The short version
All four techniques answer the same question: what colour should each pixel be? They differ in how much of the real behaviour of light they simulate, and how long that takes.
| Technique | Basic idea | Strengths | Weaknesses | Typical use |
|---|---|---|---|---|
| Rasterization | Project triangles onto the screen, then shade each pixel | Extremely fast; huge hardware support | Shadows, reflections and indirect light must be faked | Most games, real-time viewports |
| Ray tracing | Shoot rays from the camera; spawn rays for shadows, reflections and refractions | Accurate mirror reflections, refraction, sharp shadows | Slower; classic form misses soft indirect light | Hybrid game effects, older offline renderers |
| Path tracing | Trace many random light paths per pixel with multiple bounces | Full global illumination, soft shadows, colour bleeding | Noisy until many samples; heavy compute | Film, archviz, Blender Cycles, high-end games |
| Photon mapping | Shoot “photons” from lights, store them, then gather them during rendering | Efficient caustics and some indirect light | Biased; can look blotchy; memory use | Caustics, specialised offline renderers |
In everyday conversation “ray tracing” is often used as an umbrella for every ray-based method, path tracing included. In this article we use it in the narrower, classic sense so the differences are clearer.
Rasterization: what happens when an image is rasterized
Rasterization works “object first”. The GPU takes each triangle in the scene and asks which pixels it covers. Roughly:
- Vertex processing. Each vertex is transformed from 3D world space into 2D screen coordinates using the camera’s view and projection.
- Rasterization proper. The GPU works out which pixels (or fragments) each triangle covers.
- Depth testing. A depth buffer, or z-buffer, stores the distance of the nearest surface at every pixel, so a triangle behind another is discarded.
- Fragment (pixel) shading. A shader calculates the colour of each visible fragment from textures, lights and material settings, using a lighting model such as Phong or a modern physically based model.
- Output. The finished pixels are written to the frame, often followed by post-processing.
What is rasterization in a GPU? It’s the fixed-function stage of the graphics pipeline that converts triangles into pixel fragments. Modern GPUs are built to do this billions of times per second, which is why rasterization is so fast.
The limitation is that each pixel only “knows” about its own triangle and the lights. It can’t naturally see what’s reflected in it or what’s blocking light from it. So rasterized renderers fake those effects with extra techniques: shadow maps for shadows, environment maps and screen-space reflections for reflections, light probes or baked lightmaps for indirect light, and screen space ambient occlusion (SSAO) for contact shadows. These tricks can look excellent, but they break down in certain situations, such as reflections of objects that are off-screen.
Ray tracing: following rays from the camera
Ray tracing works “pixel first”. For each pixel, the renderer fires a ray from the camera into the scene and finds the first surface it hits. In the classic approach described by Turner Whitted in 1980, the renderer then spawns new rays from that hit point:
- Shadow rays towards each light, to check whether something blocks it.
- Reflection rays in the mirror direction, for shiny surfaces.
- Refraction rays through transparent materials such as glass or water.
Those secondary rays can spawn further rays, so you get reflections of reflections and correct views through glass. To make this fast enough, renderers organise geometry into an acceleration structure, usually a bounding volume hierarchy (BVH), so each ray only tests a small fraction of the triangles. Since 2018, consumer GPUs from NVIDIA, followed by AMD and Intel, have included hardware to speed up these ray-triangle and BVH tests.
Classic ray tracing is deterministic and produces crisp reflections and shadows. But it only follows a few specific directions at each hit point. It doesn’t capture light bouncing diffusely off a red wall onto a white floor, and its shadows are hard-edged unless you add extra sampling.
Path tracing: ray tracing taken to its conclusion
Path tracing, based on James Kajiya’s 1986 rendering equation, extends ray tracing to all of light’s behaviour. At each hit point, instead of following only mirror or refraction directions, the renderer picks a random direction based on the material and continues the path. It repeats this for several bounces. Each pixel averages many such random paths, which are called samples.
Because it uses random sampling (a Monte Carlo method), path tracing naturally produces:
- Soft shadows from area lights
- Indirect light and colour bleeding
- Glossy reflections of any roughness
- Depth of field and motion blur
- Caustics, though some are hard to sample efficiently
What’s the difference between ray tracing and path tracing? Classic ray tracing follows a few fixed rays per hit and mostly handles direct light, reflections and refraction. Path tracing follows random paths through many bounces and simulates full global illumination. The price is noise: with too few samples the image looks grainy, and it only clears up as samples increase. That’s why modern path tracers rely heavily on denoisers, both AI-based and traditional. If you use Blender, our guide on reducing noise in Cycles renders covers this in practice.
