Cycles X test with my Contrasted Room Scene.
Workarounds so far:
Volumetrics used for mist > planes with transparency
Multi-gpu > Multiple blender sessions with Gpu1 start frame 1 step 3; Gpu2 start frame 2 step 3; Gpu 3 start frame 3 step 3; etc. (step 3 because of a 3 gpu rig of course)
often times I have issues in a couple of frames caused by persistent data…
But I am having fun with X.
April 30 build 43789b764d9a crashes with this message:
Illegal address in cuMemcpyDtoHAsync( mem.host_pointer, (CUdeviceptr)mem.device_pointer, mem.memory_size(), cuda_stream_) (C:\b\buildbot-worker-windows\windows_custom_branch\blender.git\intern\cycles\device\cuda\queue.cpp:203)
Its not the heaviest scene in the world but it is not the lightest as well. 3866.48M used.
1dea1d93d39a , the earlier, works fine here.
Do you have a lot of procedural materials? Most of my tests are only a few minutes faster. But I’ve got some where the procedural materials are so heavy that you can’t really use eevee/material preview mode. In those the cyclesX render time is cut in half compared to 2.92.
Hi, a user on devtalk report the same and the solution.
1792x2450 resolution my necro gtx 960 4gb
just 120 samples + default checkbox OIDN (no nodes) + adaptive sapmpling
Beside the speed increase, is there any talk in Cycles X development to finally bring a proper subdivide and tesselate for Height Maps on render function? At the moment the current implementation for using Displacement Maps is clunky at best, even with the Experimental functions enabled. It should really be just a simple selection in the Render settings like…oh I don’t know…every other render engine out there?
Please tell me someone is working on this!
19 posts were split to a new topic: Why Cycles-X and not Eevee?
Is there somewhere a complete list of what Cycles-X does not touch yet.
I see YAFU’s Post
But I think the listing is just a small sample.
After a quick test, where my jaw dropped at first when I saw the render speed, I noticed that the Shadow Catcher and AOV’s don’t work.
Some things are mentioned in the official blogpost: code.blender.org - Cycles X
There are also some developer tasks that give a more complete overview:
I have edited my post and added those “lone_noel” links.
No procedural materials. The weird is that models with similar materials often work and often does not.
Right now, the biggest drawback i can see with Cycles-X is no Branched Path Tracing. I am assuming that this is removed for better sampling options in the future? however if you are using many lights, it is a significant regression to normal cycles.
From what I understand breckt feels that he can make one integrator that will make branched path tracing obsolete.
Many-light sampling should be able to remove that sampling regression and then some (because for starters, it would also work with emission materials).
The point is to sample each shading component and every light in a smart way rather than just brute-forcing everything (which is to sample every shading component and every light for every sample, regardless of their contribution).
Cycles X IS ABSOLUTELY DOPE!
Guys, whatever you did smoke, i also want a piece of it! This thing is so freaking FAST AND SNAPPY i got an 8 hours RENDERGASM!
WOW!, just WOW!
Just for fun I tried it on CPU only, and it is still snappy. You have the feeling that you work on an older GPU, while being on CPU!!! DOOOOPE!
I am already migrating to Cycles X, and only if i can’t use it i go back.
Now only 2 wishes remain…
- A GPU domainless fluid/smoke sim like nvidia FLIP VDB voxels, and
- a good caustic solver for Cycles X.
Then i have no more needs at all.
You are amazing guys!
This is the Cycles X we need, but not the one we deserve. Absolutely AMAZING!
The SSS is so freaking FAST I can’t handle it! ITS INSTANT!
I for one, welcome our Brecht & Co overlords…
That’s what I call enthusiasm!
Now change your nick to XselcyC…
Are you sure, I think we need the missing features back first (so many of my scenes use both volume shaders and AO Color/Bevel shading).