Proposal: Long-Term Rendering Modernization (DX12, DLSS, Frame Generation, Ray Tracing)

[GhervasaCristian]

Context

There have been ongoing discussions and experiments around modernizing the rendering pipeline (e.g., DX11 improvements, OpenGL exploration). I would like to open a broader discussion about a potential long-term evolution path for the renderer, focused on modern graphics features and APIs.

This is not a concrete implementation plan, but rather an attempt to evaluate whether such a direction makes sense for the project and its community.

---

Proposal Overview

The idea is to explore a new rendering path (not necessarily replacing the current one) that could include:

• DirectX 12 backend
• Temporal rendering pipeline (TAA as a base)
• DLSS / XeSS / FSR2+ support
• Frame Generation (DLSS FG / FSR3 / OptiScaler-style approaches)
• Optional hardware ray tracing (where feasible)

This would effectively mean moving toward a modern frame-based rendering architecture, closer to contemporary engines.

---

Motivation

• Improve performance scalability on modern GPUs
• Enable advanced upscaling (DLSS, etc.) for high-resolution and mod-heavy setups
• Provide a foundation for future features (ray tracing, better lighting models)
• Keep the engine relevant long-term as hardware and expectations evolve

---

Major Concerns

This direction would come with significant tradeoffs:

1. Mod Compatibility

• Many existing graphics mods modify shaders and rendering behavior directly

• A modern pipeline would likely require:

  • motion vectors
  • standardized buffers
  • stricter shader contracts

• This could break or invalidate a large portion of current mods

2. Ecosystem Impact

• Mod authors may need to:

  • rewrite shaders
  • adapt to new interfaces

• It may split the ecosystem into:

  • “legacy renderer” mods
  • “modern renderer” mods

3. Development Scope

• Estimated effort: potentially 1–2 years (depending on contributors and scope)

• Requires deep changes:

  • render pipeline restructuring
  • new data flow (temporal history, motion vectors)
  • API/backend work (DX12)

4. Maintenance Burden

• Supporting both legacy and modern pipelines could increase complexity
• Risk of fragmentation or reduced stability

---

Possible Approach

One idea would be to:

• Keep the current renderer as a stable legacy path

• Introduce a separate “modern renderer” path

  • opt-in
  • clearly documented requirements for mods

• Gradually build tooling and guidelines for mod authors

---

Key Questions for the Community

1. Direction

   • Do you think moving toward a modern rendering pipeline is a worthwhile long-term goal?

2. Compatibility vs Progress

   • How should the project balance:

     • preserving existing mods
     • enabling modern graphics features?

3. Ecosystem Adaptation

   • Do you believe mod authors would realistically adapt to:

     • stricter shader requirements
     • new rendering constraints?

4. API Choice

   • Is pursuing DirectX 12 a reasonable direction, or should focus remain on improving DX11?

5. Feature Priorities

   • Among the following, what would actually matter most?

     • DLSS / upscaling
     • Frame generation
     • Ray tracing
     • general renderer cleanup/performance

6. Project Scope

   • Is a large, long-term effort like this appropriate for the project?
   • Or should development stay incremental and conservative?

---

Goal of This Discussion

The goal is to understand whether:

• this direction aligns with the project’s vision
• the community would support such a transition
• it is realistic given the current ecosystem and contributor base

---

This is intentionally a high-level discussion, not a feature request or immediate roadmap item. The intention is to gather perspectives before any serious technical work is considered.

AI-assisted development may shorten implementation and validation time, but the biggest step remains evaluating whether this direction is beneficial for the project and its ecosystem.

[Xottab-DUTY]

Thanks for raising the topic! Although, I would prefer using AI for actual coding, not the discussion :)

@yohjimane is our current project leader and is working on modern renderer with frame graph and NVRHI, @vTurbine is our main renderer programmer, so we should consult them first regarding this topic.