Should Remotion run native or in a Linux VM?
Native macOS was about three times faster than the tested four-vCPU Linux paths on the M2 Max.
Open virtualization benchmark →Measured bakeoffs · updated 2026-07-25
Start with the answer. Drill down to every measurement.
This is the front door to RenderMac’s Remotion, virtualization, encoding, cost, and power research. Each report opens with the decision it supports, then publishes the exact setup, full tables, claim limits, and available raw evidence.
Reading rule: green “measured” labels are completed trials. Amber “mixed” reports contain a clearly marked projection, rate-card estimate, or absolute package-power calculation alongside measured data.
New · Apple WWDC26 virtualization
Container machine led the Linux options. Native macOS still rendered about 3× faster.
We ran 36 measured Remotion renders across native macOS, Colima VZ, Apple’s per-container VM, and the new persistent Container machine. The report includes every trial, startup timings, media-acceleration probes, limitations, and five raw JSON downloads.
What are you trying to decide?
How does Apple Silicon compare with a small cloud container?
The measured 3-minute M2 Max render finished in 111.5 seconds; the live Cloudflare Containers run took 2,175 seconds.
Open Mac vs cloud report →Is an M1 Air useful, or do I need an M2 Max?
The fleet report compares the same Light, Medium, Heavy, and long kinetic compositions across three Macs.
Open M1 vs M2 Max report →Does VideoToolbox materially change encode time?
The measured 20-minute 1080p H.264 encode on M2 Max VideoToolbox completed in 95.1 seconds.
Open encoding report →What does a render draw on the Mac?
The power report publishes SoC package watts and absolute watt-hours with the exact electricity math.
Open power report →What do the time differences mean in dollars?
The cost view separates measured Mac package electricity from Cloudflare’s published rate multiplied by wall-clock.
Open cost comparison →Headline results, side by side
These are orientation numbers, not substitutes for the linked reports.
3 min Remotion · M2 Max111.5 sMeasured
Same job · CF live2,175 sMeasured
20 min VT encode · M2 Max95.1 sMeasured
Heavy Remotion · M2 Max SoC~33 WPackage sample
Native HeavyMotion30.753 sMeasured median
Container machine HeavyMotion95.018 sMeasured median
Mac electricity · 3 min~$0.00017ABS Wh estimate
CF rate-card · same job~$0.0435Rate-card estimate
Two comparisons that are easy to misread
3 min Remotion · Mac
111.5 s
3 min Remotion · CF
2,175 s
The cloud row is the measured live standard-1-shaped run, not every possible cloud configuration and not Remotion Lambda fan-out.
Complete benchmark library
Measured completed trials Mixed measured data plus estimates or projections Raw data downloadable machine-readable results
| Report | Primary question | Evidence | Raw / details |
|---|---|---|---|
| Apple Container machine vs native | What does virtualization cost for Remotion? | 36 measured renders | 5 JSON files |
| Remotion: Apple Silicon vs cloud | How do tested Macs compare with small containers? | Measured + 20 min projection | Host, composition, wall time, cost labels |
| M1 Air vs M2 Max fleet | How much does Mac generation and class matter? | Measured | Light, Medium, Heavy, long kinetic |
| VideoToolbox vs software x264 | What changes when the hardware encoder is available? | Measured | 20 min, 1080p30, codec paths |
| Mac render power draw | What package power and energy were observed? | Measured package samples | Watts, ABS Wh, electricity math |
| Big-job cost comparison | How do energy and container rate-card math differ? | Measured + estimates | 3 min measured; 20 min CF projected |
| Benchmark methodology | How were workloads controlled and claims labeled? | Definitions and limits | Reproduction and interpretation rules |
More workload-specific benchmark views
3-minute Remotion · 20-minute Remotion · 5-second cold-start workload · M1 Air speed · M2 Max speed · Mac vs CF wall-clock · Cloudflare Containers cost · full research library.
How to read any RenderMac benchmark
Measured
The named job completed on the named host. Wall-clock and the available output checks were recorded.
Projected
A visible extrapolation from a measured result. It is never presented as a completed run.
Rate-card estimate
A provider’s published compute rate multiplied by wall-clock—not an invoice or a RenderMac price.
ABS package energy
macOS SoC energy including baseline draw when no idle-subtracted sample is available—not wall-outlet power.
Comparable scope
Same workload and controlled resources where possible. OS/browser differences and unlike scaling models are called out.
Claim boundary
Each report says what the test supports, what it does not support, and what was not measured.
The shortest honest conclusion: Apple Silicon performed very well for these Chromium and media workloads. The results do not prove one Mac beats every cloud configuration; they show which native paths deserve direct testing before defaulting to small Linux containers.