Reflection Probes on Mobile?

[proper zealot]

Hey devs:

Got a quick question. We're making a mobile game with some fancy bells and whistles and we're doing optimizations on it close to an early release, just noticing now that we have reflection probes in many of our scenes.
URP render pipeline.
A lot of our background objects are using a custom stripped-down PBR shader and the metalness rely on these probes. But we also mostly use a static camera angle that doesn't move so we dont get much of the benefit of active reflections.

I and another TA on the team disagree about the cost of these. I'm wondering about a second opinion.
We dont have specific GPU profile data on this yet, wondering about high level guideline type of info and it's hard to find by googling. We do have min spec that goes down to iPhone X.

Anybody willing to provide some insights?

[fiery rock]

I don't know about optimizing for mobile but ultimately you have to test to compare
Reflection probes aren't just important for camera motion, but for object rotation as well, and for an object to reflect its surroundings when going from one environment to the next
You can disable reflection probe blending and box projection from URP to make reflection probes computationally quite simple
Instead of reflection probes I guess one step down from them would be to sample a cubemap directly to get reflections but to avoid reflection volume component overhead

Old games used spheremap reflections, or matcaps even more primitively which could be implemented now
But I'm not sure if that provides a tangible advantage
Cube texture type supports "mirrorball" mapping which is a type of spheremap, that I'd tentatively guess might be cheaper to sample on platforms that don't support texture arrays

And even if matcaps were cheaper on 90s GPUs that might've been if they had a hardware function for rendering them specifically
So now matcap sampling may be less optimized than cubemap sampling

[proper zealot]

Good food for thought. Thank you!
Once we get to more in depth testing later this week, this gives us some things to try out and compare against.

[fiery rock]

My guess is that simple reflection probes likely aren't much more expensive than "fake" custom reflections
And it may come down to CPU cost depending whether there's reflection probe components in the scene to require sampling positionally
But I could be wrong

It's very hard to do anything resembling PBR without some kind of environmental specular

[proper zealot]

I mean, when I say static cameras, I mean most of our environments could be rendered as PNGs, PS1-style

[fiery rock]

Do you have any example?

PS1 games used prerendering for that so they could have very fancy shading as long as it was static
Dynamic objects had reflections painted on their color maps

Or using matcaps

(or metallic reflective materials were avoided altogether)

[proper zealot]

yeah. Like, we really arent using the benefits of 3D space for a whole lot of the environments. it feels wasteful to use PBR with all its normal and spec and detail maps when the viewing and lighting angle never change

[proper zealot]

we're under NDA so I cannot share anything, really

like you said, we'll have to test with andw ithout, and with those halfway steps you mentioned above

I appreciate it, thank you

[fast dagger]

Random stupid suggestion :
If everything is so static as you suggest, why no actually used pre-rendered images ?

[proper zealot]

I've suggested it. Got overruled.