It's strange that they don't include encoding time or SSIM metrics. I assume "other practical HEVC codec" is x265. Hopefully someone will do more in-depth benchmarks of this versus x264, x265, and VP9.
Optimizing memory consumption is nice, but I don't see the point-- HEVC video compression will saturate CPUs to the point that you wouldn't want to run other jobs concurrently, so who cares if the encoder uses 256MB or 1200MB? 1GB of RAM is very cheap!
"Optimizing memory consumption is nice, but I don't see the point"
More of the memory can't fit in cache, resulting in more cache misses and slower operations(as large swaths of memory need to be written to), gigabyte-level consumption resulting in more swapfile usage, and lowering memory for other applications, especially mobile devices(which have weak CPU/GPU and small battery capacity).
Any cache effects from using less memory would be represented in encoding speed, which is what matters.
If your video encoder pages into swap, you've lost all hope of being fast. Software HEVC encoding on mobile devices is not a use case most people care about.
5 comments
[ 3.4 ms ] story [ 23.1 ms ] threadOptimizing memory consumption is nice, but I don't see the point-- HEVC video compression will saturate CPUs to the point that you wouldn't want to run other jobs concurrently, so who cares if the encoder uses 256MB or 1200MB? 1GB of RAM is very cheap!
If your video encoder pages into swap, you've lost all hope of being fast. Software HEVC encoding on mobile devices is not a use case most people care about.