PDA

View Full Version : GPU render for lightwave 2018 Will it ever exist?



genesis1
07-03-2018, 07:22 AM
I'm running x99 system with a i7 5820k 6 core 12 thread processor and VPR doesn't half thrash the cpu fan. I think it would be a great help if an option for gpu rendering even just for vpr so as to help the cpu workload. At the moment its like a double thrash, as vpr does it first and then when you render out you get it again. Kind of expect it on render out, but would help the fan wear if we had gpu support for VPR. What do the rest of you guys think? :confused:

RPSchmidt
07-03-2018, 08:16 AM
I personally think yes, it will exist... it's really just a matter of time and priorities, but I think it's inevitable.

For my part, while I would love to have GPU rendering, right now, my personal preference (besides bug-fixing) is that the developers focus on full Substance integration. The orders of magnitude of speed I would gain in my own development process from that would be huge.

erikals
07-03-2018, 11:43 AM
don't think they have a choice.

i recall Jay Roth back in the days (15 years ago?) said yes, GPU rendering will come to LightWave eventually.

https://i.imgur.com/eQ3JVzu.gif

TheLexx
07-03-2018, 12:13 PM
Re original post, some users elsewhere have observed that cpu is "kinder" on rendering wear and tear than gpu, but I'm puzzled by the cpu fan being any particular issue since a decent cpu would surely also have a decent fan like a Noctua or similar, especially if user built. Re gpu coming to native LW, does anyone know if any of the current render engine rewrite allows any groundwork for gpu, or are the two concepts not related ?

Marander
07-03-2018, 01:28 PM
Many / most renderers offer Hybrid mode.

But I don't see it happening to LW, GPU render development requires specific skills and large dev efforts.

Marander
07-03-2018, 01:33 PM
I agree.

For dev efforts and complexity, it depends on the base architecture but shows that for previous CPU only engines like Vray or Arnold, the efforts to allow GPU / Hybrid rendering are substantial.

devin
07-03-2018, 01:41 PM
Compared to 2015, VPR in 2018 wreaks havoc on my MSI Ghost laptop. The laptop begins to generate a ton of heat and the fan kicks into overdrive. It's another reason that I've been avoiding using 2018.

rustythe1
07-03-2018, 01:44 PM
As I said in another thread, GPU is bottlenecking badly, where as CPU has been accelerating fast, if you compare the 680 780 980 and 1080 there actually isn't that much advancement in speed and power, lot more and faster ram, but fps is probably only up 130% between those cards, and also GPU revisions got to a point of 6 months, but now its been 4 to 5 years for a major upgrade, where as we are seeing CPU cores x2 every 2 years again, I think less this time with the 28 core coming out q4, and amd are revising faster and faster getting back to how it was in the early 2000s Intel re road mapped as originally I think the next revision was way of an this year was only going to be mid i7 revisions, but now a whole new lake is on the way, I still think CPU is far more user friendly and customisable, GPU is limited to to a smaller range, what happens if nvidia discover the next best gaming tech and drop cuda? Its an area where the technology is too dependant on each other, I think CPU needs to start looking more towards paralleling, then that will become a game changer,

erikals
07-03-2018, 02:56 PM
GPU is 3 times faster average,
+ there is room for 3 cards in one machine, and CPU power on top of that.

look at how Blender flies with GPU. in addition it supports (of course) CPU.

GPU+CPU is the future. (imo)

hrgiger
07-03-2018, 05:00 PM
Don't know what you mean rustythe1, GPU just smokes CPU. And its much easier to add additional GPU processing.

And no, GPU rendering won't come to LW anytime soon. Either way, no point sitting around wishing for it, just get Octane.

erikals
07-03-2018, 05:19 PM
i think he just refers to the rumor/trend...

GPU advancement is decreasing
CPU advancement is increasing

...not convinced AMD will continue the trend  (though i hope so)


as for GPU...
https://www.pcgamesn.com/intel-moores-law-no-more
https://www.forbes.com/sites/kenkam/2018/04/23/how-moores-law-now-favors-nvidia-over-intel/#190e61a65e42

so, seems like CPU passing GPU is a rumor, not a trend (?)

rustythe1
07-04-2018, 05:42 AM
Don't know what you mean rustythe1, GPU just smokes CPU. And its much easier to add additional GPU processing.

And no, GPU rendering won't come to LW anytime soon. Either way, no point sitting around wishing for it, just get Octane.

