Media lab / Intel Quick Sync

EQi12 Jellyfin QSV Test: 4K60 and Concurrency

Windows Jellyfin hardware-transcoding evidence, two- and four-stream limits, codec support, GPU use, and measured wall power.

What was tested

The EQi12 uses an Intel Core i3-1215U with integrated Intel graphics. We tested native Jellyfin on Windows and direct FFmpeg/Intel Quick Sync workloads to answer three separate questions:

  1. Can Jellyfin use QSV for a 4K60-to-1080p60 transcode?
  2. How many identical QSV workloads can remain real time?
  3. Which modern codecs decode or encode through the available QSV runtime?

These tests should not be collapsed into one marketing number. A successful codec probe does not guarantee application-level concurrency, and a direct FFmpeg run is not the same as four remote Jellyfin clients.

Native Windows Jellyfin result

Jellyfin Server 10.11.11 ran natively on Windows at port 8096. The client selected a 10Mbps playback target for an H.264 3840x2160 60fps source.

The Jellyfin-generated FFmpeg log recorded:

That log is sufficient to state that native Windows Jellyfin QSV hardware transcoding passed for this test. A prior software-only attempt is not used as the final result.

Jellyfin 10Mbps playback selection used for the native Windows QSV test
Original user capture from the native Windows Jellyfin session. The decisive QSV evidence comes from Jellyfin's generated FFmpeg log.

Two streams versus four

The concurrency test processed the same 4K60 source for 60 seconds per stream.

Concurrent workloads Completion rate CPU average GPU video-decode average Interpretation
2 about 1.01x each 21.89% 62.59% Real time in this test
4 about 0.708-0.710x each, 42-43fps 27.86% 95.48% Completed, but not real time

The supported conclusion is two simultaneous 4K60-to-1080p60 QSV workloads at real-time speed under these conditions. Four completed but fell below real time.

This was a direct FFmpeg/QSV capacity test. It is not evidence of four real Jellyfin clients with network delivery, subtitles, audio conversion, authentication, and library activity occurring simultaneously.

Two-stream and four-stream Intel QSV concurrency result
Corrected concurrency result: two matching workloads remained real time; four completed below real time. The invalid first script run is excluded.

Codec matrix

A corrected five-second source set produced the following results:

Operation Result
H.264 4K60 QSV decode passed, about 4.17x
HEVC Main10 1080p60 QSV decode passed, about 18.9x
VP9 1080p60 QSV decode passed, about 23.3x
AV1 1080p60 QSV decode passed, about 16.7x
H.264 to H.264 QSV transcode passed, about 2.31x
HEVC Main10 QSV encode passed, about 1.52x
AV1 QSV encode unsupported

The QSV runtime returned “Current codec type is unsupported” and “Function not implemented” for AV1 encoding. AV1 decoding and AV1 encoding are different capabilities; success in one does not imply the other.

An initial script run exited almost immediately because of a PowerShell parameter conflict. Those files are labeled INVALID_FIRST and are excluded from the hardware conclusion.

Intel QSV codec matrix raw result
Raw codec-matrix result used to separate decode support from encode support, including the unsupported AV1 encode path.

Power during hardware transcoding

The wall meter showed about 13W while Jellyfin was actively performing the QSV transcode. The valid FFmpeg log confirms the QSV and VPP path.

This is an observed activity reading, not a full five-minute average. Jellyfin processed ahead at more than playback speed, so the load was not constant across the entire media duration.

For context, Windows idle measured about 14W, Docker services idle about 12W, and sustained CPU load about 37W. The close idle and QSV readings illustrate why hardware media engines are useful, but the meter’s whole-watt resolution and background tasks limit finer comparisons.

Playback tests

Local Edge playback of a 4K60 file averaged about 38.58% GPU Video Decode with a peak around 42.91%; the recorded run showed 11 dropped frames out of 4,886.

YouTube 2160p60 used VP9 and Opus at 3840x2160 60fps. During an accurately bounded 30-second interval, no additional dropped frames were recorded, while GPU Video Decode averaged about 25.61% and peaked around 27.64%.

These playback observations support decoding capability but are separate from server-side transcoding.

YouTube 2160p60 Stats for Nerds capture on the EQi12
End-of-run YouTube 2160p60 Stats for Nerds capture. This playback check is reported separately from server-side Jellyfin transcoding.

Configuration guidance

Bottom line

The EQi12 passed a demanding single-stream Jellyfin QSV test and sustained two matching 4K60-to-1080p60 workloads at real-time speed. Four-way processing was below real time, and AV1 hardware encoding was unsupported. For a small household with one or two demanding transcodes, the tested result is encouraging; larger concurrency claims require different hardware or lower-complexity source media.

Continue with QSV

Use the i3-1215U codec matrix, follow the Windows QSV setup guide, or diagnose unexpected software transcoding.