Intel aims to ease the memory burden of modern games, takes on Nvidia

As part of a video presentation, Intel introduced Texture Set Neural Compression (TSNC). The technology is meant to significantly reduce the memory demands of modern games, which could offer welcome relief for gamers given the ongoing memory crunch. Anyone following recent developments in the gaming space may have noticed that Nvidia recently presented a very similar concept.

The basic idea is the same in both cases: instead of compressing textures and related material data in the traditional way, the data is converted into a learned representation. A small neural network then reconstructs the required information directly on the GPU. This could reduce download sizes, save SSD storage space and cut VRAM usage. That last point in particular is likely to resonate with gamers, as modern titles continue to demand more graphics memory while mainstream graphics cards still often ship with just 8 or 12 GB of VRAM.

According to Intel, there are two variants. A quality-focused mode achieves roughly a 9x reduction with relatively minor losses, while a more aggressive mode is said to deliver around 17x to 18x savings, though at the cost of more visible artifacts. The key question, then, is the trade-off between image quality, memory savings and computational overhead. Nvidia, for its part, cites a practical demo result for its own Neural Texture Compression, claiming a reduction in memory use from 6.5 GB to 970 MB – around 85% – while maintaining almost identical image quality. On paper, Intel’s figures appear higher. In practice, however, the two approaches are difficult to compare directly because they rely on different baselines and test scenarios.

Intel appears to offer broader hardware support

The two companies also differ in terms of hardware support. Nvidia’s Neural Texture Compression is part of the RTX ecosystem and mainly relies on the Tensor Cores in current GeForce RTX graphics cards. Although a more open path now exists through DirectX 12 Cooperative Vectors, Nvidia is still clearly positioning the technology around its own RTX platform and associated tools. Intel is taking a similar route with TSNC by offering an accelerated path for its own hardware, specifically through the XMX units in upcoming Arc GPUs. At the same time, Intel also provides a slower fallback mode via FMA that works even without dedicated AI hardware. Nvidia’s biggest advantage is that RTX graphics cards are already far more established in the gaming market, while Intel’s compatibility strategy appears somewhat more open. Which of the two standards will ultimately gain wider traction remains to be seen.