Skip to content

How it compares

On Apple platforms the choice is usually between two kinds of player:

  • AVPlayer: deep OS integration, but only the formats Apple ships.
  • A VLC- or mpv-based engine: plays almost anything, but renders its own frames and bypasses the system’s Dolby Vision, Atmos and HDR handling.

AetherEngine aims for both. It puts FFmpeg’s format coverage on top of VideoToolbox and AVPlayer, so Dolby Vision, Atmos and Match Content (the TV’s automatic HDR and frame-rate switch) keep working.

  • Whatever AVPlayer can carry plays natively: hardware decode, Dolby Vision display, Atmos passthrough.
  • Everything else plays on the engine’s own software path (dav1d, libavcodec).

KSPlayer is the closest alternative. It reaches the same result through the same AVPlayer route, but it ships as a full player with its own UI. Its free build is GPL. A paid LGPL tier unlocks AV1 hardware decoding, the full demuxer and decoder set, and a list of playback features its own feature matrix names.

AetherEngine is an engine you embed and drive from your own SwiftUI, with that codec and HDR coverage in the open-source core.

AetherEngineKSPlayerAVPlayerVLCKitlibmpv
ApproachEmbeddable engine, Apple-onlyFull player with bundled UI, FFmpeg + AVPlayer, Apple-onlyApple’s built-in playerlibVLC wrapped for Applelibmpv, cross-platform
Container & codec breadthWide, FFmpeg demuxWide, FFmpeg demuxNarrow, Apple’s setWideWide
Hardware decodeVideoToolbox, dav1d / libavcodec softwareVideoToolbox, FFmpeg SW fallbackVideoToolboxVideoToolbox plus softwareVideoToolbox plus software
Dolby VisionP5, P7 as 8.1 (with the FEL folded into its curves, or composed dual-layer, both optional), P8.1, P8.4, P20 (3D on visionOS, 2D elsewhere), AV1 P10.x, real display switchP5, P8 via AVPlayer; their matrix lists P5 HDR display without overheating as paid LGPLP5 and P8.1 onlyTone-maps, no DV displayTone-maps, no DV display
Dolby AtmosEAC3+JOC stream-copied (HDMI MAT, spatial)EAC3+JOC via AVPlayerEAC3+JOC passthroughDecodes to PCM, no object passthroughNo Atmos passthrough on Apple
TrueHD / DTS:X heightsSpeaker bed up to 9.1.6 built from the stream’s lossless 7.1, steered by its objects or heights, delivered as lossless 24-bit PCM, APAC on AirPlay (optional components)7.1 core (FFmpeg decode)Not played7.1 core7.1 core
HDR on tvOSNative Match Content switchNative Match Content on AVPlayer path, else Metal tone-mapNative Match ContentSoftware tone-mappingSoftware tone-mapping
Rendering & UIOS-native, you ship SwiftUIOwn Metal renderer, bundled controlsOS-native, you ship UIOwn renderer, bundled controlsOwn renderer, bundled OSC
Apple TV / App StoreYes, LGPL plus store exceptionFree build GPL, paid LGPL tier for AV1 hardware decode and the full decoder setYesYes, LGPLNot practical, GPL, no tvOS

Short version: all four engines that open Matroska delivered every frame. AetherEngine and KSPlayer drew about 40 mW of GPU power; VLCKit about 6 times that, libmpv more than 20 times. AVPlayer does not open Matroska.

These numbers are a snapshot taken on 2026-09-24 with AetherEngine 7.15.2. Later releases are not re-measured here. The test files are 4K HDR HEVC in Matroska, the container media servers actually serve, at 38 Mbit/s and at 90 Mbit/s (the rate of a UHD remux).

How it was measured:

  • Every engine plays each file twice: from local disk, and over HTTP from a local range origin with a fixed latency and a shared link rate. Media servers deliver over HTTP, and an engine’s network reader is not its file reader.
  • The numbers come from aetherengine-bench. This project’s author runs it, so its method, its raw data and the cases where AetherEngine does not win are all in that repository.
  • AVPlayer does not open Matroska at all. On MP4, the one container it plays, it uses less than every engine here; see that repository’s README.
  • Machine: Apple M1 (MacBookAir10,1), macOS 27.0. HTTP from a local range origin at 1000 Mbit/s shared, 20 ms latency per request. Windowed, 3840x2160 px rendered, 60 s, median of 2.
  • The MacBook Air is fanless, so sustained decode can make it throttle. The protocol has cooldowns between runs and discards throttled windows.
  • Versions: AetherEngine 7.15.2, KSPlayer 2.3.4, AVPlayer macOS 27.0, VLCKit 4.0.0-alpha.21, libmpv mpv v0.41.0.

hevc-4k-hdr10.mkv (38 Mbit/s)

