Cycles-X

I must be an odd one then, I prefer watching my render clean up in the progressive refine mode.

Apparently the slides mention (for Cycles X) that some of the more advanced algorithms do a lot better when they are not forced to work with tiles. Even the adaptive sampling Cycles has now can run into trouble if you make your tile size too small.

1 Like

Yes my GPU is embarrassing! The rig is an “old” I7…
I know I need to upgrade but the longer I wait the more it will be worth changing the whole system.

1 Like

Hi, found the link of the technical presentation on devtalk: https://docs.google.com/presentation/d/13Zo2gafKyIFEyZCBucEfFHrrrwz9RBQoIImC6nBCgEE/edit#slide=id.p

This was posted before but as for the discussion about NLM denoise and tiles, it is all inside the presentation.
Some other interessting information: https://devtalk.blender.org/t/cycles-requests/3729/397

Cheers, mib

1 Like

WHAT ? CUDA WAS SLOWER ???

Microdisplacement is working just fine. It may be only that improvements to it are coming later? It’s certainly not broken like that volume example. I have got another issue, though. This screenshot shows microdisplacement working fine (the bricks), but for some reason the pool table cloth is partially transparent all the way through the image, not even just to an object behind it or anything.

Unfortunately, I haven’t figured out a good way to see just Cycles X issues in the tracker - anyone know the best way to follow this stuff? I can’t even sort search results in that tracker - I hate that UX, it’s like it’s deliberately bad compared to every other issue/bug tracking system I’ve ever used (I am a web developer).

1 Like

Hi, I use mail list bf-committers, as it is now a branch all commits should show up there.
May be you could filter the branch commits in your mail client.

https://lists.blender.org/mailman/listinfo/

Cheers, mib

If you want to follow the progress, here are a few links :slight_smile:

https://developer.blender.org/T87839
https://developer.blender.org/T87837
https://developer.blender.org/T87836

6 Likes

OptiX should be faster than CUDA

1 Like

CPU rendering is 2.5x faster (comparing with 2.92) for me and SSS quality is better (less noise with the same sample count).

2 Likes

Oh now I got it - I thought Cycles X alone was faster than the NVIDIA APIs

Yeah, not a complete fan of phabricator either. Which, if you know which company created it, makes a lot of sense why it is: https://en.wikipedia.org/wiki/Phabricator

Everyone do a render for our homies under the ‘notable users’ section…

1 Like

So a lot of information to sort through on cycles-x. Right now the only improvements is for viewport? Also are tiles no longer a thing for renders? It’s rendering my projects slower…

Strange, my projects render faster in Cycles X. But yeah, it’s still alpha so some features may be buggy.

I didn’t read the whole thread either, but a couple of things I can answer to.

Improvements are for both view port and final render, and it’s incredibly fast right now considering it can use one GPU only. For comparison, in my tests it’s proving to be faster than E-Cycles.
By the way, the view port is so much more responsive now. It reminds me the feeling of when I started using Octane. I’m not sure that I’ll be able to go back to the stable Blender release at this point.

And yes, tiles are gone. Personally, I don’t mind.

3 Likes

Unless optix is not fully implemented yet, When I render with both my GPU and CPU it comes out to still a few seconds longer for render time.

in 2.92 on use Optix GPU only because my CPU is not the greatest…

Hi, what do you meant by Optix is not fully implemented?
For me enable CPU does nothing, cant see any workload in the task manager except feading the GPU with one core. I have a old i5 but RTX 2060.

Cheers, mib

I just tried it on my GTX 1070. Love the progressive rendering. It was indeed faster on the first test i ran, knocking about a minute off a 7 min render time… Main thing is though you can see it clearing up in real time you can then hit cancel at the point you think the noise is acceptable for the denoiser to handle and then you go into the compositor nodes and it will work… So that cut my actual render time down to 44 seconds.
And no time spent guessing how many samples to use. in fact there may as well be a default of unlimited samples at least for single frame renders, and some nice icons or labels that say finish render (or pause render and save for later when they add that) or skip to compositing.
The only thing is i was getting some strange artifacts so its obviously not production read yet hence the Alpha. Also the exposure was different to cycles, i easily fixed this with an exposure node in the compositor. All in all i am chuffed and excited that we are getting free performance increases and better usability at a time when we cant get a hold of graphics cards let alone for retail prices.

3 Likes

LOL, well, at least it’s open source? The screenshots on their site look better than Blender’s though - I wonder when it was last updated. :joy:

Anyway, anyone have any idea on the odd checkerboard transparency issue? There are no transparent materials in my scene except for the plexiglass in the ball return window on the side. :frowning:

New test.
Cycles was running on my optimal 42x42 tile size hybrid render GTX1070+ Ryzen 1700X
Cycles X using its progressive render.


Timing is in the file names… Cycles X way faster and on this scene with lots of glass materials Cycles X wins hands down look how much cleaner that is…
Material nodes for that glass is as below

5 Likes

One more test. Some complex glass shaders.


Observations… Slightly more and better caustics. Much more light getting into the glass stopper, could be a bug such as not including the volumetric absorption or it could just be that cyclesX is allowing more and better light bouncing inside the glass??? Less shadow on the white spheres could just be that CyclesX is again brighter because of better light bouncing… All in all i prefer CyclesX and its fast as hell.

4 Likes