SmoothVideo Project — GPU Acceleration slower than non-GPU?

Re: GPU Acceleration slower than non-GPU?

null@example.com (James D) — Sat, 09 Jun 2018 21:08:35 +0000

Something is wrong here:

SVP no gpu acceleration + LAV no hardware decoder, CPU 70~100%. GPU 35~50%, mostly 50% (capped at 50%?) 60FPS
SVP with gpu acceleration + LAV no hardware decoder, CPU 100%. GPU 15%, 50~54 f

Even if SVP uses differnt Auto settings that wouldn't explain how GPU usage was lower when SVP was using GPU for OpenCL.

Re: GPU Acceleration slower than non-GPU?

null@example.com (Chainik) — Sat, 09 Jun 2018 11:21:55 +0000

> GPU Acceleration slower than non-GPU?

SVP may use different settings in AUto mode with "GPU acceleration" on or off.

Also note that SVP has "Video frame - Decrease to screen size" enabled by default.

GPU Acceleration slower than non-GPU?

null@example.com (Xuefer) — Sat, 09 Jun 2018 08:08:48 +0000

laptop Gigabyte Aero 15 i7 7700hq, GTX 1060
MPC-HC 64. 24fps Video: HVC1 3840x1920 23.976fps [V: hevc main 10 L5.0, yuv420p10le, 3840x1920 [default]]

SVP pro

without 23.976fps
no SVP + LAV no hardware decoder, CPU 20~40%, mostly 30%. GPU 13~15%
no SVP + LAV using dxva2-copyback, CPU at 15~18% CPU stable around 50%
no SVP + LAV using CUVID, CPU & GPU usage both stable at around 8~10%

with SVP targets 60fps, lowest settings
SVP no gpu acceleration + LAV no hardware decoder, CPU 70~100%. GPU 35~50%, mostly 50% (capped at 50%?) 60FPS
SVP no gpu acceleration + LAV using dxva2-copyback, CPU at 35% CPU around 35%, buffer under flow, lag, about 46FPS only
SVP no gpu acceleration + LAV using CUVID, CPU 30%, GPU 11~15%, 60FPS

SVP with gpu acceleration + LAV no hardware decoder, CPU 100%. GPU 15%, 50~54 f
SVP with gpu acceleration + LAV using dxva2-copyback, CPU at 20% CPU around 40%, buffer under flow, lag, about 50FPS only
SVP with gpu acceleration + LAV using CUVID, CPU 30%, GPU 15~17%, 60FPS

hrm... performance impact of using LAV with dxav2-copyback is very hight, any idea?
CUVID is deprecated in the favor of dxva2 but how come dxva2 is slower