NVIDIA continues to improve its texture compression system to minimize the use of video memory on our GPUs. Coupled with DirectX Cooperative Vector technology from Microsoft, its RTX NTC (Neral Texture Compression) device now seems to pass a new CAP.
We were talking to you a few weeks ago AMD research on the reduction of VRAM, specifying that Nvidia was of course working on of course on devices having, more or less, the same vocation. Demonstration is made this week with RTX NTC (Neral Texture Compression) technology, which manages to significantly boost its performance when it is combined with DirectX Cooperative Vector of Microsoft.
As reported WCCFTechthe first tests in real conditions of NVIDIA techno put us on the track of a reduction of 90 % (maximum) of the TRAM occupied for the loading of textures.
This solution would provide a convincing response to the problems of video memory saturation, in particular on GPUS with 8 GB of true. A limited capacity, and yet still used on a large scale, especially on mobile and / or entry -level GPUS.
The reduction in the use of VRAM, the new nerve of war
In this case, the RTX NTC technology was tested in preview by Osvaldo Pinali Doederlein, software engineer at Google, which has engaged in a detailed presentation of its observations on X. For its test, the interested party combined the techno RTX NTC to the DirectX Raytracing (DXR) 1.2 of Microsoft – which is also part of the neuronal renders and cooperative vectors and cooperative vectors and cooperative vectors and cooperative vectors and cooperative vectors To exploit, at best, the capacities of the Tensor hearts of the new RTX 5000.
These famous cooperative vectors allow shaders GPU to work together on small matrix or vector operations. As for the neural compression texture of Nvidia, it operates neural networks to compress and decompress the textures of the games, in order to reduce their size with a minimum impact on quality. Without real surprise at the bottom, the combination of these two technologies sparks.
Osvaldo Pinali Doederlein shows in particular that by activating the Cooperative Vectors And the NTC Ensemble, in “Default” mode, the texture charged by the GPU is rendered at 2,350 fps. On the other hand, when the two devices are disabled, the same texture is made this time at 1,030 fps “only”, marking a difference of almost 80 %.
The engineer also mentions up to 90% reduction for the occupation of VRAM with this combination, specifying that textures alone can represent 50 to 70% of this occupancy rate. Tackling their compression, a fortiori with so much efficiency, therefore makes sense.
Let us specify all the same that the NTC technology is currently only available with the last Preview Pilots 590.26, and that this technology is only accessible on NVIDIA, AMD and Intel graphics cards with their own neural compression devices.
