vray gpu instance forest pack compiles long time

I have done a direct test with redshift and vray gpu. The time to compile if you have lots of instanced objects in forest pack or tyflow particles is insane, it takes certain scenes 1 to 2 minutes to compile while redshift might take 10 seconds. (Just get a detail grass preset from forest pro over a large plane and do the tests your selves)
Another note, if you proxy a forest object or tyflow in redshift, the compilation time for this particular object is down to 0, it reads directly from disk with while if you proxy a forest object in vray it has to recompile it with no added value. (same for tessellation, redshift bakes the tessellation time if you proxy an object - I think this is a very useful thing if applied to vray proxy)
Is this something that vray gpu devs will look into ?

Hey @skafimsk and welcome to the forums,

We are aware of this behavior, I have a few scenes that take a while to compile with FP. The developers will get to this soon
If possible could you please upload your setup for us, I will add it to the samples that our devs will investigate, plus it will be tested with any fixes from our side in future. Thanks!

Best,
Muhammed

any news on this subject. Its getting really frustrating working on big or heavy scenes with Vray Gpu. I feel the compile time for geometry keeps getting longer and heavier with forest pack or heavy scatters.

Hey @skafimsk

This task has our highest priority at the moment, we have multiple developers working on this at the moment(kernel 3 drastically improves time to first pixel in the examples I have)

Best,
Muhammed

I am a Houdini user.
I have the same problem with Instance on GPU.
I hope it will be improved.

Noted, thank you for the feedback @sota_oike
This topic is near the top of our current priority list at the moment

Best,
Muhammed

Hello there,

On this topic, I noticed the exact same problem, wich is a real pain to work on a day to day basis. Basicaly, I switch to CPU to work, and then siwtch back to GPU for renders.
Have you any pre-date of release for the kernel 3 you talked about ?

@FranklinTre Hello,

We are hard at work on getting kernel 3 ready for production, I don’t have a timeline when it will be ready
Time to first pixel with FP is at the top of the current GPU roadmap

Best,
Muhammed

@Muhammed_Hamed would you be able to elaborate on what we can expect from kernel 3 besides FP build times?:slight_smile:

Are portions of Kernel 3 already implemented?

Hello Abraham!

Currently only GPU LC is calculated using kernel 3, everything else is done with kernel 2
Kernel 3 has many advantages, including support for Out Of Core Texutres/Geometry and faster time to first pixel. It is going to allow us to fix some issues we have in the log, and to implement new features in future.
That being said, Kernel 3 is not production ready yet. We are hard at work on getting this done

Best,
Muhammed

We are looking forward to this, I think this is something that needs to be addressed soon especially since all archviz projects tend to be heavy on scattering/textures/geometry.

Any eta for this ?

is there an update or nightlies section for this new kernel ?

Monthly question: Is the new kernel ready? 8)

Hey guys,

We have multiple developers looking at this now, we need more test scenes at the moment. If you can share some of your scenes with us, please Email me. Thanks in advance muhammed.hamed@chaos.com
Best if the scenes are not heavy on shading

Best,
Muhammed

Hello Muhammed,

Thanks for the updates!
Are theire any news on that ? Last update improved drasticaly the Texture “cashing” (don’t know how to say that…). But as said before, we run archviz projects, it’s insanely heavy on FP and RC objects, xref and so on. Just today, I have a scene that compute in 15 sec in CPU and 2 min in GPU on RTX 3090, just to check a set of trees :expressionless:

Can you at least say to us if it’s gonna be this semester or next year, so we can know where to invest our money right now? It’s a serious issue.

Best.

I have this problem too. When GPU render normaly without any slowing issue, frame took 1:40, which is great. But inicialization for a new frame took about 5 minut. Wait, what? That is terrible. Absolutelly useless to render with GPU. Dont know if this is FP or anything else but it is way way off what make any sense of using GPU rendering.
And the scene is not so heavy. Grass, two trees, one villa building, two cars and a pool. Scene is still WIP and long time will be. It will be much heavy at the end, now its light I would say.

Same problems here. It’s not just FP, it’s a clunky experience RE TTFP with choas scatter too.

Throw a rocky ground vray proxy megascan into either and it takes forever to start rendering in IPR. It’s very frustrating. Cpu is much more responsive and tactile. We were looking to upgrade to the latest version soon hoping this issue would be resolved by now

I might be wrong, but it seems that in the latest version of V-Ray Chaos Scatter is faster at generating the scattered objects than Forest Pack. Especially if you are not using the new edge trimming feature.
BTW, I’m anxiously waiting for Kernel 3 too. Any news? :stuck_out_tongue: GPUs are much faster at rendering foliage and displacement than CPU, but the long time to first pixel kind of defeats the purpose, as others have said. I’ve been hearing about this new Direct Storage feature and how this helps the GPU access assets directly from disk. Is there something similar coming to V-Ray?

The same issue - long time lasting to start rendering on GPU…

Hello everyone!

I discussed the topic today with the team, We are aware this topic is important and we are putting more resources to this. It has been on the GPU roadmap with priority for a while as it is not a trivial fix, the plan is to find a good solution to the issue of instancing plugins rather than a workaround for the cases we have logged in our system.

We have a few scenes to test against, 2 of them are from @Alex_M
It will be helpful if you provide us with more scenes, to make sure the fix covers your use case. Please send to muhammed.hamed@chaos.com

@Alex_M Our Dev team lead is looking into Direct Storage now, I’m not sure if it will be helpful to V-Ray GPU or not. I will let you know

Best,
Muhammed