Notch TOP >1ms slower than FXPlayerHostD3D11_x64.exe with same empty layer

[dylanroscover]

Hello,
This post has a sibling discussion over on the Notch forums: Notch Block ~8ms cook time with empty layer selected - #6 by dylanroscover - Issue reporting - Notch Community Forum

What I’ve identified is the following:

1. Notch Block layers do not affect each other’s render times.
2. Empty Notch Block layers (no nodes/default render settings) running 7680x4320px@60 take about 0.9ms inside of Notch’s FXPlayerHostD3D11_x64.exe “Block player.” The same Notch block/layer/resolution/fps running inside of a Notch TOP takes about 2ms to cook. This excludes the overhead of the Notch TOP itself, which is an additional 1.7ms.
   
   
   
   
   

My two questions are:

1. Why is the reported Notch Block Render Time info par over a ms slower than what I would assume is the same value in 10-bit’s Block player?
2. Is there any way to improve performance in this regard (an empty layer inside a Notch Block in a Notch TOP)?

Thanks!

[malcolm]

I’m not sure trying to diagnose an empty test file is that productive. There is a lot of things that can happen with the GPU and driver, and if it’s not under load values you get from timings aren’t the most reliable.
Do you instead of a more heavy file that is showing similar behavior?
It’s also very possible we are just timing things differently than Notch is. We don’t get that value from Notch, we insert our own DX timing calls around the Notch render call

[eric.b]

The framerate you’ve highlighted there is reporting the framerate from ImGui, not the block render time. The block host does perform the GPU timing of the block but it doesn’t expose it in the packaged exe from what I can tell. When I modified it to see the block render time it was comparable to what we have in TD.

[dylanroscover]

Thanks @malcolm and @eric.b. Will conduct more extensive testing with the heavier layers of the file, keeping in mind what you both have explained.