libvpl 2.14 fails to achieve expected frame rate (only 32fps) when encoding 240Hz monitor capture
- Dominant language
- C++
- Stars
- 371
- Forks
- 107
- PR merge metrics
- No merged PRs in 30d
Description
System Environment Details:
OS: Windows 10 x64
CPU: 13th Gen Intel(R) Core(TM) i7-13700F 2.10 GHz
Graphics Card: Arc(TM) B580 Graphics
libvpl Implementation Details:
ApiVersion: 2.14
Implementation type: HW (Hardware Acceleration)
AccelerationMode via: D3D11
DeviceID: e20b/0
Problem Description
When using the official vpl-import-export.exe example (modified to log real-time FPS) to encode 240Hz monitor captures, severe performance limitations are observed under specific conditions: NV12
Resolution Alignment:
Non-16x multiples: Encoding stalls at 32fps (e.g., 2560x1440).
16x multiples: Achieves 230fps (e.g., 2560x1456).
Copy Mode:
Using -copy flag (direct memory copy) caps FPS at 32fps, regardless of resolution alignment.
How to improve encoding performance when adapting to different resolutions, and how to boost the encoding performance of YUV444 format specifically?
Contributor guide
Assessment
This issue has not been assessed yet.