Open oviano opened 4 years ago
To make sure user can replicate, hw/sw configures used in benchmark need to be exactly the same. That's partial reason we didn't include benchmark data in the repo. I don't have that many hardware so could not comment on the benchmark data in the link. We perform performance regression test with each release. Haven't seen the speed decrease showed in the link(what command line is used?). If you use SVT to develop your own product, the latest release or master is always recommended. The white paper in https://01.org/svt includes some official data on Xeon. SVT architecture is based on multiple different parallelism processing. The speed increases with more cores inside CPU.
All I have found so far is this:
https://openbenchmarking.org/showdown/pts/svt-hevc
Is this reliable? It seems to show a big decrease in speed from SVT-HEVC v2019-02-03 to SVT-HEVC v1.4.1 (taking Intel Core i9-9900K as an example), but maybe that was expected?
Also that one doesn't seem to show what encode settings (preset, etc) were used so it's not too useful.