Video summary

DF Direct Q+A: Will AMD Deliver Its Own DLSS 5? Will CD Projekt RED Bring Back The RED Engine?

Main summary

Key takeaways

Technology

Tech/Product Analysis Summary (DF Direct Q&A)

1) AMD vs Nvidia after “DLSS 5” (and whether AMD will copy Nvidia)

Question focus: Where AMD should go after Nvidia’s DLSS 5 launch, given that FSR is positioned as “behind DLSS.”

Key points discussed:

  • Vendor-by-vendor approaches are likely: AMD/Intel may develop their own AI/upscaling/neural filtering variants rather than matching Nvidia 1:1.
  • Skepticism about a “screen-space” neural approach as the best long-term rendering solution—more like a stop-gap (similar in spirit to earlier screen-space techniques such as screen-space ambient occlusion).
  • Adoption challenge: implementing something DLSS-5-like may require engine/developer-level oversight and integration, limiting native game support.
  • Console relevance: consoles often use AMD-based hardware, but real adoption still depends on integration cost.

2) Limits of a screen-space / post-process neural filter

What DLSS 5 is constrained by:

  • It can only infer/generate details consistent with information present in the current frame (e.g., what’s visible or inferable from available buffers like motion vectors).
  • It can’t hallucinate content not represented in screen-space inputs, such as missing off-screen ray-traced reflection data (example: Control, where reflections aren’t on screen).

Artifacts / failure modes:

  • Conflicting screen-space lighting signals can cause problems (example mentioned: Final Fantasy VII Rebirth).

Panel expectation for future iterations:

  • If future versions move closer to artist-quality reconstruction, they likely need more game-engine inputs beyond what’s typical for upscaling.
  • DLSS 5 is described as surprisingly capable even with limited inputs (including colored motion vectors), but further gains probably require more buffers/data, more training/parameters, and better integration.

3) Potential approach: deeper engine integration (Unreal Engine / Lumen-style integration)

Suggestion:

  • Instead of purely vendor-specific post-processing, vendors could pursue more fundamental engine integration, especially as Unreal Engine 5 adoption grows (e.g., Lumen).

Reality check:

  • Some integration responsibilities would likely land more with Epic/UE than each GPU vendor, though vendor-side integration is still possible.

4) DLSS 5 mods & multi-GPU experiments (and official support implications)

Mod proliferation as a major theme:

  • Mods enable DLSS 5 across unexpected platforms/configurations, including running RTX-era DLSS 5 on other GPU generations/AMD hardware (performance may be poor, but it works).
  • Reverse engineering and AI-assisted tooling are said to make experimentation faster and more widespread than before.

Multi-GPU mod discussion:

  • One mod offloads DLSS 5 processing to a second Nvidia GPU (multi-pass / filtering split).
  • Pros: keeps the tech discussion moving; demonstrates extensibility.
  • Cons: latency cost makes it impractical as a universal solution.

Official support implications:

  • Nvidia allegedly announced official support for RTX 40 series after mod pressure—panel suggests the mod activity effectively forced or accelerated official compatibility decisions.

Other observation:

  • Some mods apply the DLSS-5 filter earlier than post-upscaling, reducing frame-time cost but potentially impacting image quality.

5) Game dev agility discussions (not directly DLSS, but tech/process)

Crimson Desert / Pearl Abyss:

  • Praised for rapid patching, community response, and agility in integrating new mechanics.
  • Panel suggests a redo/review later once the game has progressed and/or around DLC release to check whether previously raised performance/image-quality issues are resolved.
  • Mentions uncertainty around whether console-related issues have truly improved and the stability of performance/features.

6) Capcom’s RE Engine feature choices (Only Musha / “why conservative visuals?”)

Core argument:

  • Enabling “higher-end” ray tracing isn’t just a checkbox—it requires engineering effort, tuning, and can introduce noise/artifacts in darker scenes (example referenced: RTGI from Requiem).
  • Capcom is described as selecting features project-by-project, balancing:
    • stable performance/image quality,
    • production schedules,
    • and development budget.

Also cited:

  • Capcom’s earlier titles used ray tracing features unevenly, rather than consistently.

7) CD Projekt Red and the “RED Engine” future vs Unreal Engine bridging

Question:

  • Is UE5 being used only as a bridge to gain time/knowledge while renewing an in-house engine?

Panel response:

  • RED Engine work continues for supporting older titles (e.g., ongoing Cyberpunk updates).
  • Panel sees no sign that Witcher 4’s pipeline will switch engines.
  • They view Witcher 4 as firmly built on Unreal already, making mid-project switching unlikely.

Alternative possibility mentioned:

  • A business partnership/co-development with Epic—panel suggests it’s more than simple “off-the-shelf” licensing, though the economics are speculative.

Sentiment:

  • “Shame” is expressed because RED Engine historically achieved strong technical results.

8) Console hardware manufacturing speculation (Sony, 3DO-style approach, and “Steam-like” future)

Speculation challenged:

  • The panel argues that third-party manufactured “console box by others” doesn’t automatically solve next-gen economics.

Key cost driver:

  • RAM and SSD modules are highlighted as the major next-gen cost factors; panel suggests Sony likely has sufficient buying power.

Ecosystem strategy:

  • Sony’s approach is framed as maintaining a closed ecosystem, which conflicts with “PC/Steam-like” distribution models.

Example mentioned:

  • Nintendo is referenced as a case where these questions are less relevant due to its ecosystem-control incentives.

Consensus:

  • 3DO-like approaches are unlikely; Microsoft may be more aligned due to incentives tied to the PC/Windows ecosystem.

Reviews / Guides / Tutorials Mentioned

  • No explicit “how-to” tutorial segment is presented; the episode is primarily expert analysis and Q&A.
  • A suggestion exists to do a future re-review/redo of Crimson Desert once the game has progressed and/or around DLC release to verify whether earlier performance/image-quality issues were resolved.

Main Speakers / Sources

  • Rich Leadbetter (host; referred to as “Rich”)
  • John Linneman (referred to as “John”)
  • Oliver McKenzie (“Oliver”)
  • Guest questioners (audience) mentioned by username: Skyrim 26 FPS edition, Carvox, Messi exclamation point, Scud, Yachtsuya, Adam Guten Tuck, Dan, Jonathan Hamilton

Original video