My @Peter_shirley style path-tracer is now more than 100x faster than when I started a few weeks ago. Threading, SIMD, cache-friendly layouts, and no recursion. Not bad, but must go faster!pic.twitter.com/CZZpw2fbHr
-
-
Here are some actual perf numbers, for 1280x720 image with 16 samples per pixel: Scalar - 15.4 sec, 2.54 Mray/sec SSE4 - 4.2 sec, 9,32 Mray/sec (3.67x faster) AVX - 2.8 sec, 14.1 Mray/sec (5.64x faster)
-
Nice! I've an implementation in C# using MonoGame. Mine is offline and online (with a serious downscaling of course). I use one thread per core, it's a nice optimization. I tried System.Numerics (SIMD vector/matrix/etc...) but no big changes for me. I'll take a look at AVX ;)
Fin de la conversation
Nouvelle conversation -
-
-
This is an offline renderer, so I'm measuring rays/sec, not FPS. Compared to scalar: SSE4 gave a 4x speedup, and AVX was about 6x. I'm testing one ray against 4 or 8 spheres in my SIMD versions.
Merci. Twitter en tiendra compte pour améliorer votre fil. SupprimerSupprimer
-
Le chargement semble prendre du temps.
Twitter est peut-être en surcapacité ou rencontre momentanément un incident. Réessayez ou rendez-vous sur la page Twitter Status pour plus d'informations.