My first test render are very fast but view port render does not work at all on Linux and some files are not rendering at all.
I am testing on Windows now.
It is all experimental and new but may be a nice week end fun to test.
Update: The render finished. Regular Cycles in Blender 2.92 took 18 minutes and 14 seconds
My next test scene in Cycles in 2.90 estimates it will take 1h30m. I’m not using adaptive sampling so I think that’s probably an accurate estimate. In X-Cycles it seems to be rendering at about 1 sample per second so with 3000 samples I guess it’ll take 50 minutes. Since it’s progressive rendering the whole image at once it’s possible I might choose to stop it earlier than 50 minutes if it looks good enough.
Update: X-Cycles took 49 minutes and 26 seconds.
EDIT: The issue seems to be VRAM related cause the rendering was pretty high resolution. And that’s probably why a lot of users went “WHAT?” when it was announced that tiled rendering will be ditched. That was a way to render demanding scenes on older yet still capable GPUs anyway (mine has 6GB, which isn’t that bad). So I hope the devs will find a way to make their progressive approach VRAM-friendly or they might disappoint a lot of users. It would be similar to Luxcore which also uses progressive rendering which makes bigger scenes pretty much impossible to render on a GPU with a normal amount of VRAM.
I know that Luxcore has tiled path rendering. But it’s not usable. It must have been an experiment that was never properly finished (even Dade admitted so in his recent post concerning X-Cycles). It’s not only slower. It’s a nightmare to set up and it does deliver different (usually worse) results than the ordinary rendering. And it can only take you a bit further as far as VRAM is concerned. It doesn’t get you anywhere near the dimensions that are possible in Cycles.
And yeah - I have a feeling the tiled rendering isn’t completely off the table for X-Cycles. We will see.
Some forum pal told that Octane solves the no-tile ram issue by placing the actual pixels on the system ram, freeing the gpu ram for geometry texture etc.
Hope it’s a viable solution, or maybe there are better ones to explore
I ran my last project through the new Cycles build, then did it again on 2.92 to compare. Even on my ancient machine, the speed differences are tremendous. It took a little over 2 1/2 minutes on Cycles X, while oldschool Cycles took around 8.
My comp’s a little too old to enjoy some of the extra perks, like the super responsive render viewport, but still…damn.