I didn't mean its as fast at the moment, my point was that development of gpu has stalled and its bound too as its limited to its PCIe and cpu is catching up at a faster rate, you may also find the new cards will be slower with some apps until they are redeveloped as the volta platform will likely be half cuda core half tensor core (as it already is on AI platforms), they could then drop cuda all together as tensor is 3x as fast, and everything becomes obsolete again?

erikals
07-04-2018, 05:57 AM
and the price of Tensor will be?
https://www.extremetech.com/computing/263751-googles-ai-cloud-tensor-processing-units-now-available-public-machine-learning-beta-test

i don't really care, just add power / price that can match rendering with 1 CPU + 3 GPU

rustythe1
07-04-2018, 06:22 AM
the quadro is already tensor, gv and gp100, and as far as I knew the volta platform was going to have tensor too? but maybe only in titan to begin with as the gv is Ģ8000 uk and your only allowed to own 1 unit! (looks like they are trying to limit bit mining?)

prometheus
07-04-2018, 06:44 AM
GPU is 3 times faster average,
+ there is room for 3 cards in one machine, and CPU power on top of that.

look at how Blender flies with GPU. in addition it supports (of course) CPU.

GPU+CPU is the future. (imo)


yup, it does, huge difference when you switch to render blender hair between CPU and GPU ..whe GPU is much much faster, same with rendering prinicpled materials and product shots...and that comparison goes between blender and lightwave as well.
it also renders fluid smoke much faster...Though when you got too much going on with shaders ..hair,, or too large fluid sims, the GPU sometimes fails, and you need to switch to CPU.

With blender....You at least have a choice and can render some stuff very quickly..and only when you are going off limit you resort to CPU..while you are locked to the slower CPU rendering in lightwave, unless having octane of course.

erikals
07-04-2018, 09:40 AM
that's a minus with Octane, if you want to use 8 GPUs, it gets to be expensive. then again, the industry should be able to afford it.

and this is yet another reason for NewTek to get into the GPU game.

https://i.imgur.com/5iDKqDX.gif

lwanmtr
11-25-2018, 04:21 AM
Every program I use (except Adobe) has a gpu render option included...even Daz and Poser...For Lightwave to remain competitive, I think it will need to have at least a hybrid.
Shouldnt have to get a 3rd party renderer like Octane to get it.

prometheus
11-25-2018, 05:40 AM
I would have to get octane if I wanted to render teh 2018 volumetrics with decent speed, from native volumetrics to TurbulenceFD, the cpu rendering of TFD is just too slow for my liking, and a part of why I find myself utilizing GPU rendered fire and smoke in """ that is way faster.

I could of course go for Lightwave octane and Tfd, though that means apart from the intital lightwave cost ..additional cost for tfd, and octan..while It cost me nothing as it stands now.

pixym
11-25-2018, 09:43 AM
On July of this year (2018), I try again Octane (V4 with denoising) and right now I have just given native LW renderer up. I use to render with 6 fast PC including may main work station, now I only use my workstation (with two gtx 1080 TI) for render my animations faster than my whole "renderfarm"…

Nicolas Jordan
11-25-2018, 10:50 AM
I think it would be nice to have the option to leverage the power of the GPU when rendering in Lightwave but I'm not going to hold me breath for that. I think as long as there is a GPU solution for Lightwave like Octane they won't bother with GPU rendering natively in Lightwave. I felt like I had to choose between GPU and CPU then building a new machine so I chose CPU. Maybe in the future if things go well I will add a 2nd 1070Ti and pick up Octane to get my feet wet with GPU rendering.

Tim Parsons
11-25-2018, 11:50 AM
NT should just use Radeon ProRender https://www.amd.com/en/technologies/radeon-prorender
Seems like the best way to go, it being free and all. Not so sure if Antti put in the hooks for his shaders or if it needs to use it's own, but this seems like a no-brainer for NT.

erikals
11-25-2018, 01:26 PM
i'd welcome ProRender for sure.

i'd also buy it if it was 3rd party.

CaptainMarlowe
11-25-2018, 01:35 PM
Yup, I would welcome ProRender too. Cross platform, supports openCL and/or Metal, which is a big plus for Mac users.

lwanmtr
11-25-2018, 01:40 PM
With Prorender you get into the argument between AMD and Nvidia...Does it work on Nvidia?

erikals
11-25-2018, 02:08 PM
Does it work on Nvidia?
Yes.

lwanmtr
11-25-2018, 02:41 PM
Thats cool then.

Marander
11-25-2018, 05:23 PM
I would have to get octane if I wanted to render teh 2018 volumetrics with decent speed, from native volumetrics to TurbulenceFD, the cpu rendering of TFD is just too slow for my liking, and a part of why I find myself utilizing GPU rendered fire and smoke in """ that is way faster.

I could of course go for Lightwave octane and Tfd, though that means apart from the intital lightwave cost ..additional cost for tfd, and octan..while It cost me nothing as it stands now.

