Open wangjingz opened 5 years ago
Hi, for M8, M9 it's relatively fast mode, and it became memory bounded rather than CPU bounded. So if testing on dual 8180, better to use taskset or numactrl to bind the encoding instances to separate node
@lijing0010 I verified the issue on last master version(CommintID:5ebc0b2 ), it can be reproduced. Because I only run one channel, so I think maybe don’t need to use taskset or numactrl to bind the encoding instances to separate node.
The sample command line is:
encoders/SvtHevcEncApp -i Netflix_Boat_4096x2160_10bit_60Hz_P420.yuv -b bitstreams/Speed_Test_M0_Netflix_Boat_4096x2160_10bit_60Hz_P420_SQ_TBR13000.265 -errlog bitstreams/Speed_Test_M0_Netflix_Boat_4096x2160_10bit_60Hz_P420_SQ_TBR13000.errlog -w 4096 -h 2160 -bit-depth 10 -fps 60 -intra-period 55 -n 0 -encMode 8 -compressed-ten-bit-format 0 -rc 1 -tbr 13000 -tune 0
4K performance have some drop between Mx and Mx+1: