Disclosure: I'm the developer of EWRE, so I have an obvious bias. I'm sharing the setup, the raw numbers, and the limits of the test so you can judge them yourself.
The question
A live wallpaper runs all day and nobody stares at it. So how much should it cost? I compared EWRE (my engine) with Lively Wallpaper, a popular free one, playing the same video on the same machine.
Test setup
| OS | Windows 11 |
| GPU | NVIDIA GeForce GTX 1080 Ti (driver 582.28) |
| Video |
dem.mp4, 1920x1080 @ ~60 FPS, H.264 |
| Method | Sampled every 1 s for 30 s of continuous desktop playback |
| Versions | EWRE (current build) vs Lively v20262.5.6 |
How they're built
- EWRE: one native process (C++20, Direct3D 11, Media Foundation). Playback happens in-process, with hardware decoding.
-
Lively: C# on .NET 9 with WinUI 3. Video is played by an
mpv.exechild process. During the test I measured three processes:Lively.exe,mpv.exe, andLively.Watchdog.exe.
Results
| Metric | EWRE | Lively | Difference |
|---|---|---|---|
| Processes | 1 | 3 | |
| Average CPU | 2.65% | 6.46% | 59.0% lower |
| Peak CPU (1 s sample) | 9.30% | 18.70% | 50.3% lower |
| Average RAM (working set) | 98.70 MB | 324.29 MB | 69.6% lower (about 226 MB) |
| Peak RAM (working set) | 101.16 MB | 331.71 MB | 69.5% lower |
Where Lively's resources go
| Process | RAM | CPU |
|---|---|---|
Lively.exe |
~157 MB | mostly 0 to 2% |
mpv.exe |
~146 to 159 MB | most of the CPU, with bursts up to 18.7% |
Lively.Watchdog.exe |
~15.5 MB | ~0% |
Lively's CPU wasn't evenly spread. There were two sustained bursts of 15 to 19% (around seconds 18 to 22 and 28 to 29), almost entirely from mpv.exe. EWRE's worst single sample was 9.3%.
Why might EWRE be lighter?
I didn't profile this, so these are hypotheses based on the architecture:
- No IPC. Playback is in-process, with no separate player to coordinate with.
- No managed runtime. Native code means no .NET or WinUI footprint.
- Hardware decoding straight into GPU textures through Direct3D 11 and Media Foundation, which keeps CPU work low.
What this test does NOT show
I'd rather you hear this from me:
- One run, one machine, one video, 30 seconds. Treat it as a data point, not a verdict. Longer runs and other hardware could look different.
- GPU and VRAM usage weren't measured. Hardware decoding moves work to the GPU, so a full comparison needs GPU load, VRAM, and ideally power draw.
- RAM is working set, summed across Lively's processes. That can overcount pages shared between them, so the real difference may be smaller than 226 MB.
- CPU samples are coarse. Values move in steps of about 1.56%, which matches Windows' 15.6 ms CPU-time tick over a 1 s window, so small differences are noise.
- Lively's player settings matter. [State here whether Lively ran on default settings and whether mpv hardware decoding was enabled.]
- 30 seconds can't prove there are no memory leaks. EWRE's RAM stayed between about 93 and 101 MB, but a proper leak test needs hours.
- Lively does more than play video. This only compares video playback.
Reproduce it
The raw per-second data for both engines is on GitHub: [link to benchmark folder]. If you run the same test on different hardware, I'd really like to see your numbers.
Try it
EWRE is free on the Microsoft Store. Feedback and bug reports are welcome on GitHub.
What would you want measured next: GPU usage, battery drain, or a long-session leak test? 👇
Top comments (0)