When doing a longer video of 2 minutes/22 seconds and no hardware, I only found a 3.3% increase.  So I think the results are very dependent.  This video was 3840x2160 with 25 framerate, av1/webm.  I rendered h264.mp4 format.

   Build:  3602 frames  251.651 secs 14.313 fps

   AppI:  3602 frames 260.165 secs 13.840 fps

But anyway. I did add a note to the Manual stating that AppImage could be slower and referred to the section on "distros with Cinelerra included" which references Andrey's RPMs, etc.

On Tue, Jan 7, 2025 at 8:27 AM Terje J. Hanssen <terjejhanssen@gmail.com> wrote:

Den 07.01.2025 15:48, skrev Phyllis Smith:

Slower AppImage is to be expected, but still disappointing.  Bottom line is that users who can take advantage of Andrey's RPMs should always do so.  Besides being faster, it provides more capabilities and opportunity to make some minor side changes.  Thank you, Terje, for doing the tests and passing along this information.

On Mon, Jan 6, 2025 at 5:35 PM Terje J. Hanssen via Cin <cin@lists.cinelerra-gg.org> wrote:

I did a few hevc and av1 OSV render tests using the latest RPM w/oneVPL
on Leap, to compare it with my own AppImage.
To get the AppImage from Slowroll to run om Leap, I had to add-install a
newer version libFLAC12.

The machine is AlderLake w/Arc A750 gpu
The OS root drive is on a std SATA disc, while the video files are
read/written to a fast M.2 disc.

Results:

hevc_qsv_8b420.mp4
** rendered 5972 frames in 19.255 secs, 310.153 fps (RPM)
** rendered 5972 frames in 23.798 secs, 250.945 fps (AppImage)

av1_qsv_8b420.mp4
** rendered 5972 frames in 18.612 secs, 320.868 fps (RPM)
** rendered 5972 frames in 21.470 secs, 278.156 fps (AppImage)

In average rendering is ca. 20% faster with the RPM.
I did also additional test with the AppImage on Slowroll and got a bit
faster rendering, but not so fast as with RPM on Leap.

==========================

Not especially news or differently, but as I also have supplemented my AppImages and native builds tests on Slowroll, I add them here.  My conclusion is that the package (RPM) build (due to most dynamically linked libs I guess?) still stands as fastest, while the AppImages supply with benefit of portability (cross Linux distributions).


On Slowroll (root and video on the M.2 NVMe SSD)
------------------

CinGG-20241120-x86_64.AppImage

hevc_qsv_8b420.mp4
** rendered 5972 frames in 20.574 secs, 290.269 fps

av1_qsv_8b420.mp4
** rendered 5972 frames in 21.612 secs, 276.328 fps
-------------

CinGG_use_system_ffmpeg-71_20241020-x86_64.AppImage

hevc_qsv_8b420.mp4
** rendered 5972 frames in 21.065 secs, 283.503 fps

av1_qsv_8b420.mp4
** rendered 5972 frames in 21.033 secs, 283.935 fps
-------------------------

Native built with Cingg's internal ffmpeg 7.0
bin/cin
Cinelerra Infinity - built: Nov 20 2024 22:06:05

hevc_qsv_8b420.mp4
** rendered 5972 frames in 21.299 secs, 280.389 fps

av1_qsv_8b420.mp4
** rendered 5972 frames in 21.313 secs, 280.205 fps
---------------------------------

Native built with dynamic linked system ffmpeg 7.1

/Cin/bin_use_system_ffmpeg-71> bin/cin
Cinelerra Infinity - built: Oct 20 2024 21:21:06

hevc_qsv_8b420.mp4
Had to comment out
# profile=main
** rendered 5972 frames in 20.988 secs, 284.544 fps

av1_qsv_8b420.mp4
Had to Set Format (hdv input) tff interlaced to Not interlaced
** rendered 5972 frames in 20.835 secs, 286.633 fps
** rendered 5972 frames in 21.024 secs, 284.056 fps