Hey everyone, Dan here.
I’m the Tech Lead for Spooker. We’re currently chasing that magic Steam Deck Verified tag and spent the last few days doing a deep dive into optimization.

I wanted to share our exact process and the steps we took to profile and fix our bottlenecks. Hopefully, this helps some of you optimizing your own native Linux builds!

To set the stage, we’ve been pretty hardcore about performance from day one. We use Addressables for manual memory load/unload, mipmap streaming for textures, and audit our code religiously. Instead of heavy loops, our codebase is reactive, using R3 and VContainer for injection, alongside zero-alloc libraries like UniTask to keep our footprint low.

Despite all that, here is where our Steam Deck (64GB LCD) was sitting:

• FPS: Solid 60

• GPU: 90% at 1520mhz

• CPU: 40% at 1949mhz

• VRAM: 2.9 GB

• RAM: 6.9 GB

While 60fps is great, sitting at 90% GPU meant we had zero headroom. If we pushed the graphics any harder with new features, it would overflow and immediately drop frames.

Before tackling the GPU, we made one quick change: we ripped out Amplify Imposters and replaced it with the new automatic LOD system in Unity 6.3. Amplify is a great package, it just wasn't working well with our use case

Result: Immediate CPU drop from 40% down to 15–20%. Huge win right out of the gate.

We ran a bunch of different tests in isolated builds to figure out exactly what was choking the GPU. Here is the exact order of operations we followed:

1. Turned off post-processing: No change.

2. Set Render Scale to 0.5: HUGE drop. This immediately told us we were likely Fill Rate or Pixel Shader bound. We confirmed this by capping the frame rate from 60 to 30fps, which yielded a similar reduction in GPU load.

3. (Side note on STP/FSR: We could have just slapped on upscaling here, but that’s a band-aid. If we fix the root cause, STP/FSR becomes either totally unnecessary or just extra icing on the cake).

4. Forward+ vs. Forward: We toggled to Forward rendering to see if the Steam Deck was choking on compute operations. No change.

5. The "White Material" Test: We replaced every single material in the game with a basic white material. This confirmed we were specifically Fill Rate Bound—meaning we were choking on memory bandwidth, overdraw, or textures.

6. Frame Debugger - The Rogue Camera: Fired up the Frame Debugger and got an instant hit. A render texture camera was turning on at the wrong point and staying active. It was a minimal impact given our setup, but a free win is a free win. Fixed.

7. Frame Debugger - The Main Culprit: The debugger caught 59 draw calls sitting squarely between SSAO and Decals. Decals aren't amazing on mobile hardware anyway, and our SSAO settings in the URP asset were absolutely maxed out.

8. The Fix: We completely disabled decals (we don't actually need them and will replace them with quad/sphere shaders later). Then, we aggressively optimized the SSAO settings down to what we actually needed for our visual style.

Result: This was the first time we moved the needle on the GPU. It dropped from a stubborn 90% down into the low 70s%.

Since we had momentum, we went through and trimmed the fat everywhere else we could:

• Bloom: Turned High Quality Filtering OFF. Not necessary for our look.

• Opaque Textures: Downsampled Opaque to 4x box. This was a fantastic tradeoff with minimal visual impact (math came out to roughly 256k pixels down to 64k).

• Terrain Holes: Turned OFF. We don't even use Unity terrain, but the tooltip claims it speeds up builds. I'm slightly dubious, but what the hay, why not?

• Lighting/Reflections: Turned OFF MainLightShadows, Reflection Probes, and Reflection Probe Atlases. We simply didn't need them for our scenes and we already had shadows disabled on individual lights

Here is where the Steam Deck is sitting now:

• FPS: Still a rock-solid 60

• GPU: Comfy 55% – 70% (at a much lower 830mhz)

• CPU: 15% – 20%

• VRAM: 2.4 GB (Down 0.5 GB)

• RAM: 6.4 GB (Down 0.5 GB)

The Best Part: The Steam Deck battery reporting at 100% charge jumped from approximately 2 hours to 4.5 hours.

Overall, we are incredibly happy with this. Taking the time to actually isolate the bottleneck instead of just throwing FSR at the problem gave us massive thermal and battery gains.

The new optimized build we be out sometime next week!