Cycle X (and OpenCL)

Discussion related to the LuxCore functionality, implementations and API.
User avatar
lacilaci
Donor
Posts: 1968
Joined: Fri May 04, 2018 5:16 am

Re: Cycle X (and OpenCL)

Post by lacilaci » Thu Apr 29, 2021 1:24 pm

In other news about recent cycles development...
persistent data!

Luxcore has become crazy fast, however it all goes away if preloading scene each frame takes several times longer than rendering the frame...

Being able to just selectively reload parts of scene would be an insane speedup for animations, even just basic support for flythrough animations would be amazing for some trivial archviz animations..

juangea
Donor
Posts: 297
Joined: Thu Jan 02, 2020 6:23 pm

Re: Cycle X (and OpenCL)

Post by juangea » Thu Apr 29, 2021 2:01 pm

I htink that would mean making LuxCore time aware (animation aware) in some way, so only flagged things are updated in the scene, being an external engine maybe this could be done with the exporter if in LuxCore there were some way of "appending" and "deleting" data from the scene.

I think it's a different approach to the one taken in Cycles, but I also think it may be easier to implement, since from the first frame an analysis can be run, so things are properly flagged on each frame, and that information can be stored and used along all the rest of the frames, so "static" things are exported in one "set" and "dynamic" things are exported in a different or several different sets, that way the engine can delete or append those sets.

Maybe this pseudo-approach is described here could make easier to implement persistent data in LuxCore, not sure :)

User avatar
Dade
Developer
Posts: 5324
Joined: Mon Dec 04, 2017 8:36 pm
Location: Italy

Re: Cycle X (and OpenCL)

Post by Dade » Thu Apr 29, 2021 2:26 pm

LuxCore API already supports it (more or less): you see it in action when you enable viewport rendering and edit the scene.

It is up to BlendLuxCore to redo the same when rendering multiple frames. It has been discussed in the past, it is doable but it is not a small amount of work.
Support LuxCoreRender project with salts and bounties

User avatar
lacilaci
Donor
Posts: 1968
Joined: Fri May 04, 2018 5:16 am

Re: Cycle X (and OpenCL)

Post by lacilaci » Thu Apr 29, 2021 2:31 pm

Dade wrote:
Thu Apr 29, 2021 2:26 pm
LuxCore API already supports it (more or less): you see it in action when you enable viewport rendering and edit the scene.

It is up to BlendLuxCore to redo the same when rendering multiple frames. It has been discussed in the past, it is doable but it is not a small amount of work.
I see, maybe current work on cycles's persistent data will make it easier then.

daros
Posts: 93
Joined: Thu Dec 12, 2019 3:25 pm

Re: Cycle X (and OpenCL)

Post by daros » Tue May 04, 2021 12:52 pm

"lacilaci"
Hi, I'm here reffering to your post of page 4.
I think Nvidia almost did not delivered any performance x Watt improvement during the past 6 years.
A new Nvidia card, if it is 4 times faster as an old one, it consumes 4 times more power.
Not the same we can say about the CPU industry where intel and amd are constantly improving their performance x watt year after year.
Today you can upgrade to CPU's which use 135W which are 6 times faster as an same wattage CPU from 8 years ago.
You see, it depends from the point of view.
Moreover Nvidia is forcing consumers to throw their hardware every two years for just a 60% speed increase. That's a shame. From an environmental point of view this is very bad.
Last edited by daros on Tue May 04, 2021 7:03 pm, edited 1 time in total.

User avatar
lacilaci
Donor
Posts: 1968
Joined: Fri May 04, 2018 5:16 am

Re: Cycle X (and OpenCL)

Post by lacilaci » Tue May 04, 2021 5:05 pm

daros wrote:
Tue May 04, 2021 12:52 pm
"lacilaci"
Hi, I'm here reffering to your post of page 4.
I think Nvidia almost did not delivered any performance x Watt improvement during the past 6 years.
A new Nvidia card, if it is 4 times faster as an old one, it consumes 4 times more power.
Not the same we can say about the CPU industry where intel and amd are constantly improving their performance x watt year after year.
Today you can upgrade to CPU's which use 135W which are 6 times faster as an same wattage CPU from 8 years ago.
You see, it depends from the point of view.
Moreover Nvidia is forcing consumers to throw their hardware every two years for just a 60% speed increase. That's a shame. From an environmental point of view this is bery bad.
4x power consumption for 4x speedup? Sounds like bs to me...

https://cudasteam.com/gpu-compare/Nvidi ... rix-980-Ti

daros
Posts: 93
Joined: Thu Dec 12, 2019 3:25 pm

Re: Cycle X (and OpenCL)

Post by daros » Tue May 04, 2021 7:01 pm

https://www.techpowerup.com/review/msi- ... -x/29.html
Compare 1080 vs 3090, 1.5 X overall performance boost for 2 X power absorption.

here even more clearly
https://www.techpowerup.com/review/msi- ... -x/34.html

daros
Posts: 93
Joined: Thu Dec 12, 2019 3:25 pm

Re: Cycle X (and OpenCL)

Post by daros » Tue May 04, 2021 7:14 pm

as you seeit's even worst...2X power absorption for 1.5x-1.7x performance boost.

User avatar
lacilaci
Donor
Posts: 1968
Joined: Fri May 04, 2018 5:16 am

Re: Cycle X (and OpenCL)

Post by lacilaci » Wed May 05, 2021 4:23 am

daros wrote:
Tue May 04, 2021 7:14 pm
as you seeit's even worst...2X power absorption for 1.5x-1.7x performance boost.
Octane bench shows score of 130 for gtx1080 vs 655 for 3090...
It's one thing that 3090 alone is a special terrible case of crazy power consumption, but to claim 6x for 6x performance, or even 1.5 performance boost is absolutely ridicilous.

The only reasonable case for cpu rendering is memory availabe, even that is only case for really massive scenes.

And if you look at how modern render engines can handle data, or UE5 demo handling crazy amount of textures and geometry... cpu rendering is ded.

daros
Posts: 93
Joined: Thu Dec 12, 2019 3:25 pm

Re: Cycle X (and OpenCL)

Post by daros » Wed May 05, 2021 9:24 am

Hi lacilaci,,
you are right, under certain circumstances the 3090 offers a 5x boost for 2x power. This means 2.5x boost for the same absorption compared to the 1080. This improvement has happened over a 5 year time frame. If we then talk about price/performance then it gets more complicated. The 1080 costed 600 euros wen out and now with the 3090, for 5 times the speed, I have to pay 5 times the price. Where is the deal?
Again, Intel and AMD under CPU can do equal or better.
Obviously, if you want path tracing rendering, you're going to be faster under GPU. But if you want the highest quality possible, you have to go for bi-directional and you still only find that on CPU.
So, Luxcore, that in its manifesto declares that quality will not be sacrificed for the sake of speed, is exactly the software that for now can't give up CPU.
And I think it is wise to work to excel in quality rather than performance because at the end of the day we are dealing with artists and not athletes. Let the corporations aim for performance and let them go at each other. In addition, in the area of quality you work more quietly, you think more and farther ahead, there is less competition because less money moves around. And there's always someone who needs quality and is even willing to come after you. Then, if one day you have the option to offer booth, quality and speed, why not.

Post Reply