ICQ+Lookahead performance with higher LA depth

[MSchm21]

Is it an expected result that on Skylake using ICQ_LA bitrate mode performance goes down quite heavy by using LA depth over 40. To give you an example here is one: ICQ_LA depth 40= 280 fps ICQ_LA depth 80= 220 fps ICQ_LA depth 100= 185 fps VBR_LA bitrate mode on the other side doesn't lose performance or a very small loss, practically the same not matter if depth 40 or 100. Memory consumption increases of course. Is it expected that VBR_LA can hold its performance with high depth numbers whereas ICQ_LA is losing lots of speed? GPU and MFX load is quite low on ICQ_LA with high depth values. And my second question, does Lookahead downsampling still work in newest drivers, because off/2x/4x seems identical.

[Harshdeep_B_Intel]

Hi Yups,

Thank you for testing and bringing this to our attention. For (1) question, LA_ICQ/LA performance is a know issue and it is due to shared memory access b/w CPU and GPU used which is more than expected. For (2) question LA downscaling is not enabled on latest drivers. But, both these issues is expected to fixed with upcoming/later driver releases for Skylake.

Thanks,

[Gilles_V_]

Hi,

is there any progress on this issue ? Il also have a 10 to 30% performance hit (closer to 30%..) when using regular LookAhead, whatever depth I choose.

I am using the 2017 SDK, on Skylake.

Thanks