Doom9's Forum - View Single Post - x264 Known Hardware accelleration problems and solutions

akupenguin · 20th December 2007, 22:09

That's strange. If you go by memory used, 4 frames of 1080p is equivalent to 9 frames of 720p, not 8.

If Level 4.1 were the only restriction, I might recommend people to restrict refs for hardware compatibility. But non-adaptive B-frames and no mixed refs? that's too harsh. Plus it's just plain a decoder bug, not a legitimate limitation. (As a codec developer, I can confidently say that adaptive B-frames do not require any explicit support. You just implement the standard, and any sequence of frame types is decodable.)

I question the DPB limit too... 1080p 16ref takes 50MB of RAM. Any video card recent enough to decode HD h264 will come with at least 256MB. Unless the decoder's memory is separate from the video memory? in which case it seems wasteful to dedicate 5% (12MB) to only h264 decoding when it could be reused.

20th December 2007, 22:09	#8 \| Link
akupenguin x264 developer Join Date: Sep 2004 Posts: 2,392	That's strange. If you go by memory used, 4 frames of 1080p is equivalent to 9 frames of 720p, not 8. If Level 4.1 were the only restriction, I might recommend people to restrict refs for hardware compatibility. But non-adaptive B-frames and no mixed refs? that's too harsh. Plus it's just plain a decoder bug, not a legitimate limitation. (As a codec developer, I can confidently say that adaptive B-frames do not require any explicit support. You just implement the standard, and any sequence of frame types is decodable.) I question the DPB limit too... 1080p 16ref takes 50MB of RAM. Any video card recent enough to decode HD h264 will come with at least 256MB. Unless the decoder's memory is separate from the video memory? in which case it seems wasteful to dedicate 5% (12MB) to only h264 decoding when it could be reused. Last edited by akupenguin; 21st December 2007 at 01:31.