Hello everyone! Today I want to write about sensitive topic. Since I am not a native English speaker, I apologize for the errors in writing. I am a user of the program Houdini for more than two years. I've always wondered possibilities this program. But there were also frustrations. Yes, I know that talking about this subject here. Peoples asked questions, “why the mantra is rendering slow?”. Micropolygons engine works fast in almost any difficult tasks. And it is true! Raytrace engine had worked slowly and now also too. This applies to everything: objects, volumetric effects, displacement, area shadows, indirect light.

Do not misunderstand me. I do not want to say that everything is very bad. Mantra features a very, very good results! But today, other engines show a good time rendering. This is Arnold (today in the subject of discussion). I spent many hours in tests of two remarkable rendering engines (PBR and Arnold): balls and planes, refraction, reflection, volumetric effects, as well as in the interior with textures. I put the settings so that the final image is of the same level of noise. In order to judge the time of calculation.
I did not set high values of the parameters, also using Area Light with Samples Shadow is 1 (usually 1 or 3 light source). I also watched a training video: https://www.sidefx.com/index.php?option=com_content&task=view&id=3048&Itemid=406 [sidefx.com]

The last image I made with the following parameters:
Pixel Samples - 4x4
Indirect Noise Level - 0.01
Max Indirect Ray Samples - 20
Diffuse Limit - 1
Arealight Sampling - 5

Arnold Render:
Camera - 4
Diffuse - 4
Arealight Sampling - 5

The results are impressive. The image produced in the PBR will get a lovely, materials, geometry, textures look distinctly bright and juicy. In Arnold all lower, but the final result is very good and not bad. All this is very well illustrated in the time that has been spent to perform the render. For me, completely unacceptable 48 minutes total time at almost identical results Arnold reveals astonishing 19 minutes of time! Each new version of Houdini, I look forward to. I always do tests PBR, but instead get the frustrations instead of a smile. Therefore, the fast and good result I can get only with Micropolygons or not best Raytrace with simulated global environment via occlusion and fill lights.

The second issue that I would like to say. Why is the new version of Houdini 14 network view window was to slow down? And I mean that I have a huge number of nodes inside. When I go to the shop network (or out, it does not matter), and then returning to / obj network I have to wait long for a long time. This is had not happened in Houdini 13.

PC: Intel Core i7-3930K 3,8Ghz, 64bit Windows 7, Nvidia Geforce GTS 250

Thanks for the tests. The 2 issues raised are how is Arnold quicker than Mantra, and, the network view is slower in H14 than H13.

The Network view may require a better graphics card. I couldn't see the amount of ram the GeForce GTS 250 has, but it may be not enough to draw the viewport and interface efficently in H14.

Mantra vs Arnold usually comes down to Mantra needing a little more ram to render a scene - tests quite often show the speed to be equal.

MartybNz
…
Mantra vs Arnold usually comes down to Mantra needing a little more ram to render a scene - tests quite often show the speed to be equal.

I can second what Marty says, we're moving from Softimage/Arnold to Houdini and have seen about the same render times Mantra vs. Arnold for our projects for the ones I've taken the time to run a comparison of the two render engines.

edit: small tip, try to keep your light samples low (starting at 1) and let the min/max for direct rays do their magic. It should save you quite a bit of time.

after almost 4-6 months of testing different HEAVY HEAVY scenes. we found out that mantra was faster (from 25% to 50% depending the scene.) and depending the artist.

memory management mantra was WAY better, Arnold crashed running out of memory several time. specially handling textures with displacement and because of that we move the hole production to mantra.
is worth to mention the amount of money saved in licenses.

MartybNz, Andy58, pelos, Guys thank you for writing a few words.

MartybNz
The Network view may require a better graphics card. I couldn't see the amount of ram the GeForce GTS 250 has, but it may be not enough to draw the viewport and interface efficently in H14.

I asked a guy who owns a graphics card NVidia Quatro, and he confirmed that the brakes or bugs he does not have. This is reverse of the medal new version. Network view currently requires large computational cost. As far as necessary? Now Netwok View is faster with a lot of nodes, but it requires greater computing resources when it does processing for display.

Andy58
edit: small tip, try to keep your light samples low (starting at 1) and let the min/max for direct rays do their magic. It should save you quite a bit of time.

