File size difference CQP mode between 8-bit and 10-bit QuickSync H.265

[ukmark62]

Hi again!

I'm noticing a huge discrepancy in file size between 8-bit and 10-bit QuickSync H.265 encoding using default CQP (Constant QP) mode. (I tried this with both the "QSVEnc" (https://github.com/rigaya/QSVEnc/) and "Handbrake" programs and the same issue was present in both). Both these programs make use of the Intel Media SDK API for QuickSync encoding.

I did approximately a 5-minute video test where the only difference was changing the video profile from "main" to "main10". The 10-bit file was over 5 times the size of the 8-bit file!!! (approx 525Mb vs 100Mb). Also, this ratio was the same if I used "fixed-function" CQP encoding. The correct file size should be around 100Mb.

I've never seen such a difference in file size. Can you test this? Looks like some issue with the CQP encoding method and 10-bit H.265 in QuickSync.

Logs are below showing the command used in QSVEnc, where the only difference was changing the "main" profile to "main10".

8-bit log:-
------------------- Video encoding using QSVEnc 3.31 -------------------

D:\StaxRip\Apps\Encoders\QSVEnc\QSVEncC64.exe --avhw --codec hevc --quality best --profile main --cqp 24:26:27 -i "input.mkv" -o "output.hevc"

B pyramid on IvyBridge generation might cause artifacts, please check your encoded video.
cop3.DirectBiasAdjustment value changed off -> auto by driver
cop3.GlobalMotionBiasAdjustment value changed off -> auto by driver
QSVEncC (x64) 3.31 (r1738) by rigaya, Feb 20 2020 11:51:27 (VC 1916/Win/avx2)
OS Windows 10 x64 (18363)
CPU Info Intel Core i7-1065G7 @ 1.30GHz [TB: 3.89GHz] (4C/8T)
GPU Info Intel Iris(R) Plus Graphics (64EU) 300-1100MHz [15W] (26.20.100.7870)
Media SDK QuickSyncVideo (hardware encoder) PG, 1st GPU, API v1.31
Async Depth 6 frames
Buffer Memory d3d9, 1 input buffer, 33 work buffer
Input Info avqsv: H.264/AVC, 1280x536, 24000/1001 fps
AVSync cfr
Output HEVC main @ Level 3.1
1280x536p 1:1 23.976fps (24000/1001fps)
Target usage 1 - best
Encode Mode Constant QP (CQP)
CQP Value I:24 P:26 B:27
QP Limit min: none, max: none
Trellis Auto
Ref frames 3 frames
Bframes 3 frames, B-pyramid: on
Max GOP Length 240 frames
Ext. Features WeightP WeightB QPOffset tskip ctu:64 sao:all
encoded 8105 frames, 76.69 fps, 2550.85 kbps, 102.79 MB
encode time 0:01:45, CPU: 9.5, GPU: 97.0, VD: 19.0
frame type IDR 1
frame type I 35, total size 1.22 MB
frame type P 2026, total size 44.95 MB
frame type B 6045, total size 56.62 MB

Start: 16:12:05
End: 16:13:52
Duration: 00:01:46

10-bit log:-
------------------- Video encoding using QSVEnc 3.31 -------------------

D:\StaxRip\Apps\Encoders\QSVEnc\QSVEncC64.exe --avhw --codec hevc --quality best --profile main10 --cqp 24:26:27 -i "input.mkv" -o "output.hevc"

B pyramid on IvyBridge generation might cause artifacts, please check your encoded video.
cop3.DirectBiasAdjustment value changed off -> auto by driver
cop3.GlobalMotionBiasAdjustment value changed off -> auto by driver
QSVEncC (x64) 3.31 (r1738) by rigaya, Feb 20 2020 11:51:27 (VC 1916/Win/avx2)
OS Windows 10 x64 (18363)
CPU Info Intel Core i7-1065G7 @ 1.30GHz [TB: 3.89GHz] (4C/8T)
GPU Info Intel Iris(R) Plus Graphics (64EU) 300-1100MHz [15W] (26.20.100.7870)
Media SDK QuickSyncVideo (hardware encoder) PG, 1st GPU, API v1.31
Async Depth 6 frames
Buffer Memory d3d9, 1 input buffer, 33 work buffer
Input Info avqsv: H.264/AVC, 1280x536, 24000/1001 fps
VPP Enabled ColorFmtConvertion: nv12 -> p010
AVSync cfr
Output HEVC main10 @ Level 3.1
1280x536p 1:1 23.976fps (24000/1001fps)
Target usage 1 - best
Encode Mode Constant QP (CQP)
CQP Value I:24 P:26 B:27
QP Limit min: none, max: none
Trellis Auto
Ref frames 3 frames
Bframes 3 frames, B-pyramid: on
Max GOP Length 240 frames
Ext. Features WeightP WeightB QPOffset tskip ctu:64 sao:all
encoded 8105 frames, 53.56 fps, 13102.75 kbps, 528.02 MB
encode time 0:02:31, CPU: 9.3, GPU: 97.1, VD: 14.0
frame type IDR 1
frame type I 35, total size 4.74 MB
frame type P 2026, total size 196.54 MB
frame type B 6045, total size 326.74 MB

Start: 16:16:56
End: 16:19:28
Duration: 00:02:32

[ukmark62]

Tried with a different input file and the findings are the same - 10-bit is anywhere from 5 to 6 times larger than 8-bit.

[lakulako]

Hi @ukmark62,
To get similar size for 10 bit QPs should be increased by 12 relative to 8 bit case.
Msdk extends QP range to the higher QP values [1; 51+6*(bits-8)].

[ukmark62]

Thanks for getting back.

@lakulako - So the extended range for 10-bit is 1-63?
The equivalent 10-bit QP is 35 for 8-bit QP of 23 etc?

I'll close this issue.