Of course, I posted the videos and stills, everyone can then decide if E-Cycles is fast enough and looking good enough for their own need.
To follow up if anybody had the same issues !
I tried to render with 256256 with lower sampling, and this time it worked correctly. It does seems to be a driver problem. I tried rendering with 3232, 6464 and 128128 tiles and i didn’t notice any significant performance diference between all of them with pure GPU rendering, so it’s good for me .
I noticed a parameter in the preference setting, in the system tab, named “texture time out”. Its default value is 120 seconds, could it have some influence ?
Good to know your problem is solved.
The texture timeout iirc is a garbage collector parameter, it frees them from memory if some textures are not needed anymore.
The best is indeed to use auto tile size.
As I mentioned earlier, it would be quite easy if I used Blender only. For example I extensively use Apple Motion - it is for me much more efficient than After Effects for simple 2D/2,5D jobs; recently started working on AR app for iPhone; I cooperate with several video editors and no one use Windows and so on and so forth… Fast hardware is no problem.
I understand. I also used Apple Motion on macOS. Great editor for 2D animation. I’m now planning to do such jobs with Blender as well (using the 3D editor and VSE). I’d rather use one tool for as much purposes as possible, in stead of switching back to After Effects.
is a built with the enhanced viewport improvements already available?
It will be the next monthly feature update, so in 4 days. Note that it actually came from a WIP I made in another thread, that a moderator brought back here. It was tested and coded like in a week or so. Because now a lot of users seem to want it like now, I release it. The default cube may be slower, it may burn your cat, etc… it is only tested on professional archviz renders with lots of lights and/or glass and complex materials.
Another preview with SSS everywhere (the bread has it, the fruits, the cheese, the cakes, etc.), loads of refraction bounces through glass, wine, water:
still even with SSS covering a fourth of the image, under 4 sec, otherwise, under 2sec.
went back to OIDN a bit, 3sec for the BMW scene at default resolution:
7sec at 100% (4 times the pixels)
WoW! Impressive! Wondering how my Titan RTX in e-cycles will perform than
For High end cards, I recommend to take the 2.7x version. It’s the version I’m developing on, so you get some preview of WIP features like the BVH building being computed only once for videos were only the camera moves or faster OIDN denoising. It makes a big difference, because in current state, the path tracing phase becomes very short, so reducing the preparation phase and the denoising phase greatly reduces the overall time.
Thanks for your great work. Just tested E-Cycles with my usual interior scene and the speed up is AMAZING!
One question: I realize a big improvement on render time with path tracing, however I can’t reproduce the same speed up when using branch patch tracing. Is it normal? And if so, Have you got plant to optimize branch patch tracing as well?
It’s very important with my work because my clients usually want an interior scene with so many light source and I need to use “Sample all direct lights” feature.
Thanks for your great work again! It helps me so much!
Is there anything special needed to get this to start going? I unzipped the Feb 2.8 version and opened blender and changed it to cycles experimental. I do see the Auto Tiles checkbox so that is activated. I don’t have any speed increase at all. I’m sure i did something wrong, please let me know
PS: I am running windows 10, 7 1080Ti cards.
Hi and welcome on board
Thanks for the feedback, Branched Path Tracing indeed gets less speedup at the moment. There is room for improvement and I’m already working on it, it’s just you are the first one using BPT, so I didn’t prioritize it until now.
It’s detailed in the FAQ, but I made it more obvious that the current biggest speedup is with regular path tracing in the first post.
I’m happy if it helps you
Do you mean you have 7 x 1080 Ti on windows 10 or you tried both windows 7 and 10?
Normally it’s just hit “render” and you get the speedup, no need for experimental or whatever. Always use the latest version (currently it’s the v20190221). If you still have problems, you can send me some logs, system information and/or a file per PM to reproduce the bug.
Yeah 7x1080Ti on Windows 10. Well my apologies, I looked through the thread and saw the viewport enhancements are only coming in 4 days. When I used the actual f12 render the results were 37 seconds instead of the default 45 in a completely unoptimized random scene. Great work man. I’ll keep doing more tests but glad to know its all setup right
There is no real speed up if there are more than three cards connected on the same motherboard. It’s bandwidth limited.
Oh my, I would like such a setup Glad to hear it works in such configurations, I would have expected the overhead of syncing 7 cards to minimize the benefits a lot. Good news
updates coming soon.
What are you talking about? Single workstation CPU has 44 (Intel) or 64 (AMD) PCIE lines. 2x CPU doubles the number of PCIE lines. Where is ‘bandwidth limitation’?