Intel has been making significant strides in visual fidelity and performance for both integrated (iGPU) and discrete GPUs. At SIGGRAPH & HPG 2025, the company unveiled advancements in high-fidelity visuals and performance enhancements, particularly with its integrated GPUs, which have dramatically evolved over the past decade. Once considered incapable of gaming, modern iGPUs now match or even outperform entry-level discrete GPUs in many cases.
Intel is focusing its efforts on expanding the capabilities of these chips to deliver even higher performance and visual fidelity, making next-generation gaming experiences accessible on low-power devices.
One major breakthrough is the optimization of Path Tracing for iGPUs. Path Tracing, a computationally intense rendering technique, has traditionally been reserved for high-end hardware. Intel is working to bring this technique to lower-powered GPUs, using a method called Resampled Importance Sampling (RIS) to improve the visual quality by up to 10x. RIS organizes samples into local histograms, utilizing Quasi Monte Carlo sampling and blue noise patterns to reduce noise and improve image quality without significant performance overhead
. This enhancement has been proven to work on complex scenes like the Jungle Ruins scene from the game Cyberpunk 2077, showing 30 FPS at 1440p on the Intel Arc B580 GPU.
Alongside RIS, Intel has also introduced Open Image Denoise 2, which optimizes ray tracing across different GPUs. This open-source library has gained popularity for its ability to work seamlessly across Intel, NVIDIA, and AMD hardware. Intel is also focusing on AI-accelerated ray tracing to further boost performance and visual quality.
They have demonstrated this technology in a “Path Tracing a Trillion Triangles” demo on the Arc B580, achieving stable 30 FPS at 1440p. To improve rendering efficiency, Intel uses one sample per pixel (1spp) and ray per bounce, which helps reduce compute and memory traffic, although it can introduce noise that requires denoising for visual clarity.
To tackle this, Intel employs a spatiotemporal joint neural denoising and supersampling model. This method is similar to NVIDIA’s Ray Reconstruction featured in DLSS 3.5 and 4, aiming to reduce noise and improve fine details in reflections, shadows, and disocclusion. The goal is to reduce flickering and moiré patterns while maintaining stable frame output and accurate visual effects.
Intel also introduced hardware-accelerated texture set Neural Compression (TSNC), which works in tandem with DirectX Cooperative Vectors to speed up texture compression. This can offer a significant performance boost, achieving up to 47x faster speeds compared to traditional compute-based implementations, with reduced texture memory footprint.
With these advancements, Intel is positioning itself as a major contender in both integrated and discrete GPU markets. As the company continues to innovate and make these features open-source, it will likely push the boundaries of what is possible with integrated GPUs, making high-performance visuals accessible to a broader audience.