We are LumenariaStudios, an independent studio building games in Unreal Engine 5. We want to share our DLSS results from our latest release, SHALTER-03, and from our next project, an unannounced data center simulation.
SHALTER-03 on Steam (demo available): SHALTER-03 on Steam
Integration details: Both games use DLSS 4.5 (NGX SDK 310.6.0, transformer model) and NVIDIA Streamline 2.11.1 with DLSS Frame Generation and NVIDIA Reflex. SHALTER-03 runs the NVIDIA Unreal Engine plugin 8.6.1 on Unreal Engine 5.6, and the data center simulation runs plugin 8.7.2 on Unreal Engine 5.8. We will keep both games updated to the latest DLSS and Streamline releases.
SHALTER-03
We integrated DLSS Super Resolution, DLAA, DLSS Frame Generation, Multi Frame Generation and NVIDIA Reflex. The game also includes an RTX (ray tracing) option with Ray Reconstruction support. We compared DLSS against the other upscaling solutions available in UE5.
Test setup: RTX 5070, 2560×1440, Epic settings, Unreal Engine 5.6
| Mode | Avg FPS |
|---|---|
| Native (no upscaling) | ~75 |
| DLSS Quality | ~75 |
| Other upscaler, Quality | ~95 |
| DLSS Quality + Frame Generation 2X | ~125 |
| Other upscaler + its frame generation | roughly equal, DLSS ahead by 5–10 FPS |
SHALTER-03 is largely CPU / render-thread bound, so upscaling alone brings limited gains. DLSS Frame Generation made the biggest difference, lifting us from ~75 to ~125 FPS. Image quality stayed intact, and in some areas it looked better than native.
The other upscaler gave higher raw FPS than DLSS Quality without frame generation. With frame generation on, it came close to DLSS in FPS. In both cases, however, image quality was far below DLSS. The biggest issue was severe flickering and shimmering on thin emissive meshes and fine detail, which gets worse with distance. DLSS kept the same content stable. The difference is clearly visible on the doorway:
Our next project: an AI data center simulation (pre-alpha, unannounced)
Our second project is a first-person simulation set in an AI compute facility. The player builds and runs a data center hall: they rack servers, fit network modules and cable every port by hand. A single hall holds thousands of individually simulated machines, switches, power strips and cable runs, so the game is draw-call heavy rather than pixel heavy.
This game is also a hard case for image stability. Its scenes are full of thin, high-frequency detail: hundreds of cable runs, dense rows of RJ45 and SFP ports, perforated rack doors, and small emissive status LEDs on every machine. This is exactly the kind of content where other upscalers flickered in SHALTER-03, which is why we are building this game’s rendering around DLSS from the start. We use DLSS Super Resolution, Frame Generation, Multi Frame Generation and Reflex, and we do not ship any other upscaler.
RTX 5070, 2560×1440, Epic settings, Unreal Engine 5.8
| Setting | FPS (displayed) |
|---|---|
| Native | ~30 |
| DLSS Super Resolution | ~35–38 |
| DLSS + Frame Generation 2X | ~50 |
| DLSS + Multi Frame Generation 4X | ~104–110 |
| DLSS + Multi Frame Generation 5X | ~130 |
| DLSS + Multi Frame Generation 6X | ~156 |
| DLSS + Frame Generation, Dynamic | ~160 |
In the current pre-alpha build, our large hall is render-thread bound (~32 ms per frame). Switching from native to DLSS Balanced cut GPU frame time from ~28 ms to ~14 ms, but the render thread stays the limit. This makes Multi Frame Generation the key technology for this game: it takes us from ~30 FPS to 100–160 FPS.
Latency is where Reflex proved essential. In Dynamic mode without Reflex, the frame rate was excellent but input latency was unacceptable. Enabling Reflex improved it noticeably, and Reflex On + Boost brought it to a fully playable level. In the meantime, we are reducing render-thread cost through draw-call reduction, instancing and Nanite, so Frame Generation can work from a higher base frame rate.
RTX 3050 Laptop GPU (4 GB), 1920×1080, Development build, Unreal Engine 5.8
| Setting | Internal resolution | Avg FPS | 1% low | GPU time |
|---|---|---|---|---|
| Native | 1920×1080 | ~15 | ~11 | ~60 ms |
| DLSS Quality | 1406×791 | ~19 (+30%) | ~14 | ~44 ms |
| DLSS Balanced | 1109×624 | ~21 (+40%) | ~17 | ~32–40 ms |
On lower-end hardware, DLSS Super Resolution alone gives a clear +30–40% boost. Frame Generation is not available on this RTX 30-series GPU, so this part of our team currently cannot test Frame Generation or Multi Frame Generation at all.
To the NVIDIA team: partnership inquiry
As a growing independent studio, we would love to explore partnership, hardware support or developer program opportunities with NVIDIA.
- Hardware: Our team currently develops on one RTX 5070 desktop and one RTX 3050 Laptop GPU. Access to RTX 50-series hardware for every developer would let the whole team build and test Frame Generation, Multi Frame Generation and Reflex directly. It would also greatly speed up our render-thread optimization work.
- Data center simulation: A game about building and running AI compute infrastructure is a natural fit for NVIDIA. With your permission, we would love to feature authentic NVIDIA hardware and branding in the game, such as GPU servers, networking and data center platforms, so that players build their halls with the real technology that powers modern AI.
We would be happy to share builds, demos, profiling data and side-by-side comparison captures with your team.
Thank you for your time and consideration.
Best regards,
The LumenariaStudios Team