Discussion in 'Article Discussion' started by bit-tech, 14 Nov 2019.
Which was always a dumb implication. You can real-time ray-trace on damn near anything as long as you are willing to sufficiently drop the quality and/or avoid actually doing any ray-tracing. Crytec did this by using extreme temporal supersampling (render once, use the same render result for many subsequent frames) and using lots of precomputed assets (the majority of the reflections) and by using raster lighting & shadowing and manually placing RT reflections only on a handful of selected surfaces. This manual placement shares the same issues manually tweaked raster effects do:
There's also the sillyness going around on many sites of "X is faster than Y in task Z!" with "when you avoid using the banks of dedicated Z accelerators in Y" conveniently left out.
It is quite impressive and it runs like butter on my system. A 3700X + 1080Ti combo gets you 9.1k+ in score.
Didn't run too badly on my laptops 1050MQ with 4Gb of RAM though a less impressive score of 3.1k but it is well below spec. It even runs on the 2400G but is ~20% slower.
My 1080Ti did 13k at default, GPU barely went over 80% so guess I need to up res, looks nice though even ow res and full screen. 1080p was 9.2k, 4k was no really workable hitting similar rates to onboard GPUs at 3k, extra res didn't make it look any better.
would be interesting to see difference between an hardware generated ray tracing and an emulated one.... well if you have many CPU core this might not be a problem
Separate names with a comma.