I tried to render a VDB object in acceptable quality in LW 2018 today but after about 2 hours (on 12 cores) I gave up. I know I probably could have optimized the scene (I used a preset from the OD Quick Setup) but I don't have that time to fiddle with LWs render settings. I'll give it another try sometime later but I find little to no use for the LW2018 volumetrics or renderer in its current state. Octane seems the only feasible solution to me. Now that version 4 is out with all the plugins included it's an interesting option.

TFD renders pretty fast for me but not in LW (where it crashes anyway).

lwanmtr
11-25-2018, 05:44 PM
LW's volumetrics seem pretty quick to me..been going back n forth alot with real flow lately and musch faster than it was before.

Marander
11-25-2018, 05:46 PM
Thanks for the feedback. I guess I'll need to take some more time to find good settings then.

If there would be an example scene for rendering a VDB object in volumetrics in 2018 in good quality, that would be very helpful. Also great would be some nice nodal setup for good volume shading.

lwanmtr
11-25-2018, 05:50 PM
i havent tried an open-vdb yet cause I dont have any files...but the straight volumetric is pretty fast at least what Ive been doing.

jwiede
11-26-2018, 10:09 AM
i havent tried an open-vdb yet cause I dont have any files...but the straight volumetric is pretty fast at least what Ive been doing.

When you say "straight volumetric" do you just mean for env. fog or something else (and if so, what)?

prometheus
11-26-2018, 12:14 PM
I tried to render a VDB object in acceptable quality in LW 2018 today but after about 2 hours (on 12 cores) I gave up. I know I probably could have optimized the scene (I used a preset from the OD Quick Setup) but I don't have that time to fiddle with LWs render settings. I'll give it another try sometime later but I find little to no use for the LW2018 volumetrics or renderer in its current state. Octane seems the only feasible solution to me. Now that version 4 is out with all the plugins included it's an interesting option.

TFD renders pretty fast for me but not in LW (where it crashes anyway).

I suspect it depends on the VDB resolution.

Houdini has a VDB smooth node, which I believe is required to add before exporting the vdb.
Blender has none as I am aware of...meaning you need to have really high res simulation.
Within Lightwave there is only three options to deal with it in the interpolation mode.

Crap in ..is crap out.


We had a huge cloud sample in a thread here somewhere, from disney...which some of us did try.
So itīs not impossible to get good quality....may take a bit of work though.

prometheus
11-26-2018, 12:18 PM
i havent tried an open-vdb yet cause I dont have any files...but the straight volumetric is pretty fast at least what Ive been doing.

Depends on the step size you set, at good quality levels, I donīt think the new volumetrics is that fast...not if I compare to rendering volumetric PBR material withGPU in cycles for bl.

prometheus
11-26-2018, 01:26 PM
A quickie test..since the topic has evolved a bit in to openvdb questions...

Lightwave bullefx scene exported as mdd, fluids simulated in blender, exported back in to Lightwave for VDB and VPR render.

Its just a quickie test of running a bullet sim in lightwave from the 2015 content bullet/forces/ explosion scene.
Then sending to blender by mdd baking, then painting a weightmap around the main break area with blenders weight paint..and then assigning that map to serve as fluid density map..from where blenders fluid is generated, simulate it in blender at a resolution of 101 divisions, and smoke high resolution at 4 divisions.
Saving out the simulation as vdb uncompressed.

Back to lightwave and adding a null, set it to openVDB and load the vdb file, creating a manual gradient in emission and in scatter channel, render out with VPR.
I only have discovery edition so canīt get the best quality in there.
I think I also set the step size to high ..so the quality could be better.
Interpolation mode quadratic.


Arggh...vimeo embeds Still not able to work within this forum..how do we solve that? should perhaps have posted it on youtube..which I didnīt because it was only a small test I didnīt want to put up there.
A bit short in length..not the same shading match for volumetrics though, donīt have time for such tweaks for a test like this.

https://vimeo.com/302915512



https://vimeo.com/302915512

https://forums.newtek.com/attachment.php?attachmentid=143425&d=1543264883

143425

lwanmtr
11-26-2018, 01:53 PM
I've been attempting to get granular effect (importing realflow particles)..which hasnt been successful so much..but has given me the chance to fill my scenes with what does come out to be clouds and fog, which to me are rendering acceptably fast, even on my laptop. When I say straight volumetric, I mean setting the object type to volumetric, as opposed to mesh, etc.

I'll admit, Ive been tempted to look at blender's volumetrics, just to see how fast they go.

I can say that gpu render in Poser is much faster than it's cpu render, and in Daz the gpu render actually renders more realistic (daz's cpu render is really bad).

prometheus
11-26-2018, 02:47 PM
I've been attempting to get granular effect (importing realflow particles)..which hasnt been successful so much..but has given me the chance to fill my scenes with what does come out to be clouds and fog, which to me are rendering acceptably fast, even on my laptop. When I say straight volumetric, I mean setting the object type to volumetric, as opposed to mesh, etc.

