Yes—but only in games and rendering software that implement NVIDIA’s neural texture compression (NTC), and the size of the benefit depends on how the developer uses it. NTC can reduce the memory occupied by textures by reconstructing their values with a small neural network. In NVIDIA’s SDK example, inference during texture sampling uses less resident VRAM than conventional block compression; inference during loading saves disk and transfer space but uses the same VRAM as the block-compressed version.
What neural texture compression does
Conventional GPU texture compression stores image data in compact blocks that the GPU can sample directly. NTC takes a different approach: it stores a compact learned representation of a material’s textures and mipmaps, then uses a small neural network trained for that material to reconstruct texture values when needed. The aim is to keep access suitable for rendering while reducing the amount of texture data that must be stored.
As an Amazon Associate I earn from qualifying purchases.
NVIDIA Research’s 2023 paper, Random-Access Neural Compression of Material Textures, describes reconstructing material texture data on demand. NVIDIA says the method can provide two additional levels of detail—16 times as many texels—at a low bitrate. Its project-page comparison shows NTC at four times the resolution and 16 times the texels of the displayed BC-high example while using 30% less memory. Those figures describe the paper’s assets and settings, not a guaranteed result for every game. The work was published August 6, 2023, and was presented at SIGGRAPH 2023, where it received an honorable mention in the technical papers awards. NVIDIA Research: paper and project page; NVIDIA Research publication record.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errorsWhich NTC mode can reduce VRAM?
NVIDIA’s RTX NTC SDK describes two ways to use the compressed textures. The distinction matters: only on-sample inference keeps the compact representation resident for sampling. On-load inference expands the data into a conventional GPU texture format when the asset is loaded.
#1 Best Overall
- AI Performance: 767 AI TOPS
- OC mode: 2632 MHz (OC mode)/ 2602 MHz (Default mode)
- Powered by the NVIDIA Blackwell architecture and DLSS 4
- Axial-tech fan design features a smaller fan hub that facilitates longer blades and a barrier ring that increases downward air pressure
- A 2.5-slot design maximizes compatibility and cooling efficiency for superior performance in small chassis
| Representation or mode | Disk size | PCIe traffic | VRAM |
|---|---|---|---|
| Raw images | 32 MB | 32 MB | 32 MB |
| BCn | 12 MB | 12 MB | 12 MB |
| NTC, inference on load | 2.50 MB | 2.50 MB | 12 MB |
| NTC, inference on sample | 2.50 MB | 2.50 MB | 2.50 MB |
These are the SDK’s figures for a 2K-by-2K texture bundle; mip chains are excluded. The on-load path is useful for reducing stored and transferred asset size, but its textures are transcoded to BCn, so its example does not reduce resident VRAM compared with BCn. On-sample is the mode relevant to reducing texture VRAM, because the compact representation remains resident and the decoder reconstructs sampled values during rendering. NVIDIA RTX NTC SDK documentation and repository.
How large are the demonstrated savings?
NVIDIA’s January 6, 2025 announcement says RTX NTC saves up to 7× more VRAM or system memory than traditional block-compressed textures at the same visual quality. “Up to” is NVIDIA’s stated maximum, not a universal saving or an independently established result across released games. Actual outcomes depend on the assets, quality target, inference mode, and implementation. NVIDIA announcement.
Rank #2
- Powered by the NVIDIA Blackwell architecture and DLSS 4
- Powered by GeForce RTX 5070 Ti
- Integrated with 16GB GDDR7 256bit memory interface
- PCIe 5.0
- WINDFORCE cooling system
NVIDIA’s OptiX technical blog offers a separate production-scene demonstration: more than 100 8K UDIM textures with five layers each would take more than 32 GB uncompressed, while the illustrated NTC representation occupied less than 3 GB. NVIDIA rendered that example on a 16 GB GeForce RTX 5080 and says the compressed footprint was about half the size of BC-compressed textures. This is a specific scene example, not a forecast for a game’s total memory use. NVIDIA OptiX technical blog.