SourceGPU powerCPU loadRSSPeak footprintPlays
AetherEnginedisk41 mW4.4% of a core342 MB323 MBYes
HTTP40 mW5.4% of a core336 MB350 MBYes
KSPlayerdisk43 mW5.8% of a core333 MB772 MBYes
HTTP41 mW6.4% of a core357 MB496 MBYes
AVPlayerdisk, HTTPNo (“This media format is not supported.”)
VLCKitdisk266 mW15.0% of a core111 MB281 MBYes
HTTP244 mW17.0% of a core122 MB300 MBYes
libmpvdisk887 mW20.7% of a core475 MB933 MBYes
HTTP892 mW21.5% of a core629 MB1191 MBYes

Launch failures (crashes that were retried, not refusals, see the aetherengine-bench README): KSPlayer 1 (HTTP).

hevc-4k-hdr10-90m.mkv (90 Mbit/s)

SourceGPU powerCPU loadRSSPeak footprintPlays
AetherEnginedisk40 mW6.6% of a core577 MB681 MBYes
HTTP40 mW9.3% of a core557 MB700 MBYes
KSPlayerdisk41 mW6.7% of a core589 MB1040 MBYes
HTTP41 mW7.6% of a core578 MB703 MBYes
AVPlayerdisk, HTTPNo (“This media format is not supported.”)
VLCKitdisk235 mW16.2% of a core117 MB314 MBYes
HTTP235 mW18.4% of a core135 MB325 MBYes
libmpvdisk953 mW21.5% of a core501 MB980 MBYes
HTTP962 mW22.3% of a core684 MB1188 MBYes

Launch failures (crashes that were retried, not refusals, see the aetherengine-bench README): KSPlayer 4 (disk).

What the columns mean.

  • Power figures are package power with an idle baseline subtracted, so they are attributable to the run and not to the machine.
  • Peak footprint is the highest physical footprint the kernel saw for the player process over its lifetime, load phase included. That footprint is what the system’s memory limit acts on.
  • RSS (resident memory) is the mean over the measured window.
  • Decoding that happens in a system service (VideoToolbox) is in neither.
  • libmpv is measured with --hwdec=auto-safe (hardware decode), not mpv’s own software-decode default; see “Fairness decisions” in the aetherengine-bench README.
Frame delivery per run, and measurement notes

No engine dropped a frame. The table shows frame delivery and output as medians. Resolution, bit depth and HDR transfer are informational per engine only, and show as ‘varies across repeats’ when they disagree (see the aetherengine-bench README).

FixtureEngine, sourceDelivered / expectedOutput
hevc-4k-hdr10.mkvAetherEngine, disk1438 / 14383840x1714, 10-bit, hdr10
AetherEngine, HTTP1441 / 14383840x1714, 10-bit, hdr10
KSPlayer, disk1440 / 14403840x1714, 10-bit, SMPTE_ST_2084_PQ, served via KSMEPlayer
KSPlayer, HTTP1440 / 14403840x1714, 10-bit, SMPTE_ST_2084_PQ, served via KSMEPlayer
VLCKit, disk1451 / 14403840x1714, bit depth and colour transfer not reported
VLCKit, HTTP1454 / 14403840x1714, bit depth and colour transfer not reported
libmpv, disk1443 / 14403840x1714, 10-bit, pq
libmpv, HTTP1443 / 14403840x1714, 10-bit, pq
hevc-4k-hdr10-90m.mkvAetherEngine, disk1436 / 14383840x1714, 10-bit, hdr10
AetherEngine, HTTP1439 / 14383840x1714, 10-bit, hdr10
KSPlayer, disk1440 / 14403840x1714, 10-bit, SMPTE_ST_2084_PQ, served via KSMEPlayer
KSPlayer, HTTP1440 / 14403840x1714, 10-bit, SMPTE_ST_2084_PQ, served via KSMEPlayer
VLCKit, disk1454 / 14403840x1714, bit depth and colour transfer not reported
VLCKit, HTTP1452 / 14403840x1714, bit depth and colour transfer not reported
libmpv, disk1442 / 14403840x1714, 10-bit, pq
libmpv, HTTP1443 / 14403840x1714, 10-bit, pq

Every run rendered into 3840x2160.

CPU package power was measured for every run but is not published above. The idle baseline drifts with load so much that one engine’s own repeats can disagree more than the column would show between engines. For example, on hevc-4k-hdr10.mkv KSPlayer measured 76 to 267 mW of CPU package power across its own repeats, a 3.5x range. The values are recorded for every run in Results/; see “Known gaps” in the aetherengine-bench README for the full reasoning.

The 90 Mbit/s rows come from a second run of the same protocol, build and origin, on the same day and machine.

  • During the first half of the main run a tvOS Simulator was running in the background. It raised idle package power to 188 to 214 mW (19 to 26 mW once it was shut down).
  • Both hevc-4k-hdr10-90m.mkv baselines failed validation under it, so that fixture was measured again.
  • The hevc-4k-hdr10.mkv baselines were taken under it too and passed. None of the published columns moved with it: the GPU idle baseline stayed under 0.05 mW throughout, and CPU load, RSS and footprint are not baseline-subtracted.

Method and raw results: aetherengine-bench.