AMD's (ProRender) Plugin for Blender Previewed


(ajm) #261

Microsoft pushed an update recently and wiped out the drivers, and I lost the OpenCL.dll, so the most recent driver I downloaded was 388.59. (really disliking the Windows 10 update process). Anyway, been rendering with that lately, but the bug persists. I have been able to render with RPR, just not use any denoising.


#262

bad win10, bad… :no:
i’m using 376.84 drivers

Further development in my case:
Looks like IBL was too large or something - RAM consumption fluctuated over the top & down to ‘zero’, before freezing. IBL map removed. But still won’t render - freezes as soon as sampling starts (@ 2nd-4th iteration). Modifiers applied, geo clean, scale 1, normals ok, shaders basic/simple… Maybe my mind melted :spin: … can’t find the reason, so if anyone is willing to look into, here’s the scene file: Entrance-bughunt.zip (6.32 MB)

PS
Works fine in preview


(juang3d) #263

Is there any roadmap for future evolutions of RPR?

Cheers.


(thomascheng) #264

I tried it with a 16k HDRI from HDRIHaven. Preview works, but when I tried to render it, seems to get stuck at the initialzing phase. I would submit the file with a bug report.


(bsavery) #265

If you check back I posted a link to our discord server, come chat with us there, many updates there on things coming…


#266

Thanks for the feedback.

Old bugger still lingers: poles & n-gons don’t work nice (shading issues, black tris) - most likely NVidia only… :confused:? Bug report i must…? Is it ray epsilon (aka Secondary Rays Bias) problem or…?

On fresh, squeaky clean scenes this engine flies otherwise…

Discord? I really don’t wanna register anymore just to read :rolleyes:


(jovlem) #267

Could someone make some renders / screenshots from:

  • ProRender with viewport denoiser enabled
  • ProRender with viewport denoiser disabled
  • Cycles in viewport

I like to know what are the differences in time and samples to get an acceptabel image for a preview.
I would suggest to use some glossy and glass materials because these take longer to render.

My computer don’t supports OpenCL 1.2 so I can’t test it myself.


(juang3d) #268

Interesting, I´ll do it soon, I´m not used to Discord, I find it a bit uncomfortable, but I´ll try to get used to it since many people is using it :slight_smile:

Cheers!


#269

Here’s something fast from tests/studies am performing… 128 passes on RPR VS 128 samples on Cycles Same clamping value (20) & the number of bounces/light paths (12) - times are near but clarity, obviously not… :eyebrowlift2:

(unequal colors are my fault - made mistake while transfering hex values with different tone mapping in place :rolleyes:, also i still need to figure out some details about Cycles vs. RPR relation: samples VS passes, light/emission values…)


(bsavery) #270

Hi Burnin. The water looks much darker in RPR than cycles, so not sure about that. But samples vs passes should be the same as long as your’e not using square samples in cycles. I kind of want to rename the interations parameter to samples.


(tobbew) #271

It looks like caustics were on in cycles but not rpr


(mathieu) #272

@burnin Can you provide the blend?


(Roken) #273

Another Arch user here wondering when builds for something that isn’t 'buntu will be arriving?


#274

Yes indeed (as mentioned the difference in color, even for water transparency, is my fault). Rendered are two different files using different shading. Wasn’t quite ‘academic/scientific’ with my goals when i set those in October (and haven’t touched since).
With first RPR tries i didn’t put any attention to shaders until recently :wink: fast compile & fast render.
Then again, there’s no Principled neither Glass shader in RPR & I know i must set scenes as close as possible (technically the same) to make valuable ‘comparative study’… :o

Here it is… cornell_RPR.zip (7.77 MB)


(chafouin) #275

Thanks for the file burnin. Gave your scene a try and I have issues with CPU rendering and denoising, it renders transparent with Edge Avoiding Wavelets or Bilateral, or very slow (hangs every 4 samples) and black with Local Weighted Regression?
No problem with GPU only.


#276

Yes, same here (sorry, i don’t repeat bugs found :wink: ) i assume CPU is in experimental phase as the main point & priority are AMD GPUs, APUs, OCL… “Boltzmann Initiative”


(mathieu) #277

Thanks for the file. But the Cycles materials are missing. I did a quick test and my Cycles render were much cleaner than yours. Could you provide a file with 2 scenes (one for cycles and one for RPR)?


#278

Yes, as mentioned :smiley: two different scene files were made… here’s the Cycles version: cornell_Cycles.zip (3.03 MB)
Too much difference/inconsistency between engines (definitions, shading, lighting…) & not enough documentation make me wonder more than i am willing to accept for now (for intricate study).
ie. Lux Render engine documentation is really an admirable example. The workings & engines there are very well explained/exposed.

So please, share any findings and your thoughts on workings, models, values, their relations & such… between RPR & Cycles.

TIA & i wish you joyful holidays!


(mathieu) #279

Deactivate Clamping in rpr and you will get much more noise. Nonetheless, it made me try it again and it has improved a lot stability and performance-wise. I’m really impressed.


#280

Yup, i use clamping (great tool) in all engines that allow me to… usually from 0.1 -10000 - scene dependent

yeah, when it runs it RUNS! :yes: