3m (ABS Wh) | 111.5 s | ~$0.00017 |
What this means for RenderMac
- Lead with faster Remotion wall-clock on Mac and macOS-native tooling, not “idle watts as crypto.”
- Do not claim a single Mac beats Remotion Lambda’s many-function fan-out on wall-clock; Lambda’s advantage is parallelism.
- M1 Air overnight supply is viable for Remotion-class jobs - the 3-minute kinetic stayed in the 2.2–2.4 minute band.
- Cloud containers still win on ops elasticity; Macs win when Chromium Remotion is the bottleneck.