Thanks. Good tip.

pelos
after almost 4-6 months of testing different HEAVY HEAVY scenes. we found out that mantra was faster (from 25% to 50% depending the scene.) and depending the artist.

memory management mantra was WAY better, Arnold crashed running out of memory several time. specially handling textures with displacement and because of that we move the hole production to mantra.
is worth to mention the amount of money saved in licenses.

Good, but why you have chosen in the direction of the mantra, and did not chosen, for example RenderMan? It is also worse in speed when a very heavy scene?

From what has been said. In my dreams to see PBR to all users. For those who need to do to render simple scenes with a stunning time of rendering the result, and as well as for those who work with huge amounts of data. I think Mantra render engine has good prospects to be the best production renderer in the world, to be a leader in the processing speed, both complex and simple scenes. But now, small studios that make simple TV commercials or very simple cartoons, do not have good rendering farms, they will choose another renderer or engine in Houdini.

del

Fancy posting up some of test file scenes

Rob

Why are there artifacts on the penumbra regions - best seen on the round couch example - in the Mantra renderings?

McNistor
Why are there artifacts on the penumbra regions - best seen on the round couch example - in the Mantra renderings?

It's not enough smoothing (option “Render Polygons As Subdivision Mantra”). It's funny, but in rendering Arnold, I did not turned on anything for smoothing.

Felix6699
It's not enough smoothing (option “Render Polygons As Subdivision Mantra”). It's funny, but in rendering Arnold, I did not turned on anything for smoothing.

Have you checked the normals on that model, perhaps put down a normal SOP again at the end?

Andy58

Felix6699
It's not enough smoothing (option “Render Polygons As Subdivision Mantra”). It's funny, but in rendering Arnold, I did not turned on anything for smoothing.

Have you checked the normals on that model, perhaps put down a normal SOP again at the end?

Nope 8)

Felix6699

Andy58

Felix6699
It's not enough smoothing (option “Render Polygons As Subdivision Mantra”). It's funny, but in rendering Arnold, I did not turned on anything for smoothing.

Have you checked the normals on that model, perhaps put down a normal SOP again at the end?

Nope 8)

Looks like a raytrace bias issue - have you tried raising the bias in your mantra rop?

Mantra -> Rendering tab -> Shading rab “Raytracing bias”.

Hampus Hedberg2

Felix6699

Andy58

Felix6699
It's not enough smoothing (option “Render Polygons As Subdivision Mantra”). It's funny, but in rendering Arnold, I did not turned on anything for smoothing.

Have you checked the normals on that model, perhaps put down a normal SOP again at the end?

Nope 8)

Looks like a raytrace bias issue - have you tried raising the bias in your mantra rop?

Mantra -> Rendering tab -> Shading rab “Raytracing bias”.

:idea: You have solved this problem! “Ray Tracing bias” parameter eliminates artifacts as “stair” on geometry.

What is the best practice to avoid this scaling issue?
Consider one H unit as… 1 cm, 1 m, 1 mm?

McNistor
What is the best practice to avoid this scaling issue?
Consider one H unit as… 1 cm, 1 m, 1 mm?

Houdini units are by default set to meters. its best to leave it at that. As a rule of thumb, the raytracing bias should be 0.01 of the size of your scene.

rob

or use “Bias along normal”
worked for me, when i was doing droplets on bottle.

pelos
memory management mantra was WAY better, Arnold crashed running out of memory several time. specially handling textures with displacement and because of that we move the hole production to mantra.
is worth to mention the amount of money saved in licenses.

That's not what we usually hear, but that's entirely possible. The first things to check would be if the textures have been converted to a tiled, mip-mapped format (using maketx for example) and if the subdivision iterations are not crazy high.

I'd love to have repro scenes or even just detailed logs of some of your problematic renders. Which company was this in? I didn't hear from this from support but maybe I missed it. Please PM me with the info if you are keen.

unfortunately memory management is not Arnold fault, but its bad behavior of Houdini it self. Houdini is very bad with memory management while rendering.

I have encountered a case where the SOP cache kept growing frame after frame for heavy geometry translation (ticket #375), the workaround was to call sopcache -c in the Post-Frame script to flush the cache.

Actually… that ought to be an option in the Arnold ROP, created #667 for that