BlobTrack TOP reports different GPU Cook Time info depending on the processing done on its
blobtrack1_blobs table output.
Here is a demonstration:
BlobTrackBug.toe (29.4 KB)
Blobtracking a 256x256 texture with one blob continuously present and moving, converting the blobtrack data to CHOPs, and doing some heavy processing on the CHOPs.
If the heavy processing on the CHOPs are bypassed the blobtrackTOP reports ~5ms GPU Cook Time in this case, and double that (~10ms) when the heavy CHOP processing is enabled.
This signals some inconsistency in how the GPU Cook Time is calculated/reported for this operator, as downstream CPU processing should not affects its GPU time.
Seems like actual GPU usage is not affected though, but cannot confirm 100%.
Thanks for the report.
I can see a similar behavior as described in your report.
This is logged for a developer to look into it and get back to you.
This one is interesting. If I bypass the Filter CHOPs I see my GPU usage go up (in the Windows Task Manger), and the GPU cook time go down. If I unbypass them then the GPU usage goes down and the GPU cook time goes up. I think because these CHOPs are taking so long to process (2ms each), they are causing the GPU to get under utilized since the CPU isn’t giving it commands as quickly. I see the same behavior if I instead turn the Window Length of the Trail CHOP down to 2 seconds instead of 60. Reducing the CPU work helps with GPU load.
Overall I’d say this is just an issue with measurement which I need to look into, and not an indication of actual extra work occuring on the GPU.