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →What this could mean for games
Texture memory is only one part of a game’s GPU memory budget. Geometry, render targets, shaders, acceleration structures, and other allocations also use VRAM, so reducing texture memory does not translate directly into the same percentage reduction in total VRAM use. The practical benefit may be more room for detailed textures, lower pressure on a card’s memory budget, or smaller assets to store and transfer—but the developer has to decide how to use the saved capacity.
Rank #3
- Powered by the NVIDIA Blackwell architecture and DLSS 4. System Requirements: Minimum 850W PSU with 16-pin 12V-2x6 (12VHPWR) connector required. Verify before purchasing.
- Military-grade components deliver rock-solid power and longer lifespan for ultimate durability. Compatibility: 348mm (13.7") length, 3.6 slots, 4.3 lbs. Confirm case clearance and slot spacing. GPU bracket included.
- Protective PCB coating helps protect against short circuits caused by moisture, dust, or debris
- 3.6-slot design with massive fin array optimized for airflow from three Axial-tech fans
- Phase-change GPU thermal pad helps ensure optimal thermal performance and longevity, outlasting traditional thermal paste for graphics cards under heavy loads
NTC also trades some runtime work for a compact representation. With on-sample inference, the renderer must reconstruct texture values while rendering; the effect on frame time depends on the workload and implementation. On-load inference avoids that per-sample reconstruction path by transcoding to BCn, but it gives up the resident-memory advantage shown in the SDK’s example.
Compatibility and performance depend on the renderer
RTX NTC is a developer technology, not a switch in a graphics driver that automatically compresses every game’s textures. A game must ship an implementation in its renderer and use relevant NTC texture assets. NVIDIA’s materials establish an SDK and demonstrations, but do not establish broad support in retail games or a user-facing setting for enabling NTC in existing titles.
Rank #4
- Powered by the NVIDIA Blackwell architecture and DLSS 4
- Powered by GeForce RTX 5060
- Integrated with 8GB GDDR7 128bit memory interface
- PCIe 5.0
- WINDFORCE cooling system
The SDK includes a compression and decompression library, a command-line tool, an interactive explorer, and a sample renderer. It supports Windows and Linux configurations with DirectX 12 or Vulkan, subject to API and driver caveats documented in the repository. Its hardware guidance varies by operation:
- Decompression on load: Shader Model 6 hardware is the stated minimum; NVIDIA Turing RTX 2000-series and newer are recommended.
- Inference on sample: listed as functional on Shader Model 6 hardware, but very slow there; NVIDIA Ada RTX 4000-series and newer are recommended.
- Compression: NVIDIA Turing RTX 2000-series is the stated minimum; Ada RTX 4000-series and newer are recommended.
The SDK says Cooperative Vector paths can accelerate inference on newer GPUs. Its DirectX 12 LinAlg/Cooperative Vector path is marked preview/testing-only in the repository, so developers need to check the current SDK requirements before treating that path as deployment-ready. GPU guidance is not a promise of a particular frame rate or memory reduction; those depend on the renderer and workload. NVIDIA RTX NTC SDK repository.
Best Value
- Powered by the NVIDIA Blackwell architecture and DLSS 4 OC mode: 2640MHz/Default mode: 2610MHz (Boost Clock)
- Military-grade components deliver rock-solid power and longer lifespan for ultimate durability
- Protective PCB coating helps protect against short circuits caused by moisture, dust, or debris
- 3.125-slot design with massive fin array optimized for airflow from three Axial-tech fans
- Phase-change GPU thermal pad helps ensure optimal thermal performance and longevity, outlasting traditional thermal paste for graphics cards under heavy loads
What players should expect
For players, NTC is a potential way for developers to fit higher-detail texture data into a constrained memory budget—not a feature that comes automatically with an RTX card. Until a particular game’s developer implements it, buying compatible hardware or changing a driver setting will not make that game use NTC. Even after implementation, the benefit will depend on which mode and assets the game uses.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