Most offline production renderers today, including Cycles, Arnold, RenderMan, V-Ray and Corona, are path tracers or path-tracing hybrids.
Photon mapping vs ray tracing
Photon mapping, developed by Henrik Wann Jensen in the mid-1990s, is a two-pass method designed to capture lighting that ray and path tracing struggle with.
- Photon pass. The renderer emits many “photons” from each light. They bounce through the scene and are stored where they land, in a data structure called the photon map.
- Rendering pass. A ray tracer renders the image as usual, but at each visible point it estimates indirect light by looking at how densely photons gathered nearby.
The big win is caustics: the bright, focused patterns light makes through a glass of water or off a ring. These are very hard for a camera-first path tracer to find by chance, but easy for photons travelling from the light. Photon mapping also gives smooth indirect light with less noise than plain path tracing at low sample counts.
The downsides: photon mapping is biased, because the density estimate blurs lighting slightly. It can produce blotchy artifacts if too few photons are used, and storing millions of photons costs memory. Progressive variants reduce the bias over time. Today photon mapping mostly appears as an option alongside path tracing, for example for caustics or as a GI cache, rather than as the whole renderer.
So photon mapping vs ray tracing isn’t quite either-or: photon mapping uses ray tracing for its rendering pass and adds a light-first pre-pass on top.
Hybrid rendering in modern games and engines
Real-time ray tracing in games is almost always hybrid. The engine rasterizes the main image, which is fast and reliable, then uses ray tracing selectively for effects that rasterization fakes poorly: reflections, shadows, ambient occlusion and global illumination. Upscalers and frame generation (DLSS, FSR, XeSS) and real-time denoisers make up for the small number of rays each frame can afford.
A few games now offer full path-traced modes. Examples include Cyberpunk 2077’s “Overdrive” mode, Alan Wake 2, and remastered versions of older games such as Quake II RTX and Portal with RTX. These replace most of the faked lighting with path tracing at a heavy performance cost.
What this means in Blender and other 3D tools
- Blender EEVEE is built on rasterization, with screen-space and ray-traced extras. It’s fast enough for real-time previews.
- Blender Cycles is a path tracer. It’s slower, but it handles glass, indirect light and soft shadows properly. See our Cycles vs EEVEE comparison.
- Real-time archviz tools like Twinmotion, Lumion, Enscape and D5 use hybrid approaches, and some offer path-traced output modes.
- GPUs speed up both, but in different ways. Rasterization uses the traditional pipeline, while ray and path tracing benefit from dedicated RT hardware. More on that in GPU vs CPU rendering.
Gaming questions answered
Does ray tracing make a difference? It depends on the game and the effect. Ray-traced reflections and global illumination can change the look of a scene a lot, especially in environments with lots of glass, water, or indoor bounced light. In games with heavy baked lighting, the difference can be subtle. Many players find it worthwhile only when their hardware can keep frame rates comfortable.
Does path tracing reduce FPS? Yes, substantially. Path tracing is the most demanding lighting option in games that offer it. Even top-end GPUs typically rely on upscaling, and often frame generation, to reach smooth frame rates with it on.
Can an RTX 4090 do path tracing? Yes. High-end cards like the RTX 4090 can run path-traced modes in current games, though usually with DLSS upscaling at high resolutions. Lower-tier cards may run them at reduced resolutions or settings.
Is the RX 9070 XT bad at ray tracing? AMD’s newer RDNA 4 cards improved ray-tracing performance noticeably over earlier generations. In the heaviest path-traced titles, NVIDIA’s high-end cards have generally kept an edge. Independent benchmarks for the specific games you play are the best guide.
Cyberpunk 2077: ray tracing or path tracing? If your GPU can hold a frame rate you’re happy with, path tracing (Overdrive) gives the most realistic lighting. If not, standard ray tracing, or even selected ray-traced effects, is a better balance. It’s a matter of taste and hardware, so try both in the same scene.
Is rasterization still used?
Very much so. Rasterization remains the backbone of nearly all real-time graphics: games, VR, user interfaces, CAD viewports and 3D app viewports. Even games marketed for ray tracing rasterize the primary image and add ray-traced effects on top. Path tracing has taken over offline rendering, but for anything that must run at 60 frames per second or more, rasterization isn’t going away soon.