Unstable images, GPU max

[laultman]

I am using StereoKit.HolographicRemoting in a console app rendering to my HoloLens 2. My Microsoft Surface Pro 7 (i7 with Iris GPU) hits Max as my models get more complex (see image).

So here are alternatives: 1 Scan at different resolutions for the type of use case, 2 up-scale the GPU to brute force the solution. Naturally, the best solution is, both. I should use common sense and scan for the type of presentation. I should also have sufficient GPU for the worst-case scenario.

The rendering is excellent for stable platform such as a 3D presentation on a fixed headset such as Windows Mixed Reality Headset and most likely any of the VR types. I am satisfied I could go on to about 20-40 million polygons in fixed position. HoloLens 2 is completely in another class in just about every way imaginable.

UPDATE…
I wrote all the above based on running in the development tools (debugger, etc.)
Running the same app in a console window without tools stabilized the image and cleaned up the individual models’ visual appearances. However, here is where my ignorance of GPU is glaringly obvious, I don’t know what to expect the utilization to be in the world of GPU. The GPU numbers did not change, still about 95%. My expectation is that the GPU numbers would move up or down based on the complexity of the object. I had 14 LVSRs at 182K polys and texture size of 29mBytes, 12 7tons, 12 LAVs, 16 Humvees with all about the same poly count but texture size about 10mBytes each. That’s about 8.1 million polys.

However, again my ignorance of GPU. I loaded all the models in such a way that they are cached on the GPU (per recommended StereoKit docs). I used a different pose position to display the models in different places in the frame (StereoKit Multiple Objects demo). So, do I really have 8.1 million or do I have < 400K polys? Is that why the GPU numbers are constant?