High-Performance Raycasting on the GPU
How open is the view from every point of a city?
Technion Computer Science project by Yossi Gitelman · Supervisor: Boaz Sternfeld, CGGC · 2026
The openness score comes from published work at the Technion's Faculty of Architecture: from a person's eye, rays go out over the half sphere, and each one is measured to the first surface it hits, up to a distance limit (2,070 rays and 500 m in their published setup). They report about 40 minutes of computing per square kilometre.
This project computes the same measure exactly on the GPU's ray-tracing cores (NVIDIA OptiX), with the CPU (Intel Embree) for comparison, and shows every ray in the browser: for one point, along a route, or for the whole city at once, with or without its big parks. The published setup is the default, and every setting can be changed.
Take a look
How it works
01. Read the result
02. Change the settings
03. Draw a route
04. Score the whole city
05. Every run is saved
06. GPU against CPU
75× faster on the GPU
On the lab PC's NVIDIA GeForce RTX 5090, Manhattan without its big parks (1,268,317 standing points × 2,070 rays = 2.63 billion rays) takes 0.18 s, against 13.9 s on its Intel Core i7-14700 CPU (20 cores, 28 threads): 75 times faster. Same points and rays on both; each time is the median of 5 runs after a warm-up.
​
​Both engines were checked against an exact float64 reference, and the scores reproduce the architects' published table within 2 points in 25 of 28 neighbourhoods.