Summary of "AMD, Nvidia and Intel Just Solved EVERYTHING!"

What this is

Neural texture compression (NTC) is an AI-based texture/material compression approach that plugs into the render pipeline. Small neural networks (neural materials / latent representations) decompress textures at runtime instead of using traditional block formats (e.g., BC7). This decompression is deterministic (not generative like DLSS5) to avoid hallucinations and preserve visual fidelity.

Key benefits and claims

Massive VRAM reductions
- Nvidia demo (“Tuscan Villa”) reduced VRAM use from 6.5 GB to ~970 MB (~6.7× lower).
- Nvidia suggests scenes that needed 16 GB could need ~2–3 GB with NTC.
Storage reduction
- Textures stored in compressed neural form reduce game install sizes.
Render performance gains
- Nvidia reported 1.4×–7.7× faster render times at 1080p for some workflows using “neural materials.”
Practical impacts
- Helps with VRAM/memory shortages, reduces shipping/storage costs and can enable richer scenes on constrained hardware.

Vendors, branding, and status

Nvidia — “RTX Neural Texture Compression”
- Paper announced in 2023; demoed at GTC.
- SDK released that supports Nvidia GPUs and offers cross-vendor support for AMD and Intel.
- Recommends RTX 4000-series (heavy tensor/core compute) for good runtime performance.
- Non‑Nvidia GPUs can function but will generally be slower at decompression.
Intel — “Texture Set Neural Compression” (TSNC)
- Demoed cooperative vectors; plans an SDK later this year.
- Public claims include compression ratios up to ~9× and ~18× (scene and comparison dependent).
AMD — “Neural Texture Block Compression” (NTBC)
- Public focus is on AI improving block-compression (BC-style) rather than full runtime neural decompression.
- Likely working on runtime options, but public materials are less mature.
Microsoft
- Added cooperative vector support in DirectX 12 Shader Model 6.9 — a low-level shader feature that enables these neural decompression techniques.

Hardware/software compatibility and caveats

SDK validation
- Vendors have validated SDKs on older AMD RX6000 and Intel Arc A-series GPUs, but practical runtime decompression is compute-heavy.
- Newer hardware (Nvidia recommends RTX 4000-series) will deliver the best real-world performance.
Cross-vendor performance
- Nvidia’s SDK supports AMD/Intel GPUs, but decompression speed and efficiency will generally be best on Nvidia hardware due to architecture/tensor core optimizations.
Variability and adoption
- Real-world benefits depend on developer adoption — game engines and studios must integrate the tech.
- Compression ratios and performance vary by scene and implementation; vendor claims are not directly comparable without identical testing conditions.

Related tech and context

Nvidia is also working on “auto shader compilation” (compile shaders while the system is idle to reduce in‑game shader stutter).
This wave follows Nvidia research (2023 paper) and GTC demonstrations.
Microsoft’s DirectX 12 Shader Model 6.9 (cooperative vectors) is the enabling platform-level feature.

Actionable points for developers and gamers

Developers
- Target Shader Model 6.9 (cooperative vectors).
- Experiment with and adopt vendor SDKs (Nvidia, Intel, AMD as available).
- Test across GPU vendors for performance and fidelity; validate compression ratios on target scenes.
Gamers
- Expect future updates that reduce VRAM and storage requirements.
- Older GPUs may not see full runtime benefits until they are upgraded or until vendor implementations are further optimized.

Main speakers and sources referenced

Video host: Gamer Mel (Gamer Meltd / Gamer Mel)
Primary technical sources:
- Nvidia (GTC demo, 2023 paper, SDK)
- Microsoft (DX12 Shader Model 6.9 / cooperative vectors)
- Intel (TSNC demo & upcoming SDK)
- AMD (neural/block-compression work)