an OSL shading node such as Code Snippet is nice.

A camera sequencer similar or better than the one in maya.

if a scene has tenthousands of lights, now Arnold can render it very very very fast. wish Karma can do it too.
katana's renderboot can save much memory, wish solaris's husk can do better than it.

1. native sparse GPU solver
2. Karma XPU pyro shader that works with scattering
3. Nvidia's Neural vdb to save 100x disk space on volume sims
4. fixes to the Layout LOP in Solaris since its quite buggy and needs some fixes for "paint brush" pscale.

That promised new quad remesher pretty please.

I'd welcome a separation of manipulator orientation picking from picking its location.

Currently, "Start Orientation Picking" moves the manipulator to a different location. So "orientation picking" is a bit misleading.

To counter that (in order to maintain manipulator's position), we need to temporarily store Translate values (or Pivot Translate if we're operating on a pivot) and then after orientation is "picked", paste those values back.

This significantly slows down the workflow, especially in modeling tasks where such operations are frequent.

luoqiulin
an OSL shading node such as Code Snippet is nice.

In material context press the tab and type OSL, you will see a node like a snippet for the OSL.
Isn't it what you want?

360VR camera in Karma.

(if this is already possible please let me know)

Houdini is amazing. The only thing I desperately, desperately wish is for Karma (normie and/or XPU) to be faster. I'm cranking out tons of renders in Maya and Arnold now, won't win any speed awards but it's useable. I tried the same scene in Karma XPU (XPU!) and it was EIGHT times slower than Arnold to get the same noise level. It's the difference between finishing a set of 20 stills in 5 hours versus 40... I love how Karma looks and I love Houdini, but I just flat out can't use it for the projects I'm doing right now. It's a shame.

What are you rendering? Karma in general should be pretty quick, XPU for sure, but I'm curious what
you're actually rendering that could be 8 times slower. Doesn't sound right at all.

Do you have some examples you can share?

L

BrianHanke
I tried the same scene in Karma XPU (XPU!) and it was EIGHT times slower than Arnold to get the same noise level

I agree with lewis_T, this doesn't seem right...

What exactly are you rendering?
Are you noticing this slowness in the viewport? or just offline? (ie husk)
Is it possible your GPU has run out of memory, and its falling back to CPU-only?

thanks!

brians

BrianHanke
I tried the same scene in Karma XPU (XPU!) and it was EIGHT times slower than Arnold to get the same noise level

I agree with lewis_T, this doesn't seem right...

What exactly are you rendering?
Are you noticing this slowness in the viewport? or just offline? (ie husk)
Is it possible your GPU has run out of memory, and its falling back to CPU-only?

thanks!

Here's a quick comparison in the scene I'm working on now. Matching time: 15-20m to Arnold finished frame compared with 15-20m XPU, around 500 samples. Cleaning up XPU to match noise level requires 2000+ samples, looking at 1.5 hours or so.

Ok, so I can tell by the Optix readout in the first image, that the GPU has not run out of memory.
We have already improved XPU glass for the next version of Houdini.
Is it possible for you could post a version of this scene I could use for testing?
Thanks!

brians
Ok, so I can tell by the Optix readout in the first image, that the GPU has not run out of memory.
We have already improved XPU glass for the next version of Houdini.
Is it possible for you could post a version of this scene I could use for testing?
Thanks!

Here you go, thanks for your prompt attention. SideFX rocks!

I took a look at your scene. A few points...

1)
Sadly the new feature "direct refraction subset" didn't do anything to improve the glass rendering. It's still the same level of noisy.

2)
I noticed the level between GPU and CPU was fairly low.
It was Optix=85% CPU=15% on my CPU.
Normally I get something like Optix=97% CPU=3% (I have a crappy CPU)


This indicates that the GPU is not working as fast as it could be on your scene. The high level of refraction bounces (ie 10) means there will be a lot of a thing called "divergence". We don't have a quick fix for that, but NVidia's latest GPUs now have special hardware to address it. Its called "shader execution reordering". So in the next version of Houdini, and with an ADA Nvidia GPU, we can expect refraction scenes to run much faster.

3)
KarmaCPU supports a feature called "automatic convergence mode". This is a fancy name for adaptive sampling. XPU does not support this yet, but we have plans to support something like it in future versions. It's the reason I think Arnold looks like it has less noise in this refraction-heavy scene. So again no quick fix sorry, but we do have this kind of thing on our radar.

thanks very much for providing the scene, sorry I could not be of more help.
Brian

Thanks for looking into it! I'll keep checking back when new major versions of Houdini are released. Solaris/Karma has come so far so fast, I won't be surprised to see equally exciting developments going forward.

brians
Sadly the new feature "direct refraction subset" didn't do anything to improve the glass rendering. It's still the same level of noisy

Oh, and I should clarify this comment here.

"direct refraction subset" didn't do much for your scene, but that doesn't mean it doesn't do anything. It does help in other refraction-related cases. Here it is applied to a glass sphere with a dome light with HDRIHaven_skylit_garage_2k.rat, rendered with sample-count of 1.

It's been implemented in KarmaCPU for a while but is now working in KarmaXPU. Sadly due to the large amounts of code changes, we are unable to back-port it to H19.5, so it'll have to get released in the next Houdini version.

Off


On

brians
sorry I could not be of more help.

And I was also probably too hasty to say this.
After chatting with a colleague there might be some more stuff we can do to improve refraction noise.
We'll let you know of any developments.

Brian

brians
After chatting with a colleague there might be some more stuff we can do to improve refraction noise.
We'll let you know of any developments.

Sounds good! I was testing my scene with Karma CPU using the diagnostic AOVs. Most layers clean up just fine, but glossy transmission and indirect glossy reflection need massive sample multipliers thrown at them. Would love to see some improvements in that area in future Houdini releases.

While we're on Karma XPU/MtlX

- are there any plans for an up to date Physical Sky implementation for Karma?
- also decoupling IOR for Specular and Refraction would be nice to have for MtlX
- will we see Transmission Depth, Scatter etc. implemented for H20?
- will there ever be nested dielectrics for Karma XPU?

brians
I took a look at your scene. A few points...

1)
Sadly the new feature "direct refraction subset" didn't do anything to improve the glass rendering. It's still the same level of noisy.

2)
I noticed the level between GPU and CPU was fairly low.
It was Optix=85% CPU=15% on my CPU.
Normally I get something like Optix=97% CPU=3% (I have a crappy CPU)

Image Not Found

This indicates that the GPU is not working as fast as it could be on your scene. The high level of refraction bounces (ie 10) means there will be a lot of a thing called "divergence". We don't have a quick fix for that, but NVidia's latest GPUs now have special hardware to address it. Its called "shader execution reordering". So in the next version of Houdini, and with an ADA Nvidia GPU, we can expect refraction scenes to run much faster.

3)
KarmaCPU supports a feature called "automatic convergence mode". This is a fancy name for adaptive sampling. XPU does not support this yet, but we have plans to support something like it in future versions. It's the reason I think Arnold looks like it has less noise in this refraction-heavy scene. So again no quick fix sorry, but we do have this kind of thing on our radar.

thanks very much for providing the scene, sorry I could not be of more help.
Brian

I do have to say that reading this thread I’m amazed at the level of Client interactions and support that SideFX provides! AMAZING!