Intel Talks How It Is Enabling High-Fidelity Visuals & Faster Performance on Built-in GPUs, Demos Ray Reconstruction-Like Denoiser For Path Tracing On Arc B580

At SIGGRAPH & HPG 2025, Intel talked about its improvements to visual fidelity & performance for built-in and discrete GPUs.

Intel Is Focusing On Expanding Visual Image Quality & Performance on Built-in GPUs, Arc B580 Path Traced Demo With A Trillion Triangles Showcased Running at 30 FPS @ 1440P With Ray Reconstruction-Like Denoiser

Built-in GPUs or iGPUs (Integrated GPUs) have come a long way. Only a decade ago, iGPUs were used for media and video purposes & everyone knew that gaming on them wasn't great. But a lot has changed in the past few years. Most iGPUs can comfortably offer the performance of entry-level discrete GPUs, and Intel is planning to focus more on expanding the visual fidelity and the performance of these chips.

To achieve these goals for the next-generation of iGPUs, which will also be added to dGPUs, the Blue Team is focusing on:

• Efficiency Improvements for Path Tracing
• Neural Graphics
• New Physically-based effects (like fluorescence)

The first goal is to achieve high-fidelity visual effects, such as Path Tracing, on low-power devices that feature iGPUs. Path Tracing is costly and uses a lot of photon paths to simulate. Even if these paths are carefully sampled, they still need to be denoised, and that is an extra layer of GPU resources being burdened. The solution comes in the form of Resampled Importance Sampling, which improves visual quality by a factor of 10x.

The following is how this improved quality is achieved:

The work, accepted to SIGGRAPH 2025, improves real-time path tracing by enhancing Resampled Importance Sampling.  Samples are organized into local histograms and Quasi Monte Carlo sampling with antithetic patterns is employed, reducing noise with minimal overhead.  Combined with blue noise, this method significantly improves visual quality, achieving up to 10× better results.

This work improves on top of the current state of the art used in AAA games like Cyberpunk 2077 and brings high-end experiences closer to our low-power hardware.

In spite of many challenges, we have come a long way from initial explorations and reconstructing simple scenes to the extremely challenging large scale Jungle Ruins scene featuring animated trees, vegetation, foliage, different materials, dynamic shadows and lighting conditions, fully path traced at 1 SPP achieving 30FPS at 1440p on Intel B580 GPU.

via Intel

Intel is also introducing Open Image Denoise 2 while talking about what comes next. Interestingly, Intel is also focusing on AI-accelerated ray tracing for everyone. The Intel Open Image Denoiser library is very popular due to its open-source nature, and with the second iteration, it introduces new optimized cross-vendor support across all major GPUs (Intel/NVIDIA/AMD).

Intel is also simultaneously working on the development of the next version, which will employ a neural network architecture for improved visuals and performance. The company recently offered a look at its "Path Tracing a Trillion Triangles" demo running on an Intel Arc B580 GPU at 1440p and offering a stable 30 FPS.

Performance and image quality are proportional to the number of rays at each stage of the path tracing.

To save on compute and memory traffic we use 1spp and 1 ray on every bounce. Due to the stochastic nature of path tracing, the rendered image has significant noise. Each pixel is determined by a single random light path, causing extreme fluctuations in brightness and color, especially in complex lighting scenarios such as indirect illumination, caustics, soft shadows, etc. To remove noise and reconstruct details, we use our spatiotemporal joint neural denoising and supersampling model.

via Intel

The following are the main highlights of this demo:

• Reducing the cost of path tracing to achieve real-time performance is a significant challenge and an active area of research in both industry and academia. In this series of blog posts, we share our practical findings on a real-time path tracing of the animated one-trillion-triangle Jungle Ruins scene, achieving 30FPS at 1440p on an Intel Arc B580 GPU.
• This blog series focuses on the practical application of one sample per pixel (1spp) denoising and supersampling, the metric used for visual quality evaluation, handling animations in the high-complexity scene made of 1 trillion instanced triangles, tradeoffs of content creation, and performance.

Interestingly, it looks like Intel is looking to reconstruct the details and remove the noise using a spatiotemporal joint neural denoising and supersampling model, applying both at the same time. This sounds a lot like NVIDIA's Ray Reconstruction, featured first in DLSS 3.5 and later in DLSS 4, and AMD's upcoming Ray Regeneration for its FSR Redstone technology.

• Fine texture details – Denoisers generally tend to produce smoother results because they are optimized to reduce visible noise using a loss function that favors averaging. As a result, finer details can be lost, especially when the model cannot distinguish between high-frequency noise and the actual signal.
• Flickering - While a single denoised frame might look clean, small inconsistencies from frame to frame can result in visible shimmer over time. The inconsistencies can occur due to changes in lighting, motion or lack of temporal context in the model itself. A good temporal loss can encourage the model to keep outputs stable, but if it is aggressively used, we end up with ghosting artifacts.
• Moiré patterns - appear when high frequency details are undersampled, causing interference between the scene detail and the pixel grid. This results in wavy patterns that are not present in the scene but emerge due to insufficient resolution or sampling precision. One way to address this is by training the model on more samples with the textures and structures that commonly cause these artifacts. With sufficiently diverse and representative training data, the model learns to resolve these issues while denoising.
• Shadow Reconstruction – Shadows are inherently tricky for denoisers when there is no supporting information in motion vectors or guide buffers. The model relies solely on noisy color input. When we introduce training samples with different lighting conditions and animation, we see our model gradually learns to reconstruct shadows more effectively.
• Disocclusion – One of the most challenging aspects of our model is handling disocclusions. These occur in regions that were occluded in the previous frame but become visible in the current one due to camera or object motion. Because these newly visible areas lack information from the previously denoised frame, reconstruction becomes difficult. The absence of consistent patterns makes it hard for our model to generalize, sometimes resulting in ghosting artifacts. As with other artifacts, adding diverse and representative training data can help mitigate the issue.

• Reflections – Similar to shadows, reconstructing a reflection model relies only on noisy color input. Providing the first non-specular hit in the auxiliary buffers can significantly improve the quality of the reflection, especially for mirror-like surfaces.

via Intel

Coming back to enabling high-performance and high-quality visuals for low-power GPUs, Intel has hardware-accelerated texture set Neural compression or TSNC, which works in tandem with DirectX Cooperative Vectors. These cooperative vectors leverage the full potential of hardware-accelerated AI capabilities on modern-day chips and can help in achieving up to 47x speedup versus a compute-based implementation that uses FMA (Fused Multiply Add). Some performance metrics were shared:

• Intel Arc 140V (Lunar Lake): 2.6ms (BC6 baseline) / 2.1ms (TSNC with Cooperative Vectors)
• Intel Arc B580 (Battlemage): 0.55ms (BC6 baseline) / 0.55ms (TSNC with Cooperative Vectors)

Intel states that TSNC performs on par or better than regular BC6 compression at a fraction of the texture memory footprint, so you also get lower memory usage, freeing up resources, and getting faster performance in return. Bonus video from Compusemable below for an insight into how NTC brings better looking textures and lowering VRAM usage:

The demos and blog posts by Intel give us an insight into where the company is going next. Intel isn't the Intel we know from the past; it is a different beast altogether, one that is getting ready for the future and innovating tremendously within the GPU segment. Architectures like Xe2 have shown that Intel is a strong contender, both in the entry-level discrete GPU segment and even more so within the integrated space. With these innovations, Intel can really take the integrated GPU segment to unimagined heights, and we hope we really see these implemented soon, thanks to their open-source commitments.

Follow Wccftech on Google to get more of our news coverage in your feeds.