I'll admit, Ive been tempted to look at blender's volumetrics, just to see how fast they go.

I can say that gpu render in Poser is much faster than it's cpu render, and in Daz the gpu render actually renders more realistic (daz's cpu render is really bad).


Daz GPU iray render is great in speed and quality, its spectral environment is one of the better ones, if you compare blender, lightwave, houdini and daz (excluding full volumetric atmosphere from terragen and vue)
But for Blender and Daz, it takes an initial sending time to the graphic card ..preparing the scene, in Blender it seems ok, in daz it can be a bit too slow for that initial process, once you have done that for the scene..it is a joy to work with.

If you try blender fire and smoke, and use the GPU...and have large voxel sets, I suggest that you go to render panel and change volume step rate to something around 1 or more...before activating cycles render preview, meaning it will render faster ..lesser quality initially...but take that step to ensure the GPU can handle it..otherwise Blender may return an error due to lack of memory or something, and you have to re-boot blender and the scene to get it working again, so start with higher step rates and see how much you can decrease the step rate level to finer quality..before it gives up on you.

As for vdb from blender, hard to say...there is undoubtly a showcase of blockiness in the VDB file, as mentioned..maybe blender needs a smoothing modifier when exporting ..or Lightwave needs a way to implement that better than the 3 interpolation modes we got today.

https://forums.newtek.com/attachment.php?attachmentid=143429&d=1543269206


Houdini I think does this better with a smoothing vdb node.143429

lwanmtr
11-26-2018, 02:50 PM
Indeed, Daz is very slow at that initial setup...and the larger the objects, textures, etc, the longer it takes. I 've noticed it will begin to really crawl after a while if youre moving the camer around alot or making many surface changes.

jwiede
11-26-2018, 05:54 PM
...or Lightwave needs a way to implement that better than the 3 interpolation modes we got today.

This is kind of the real problem, LW needs at least some minimal native ability to interactively clean/fix/smooth the VDB data prior to rendering, beyond a few interpolation modes.

Forcing the user to round-trip all the way back through the data export process again just to tweak/smooth their data a little is much too costly. A few round-trips like that and the user will start looking at rendering in the same app as where the data is generated.

Dan Ritchie
11-27-2018, 12:35 PM
Compared to 2015, VPR in 2018 wreaks havoc on my MSI Ghost laptop. The laptop begins to generate a ton of heat and the fan kicks into overdrive. It's another reason that I've been avoiding using 2018.

That's how you know you're getting proper utilization of your cpu.

Marander
11-30-2018, 01:54 AM
I suspect it depends on the VDB resolution.

Houdini has a VDB smooth node, which I believe is required to add before exporting the vdb.
Blender has none as I am aware of...meaning you need to have really high res simulation.
Within Lightwave there is only three options to deal with it in the interpolation mode.

Crap in ..is crap out.


We had a huge cloud sample in a thread here somewhere, from disney...which some of us did try.
So itīs not impossible to get good quality....may take a bit of work though.

The source is not the problem. Not talking of the Disney examples here (I created that thread) but VDBs I created in another app.

LW takes ages to render it in acceptable quality, but as I mentioned I also didn't take the time to find better render settings.

In Redshift the same thing renders in seconds with better quality right out of the box.

Additionally to that I have proper control in the viewport how it is displayed and I don't need to use the weird LW 'workflow' with a Null object but just create a Volume object with a Volume material and tweak it from there.

Also some types of VDBs cannot be loaded in LW (for example SDF - Signed Distance Fields but also some Fog volume objects I created).

It's difficult to find any use for LW these days.

prometheus
11-30-2018, 04:28 AM
The source is not the problem. Not talking of the Disney examples here (I created that thread) but VDBs I created in another app.

LW takes ages to render it in acceptable quality, but as I mentioned I also didn't take the time to find better render settings.

In Redshift the same thing renders in seconds with better quality right out of the box.

Additionally to that I have proper control in the viewport how it is displayed and I don't need to use the weird LW 'workflow' with a Null object but just create a Volume object with a Volume material and tweak it from there.

Also some types of VDBs cannot be loaded in LW (for example SDF - Signed Distance Fields but also some Fog volume objects I created).

It's difficult to find any use for LW these days.

The source is One part of the problem I would say, you can export out vdb...without smooth vdb node applied, and it wouldnīt be nice when rendering in Lightwave.
Now we are talking quality, not speed.
Speed is however another problem.

As for null items as volumetric, Lightwaves procedural volumes once applied through the more tedious interface nowadays, it is pretty nice...if you are to use any geometry in blender, make it a principled volume material...you would find that it is very hard to get anywhere near the
volumetric style Lightwave has for volumetrics, but then again..I am all for making any geometry volumetric..If it can keep itīs elegance on how the texture and density is applied on the object.