Hey all -

I'm working on a render right now, and I've spent a few days getting the right balance of quality/render time in the IP. Now I've sent it to hbatch for final render, and each frame is taking 4x as long!

Admittedly, the quality is better, but 2 hours/frame wont work for our deadline. What explains this disparity?

To clarify, I definitely do not have “preview” or “auto-update” checked in the IP. I assumed that what I saw in IP was identical to what I would get out of hbatch.

Thanks!
-Mike

tk421storm
Hey all -

Admittedly, the quality is better, but 2 hours/frame wont work for our deadline. What explains this disparity?

Thanks!
-Mike

What exactly did you crank up in quality ? If you can give more details it would really help.

Rob

The problem is, I haven't changed anything!

I render in houdini to the IP, and it takes about 20 minutes. I save and close the project, close houdini, then open the project with hbatch, tell it to render the exact same mantra node, and it's taking 4 times as long.

Hi,
You said in your prior post that the quality settings were better ?. Are you rendering a tiny res via IP and 4k or something in Hbatch ? . Are you rendering a sequence or a still ? , motion blur ?. Certainly some self help is required to debug what is going on but without a scene file and fine details it makes it very hard to debug whats going on.

rob

I'm rendering same res (720p) in both contexts - with the same settings. In the IP, I'm rendering one frame at a time, and of course with hbatch I'm rendering a whole sequence.

There is motion blur - would the calculation of the whole sequence be different than a calculation of a single frame?

When I say the quality was better - it's just that. I haven't changed any settings, but the hbatch render is rendering with no noise at all - it seems like the samples are cranked way up.

I have object level ray min/max settings on some objects, is it possible those settings are being ignored by the IP but used by the hbatch mantra?

Sorry to be confusing, I have no idea what's going on here. Same file, same machine, same output, but wildly different render times.

-Mike

tk421storm
I'm rendering same res (720p) in both contexts - with the same settings. In the IP, I'm rendering one frame at a time, and of course with hbatch I'm rendering a whole sequence.

There is motion blur - would the calculation of the whole sequence be different than a calculation of a single frame?

When I say the quality was better - it's just that. I haven't changed any settings, but the hbatch render is rendering with no noise at all - it seems like the samples are cranked way up.

I have object level ray min/max settings on some objects, is it possible those settings are being ignored by the IP but used by the hbatch mantra?

Sorry to be confusing, I have no idea what's going on here. Same file, same machine, same output, but wildly different render times.

-Mike

By IP are you referring to the IPR render in the view port ? if so that uses the raytracing engine to render. Whats your mantra rop set to ? . Ref Motion blur of course that can be an issue, if you have something fly past close to the camera focal plane , motion blur calculation will increase. Ray min / max values I tend to set in the mantra rop and not on per object basis.

Rob

I guess I mean the “Render View” - not the Interactive one available in the 3d viewport. I use the term “IP” because that's what auto-populates the destination field of a render node, and I assumed that meant send it to the image viewer.

I'm using PBR, and I'm assuming that what I see in the render view would be the same.

Even with motion blur, I'm assuming that what I see in “Render view” is what I'll get when I do hbatch. If something takes forever to render w/ motion blur, I should see that in the Render View, correct?

So now I've picked up the render again, while I'm sitting at the machine, and it seems to be taking the expected amount of time per frame. I wonder if something is happening when I'm away from the machine at night, slowing down the process.

Thanks for talking me through this! I love houdini, but often it'll do something I can't understand, and with a tight timeline I was extra worried.

-Mike

Try comparing the render view render with a render from the mantra ROP in houdini. Also check if your camera is the same as the one used in the render view (I'm assuming the ROP is the same).

Spoke too soon!

I don't know what's going on here. I've attached two screengrabs - both from exrs read into nuke - one from the render viewer, and one from the mantra hbatch render.

Not only does the mantra render take 4x as long (2hours vs 30 minutes) but the lighting looks totally different. It's something I could fix with CC, but why is it taking so much longer?

I must be missing something about how the render viewer works. Am I wrong in assuming the result I see in the render viewer should be identical to the result from a command line renderer?

tk421storm
Spoke too soon!

I don't know what's going on here. I've attached two screengrabs - both from exrs read into nuke - one from the render viewer, and one from the mantra hbatch render.

Not only does the mantra render take 4x as long (2hours vs 30 minutes) but the lighting looks totally different. It's something I could fix with CC, but why is it taking so much longer?

I must be missing something about how the render viewer works. Am I wrong in assuming the result I see in the render viewer should be identical to the result from a command line renderer?

I have already told you the IPR uses ray tracing so clearly your render engine on your mantra rop is not set correctly. I can see from your render you have just about every bell and whistle turned on !

Rob

I'm sorry if I'm seeming dense, I'm just trying to understand this.

In my render ROP, i have an option to choose “Render Engine”. You mean that no matter what I choose there, if I render this ROP in the “Render View” pane, it will always render with the ray-tracing engine?

I just threw together a simple project with a torus and a light, and when I changed engines on the ROP I did see changes in the “Render View”.

AND if this is true, how can I make my mantra ROP render the same thing I'm seeing in the render view? Just change the engine to “Ray Tracing”?

tk421storm
I just threw together a simple project with a torus and a light, and when I changed engines on the ROP I did see changes in the “Render View”.

AND if this is true, how can I make my mantra ROP render the same thing I'm seeing in the render view? Just change the engine to “Ray Tracing”?

Correct.
I will freely admit I never use the IPR to render anything, I always launch a render through the link to the output driver menu. ( the roll of film logo in the viewport. I find the IPR does not even update correctly call me old school but I keep it simple and find using Mplay far better to look at my renders as they come through.

rob

Hey Rob -

Thanks SO MUCH for sticking with me!!

Wow, that is a crazy limitation. I'm shocked that I can't find anything about that in the documentation either. All the tutorials (even from sideFX) have taught me to use the RenderView, and never once mentioned that huge limitation. Houdini is usually what-you-see-is-what-you-get, so to have the undocumented is a huge shortcoming.

For clarification, in my renderROP I should set the output to be “ip” and then click the film roll on the scene side? Then, when I'm ready to render the sequence, I can set a filename instead of “ip” in the output, correct?

Thanks again, man.

No problem

Limitation, not really if you don't use a feature its no limitation. First thing you have to learn is nothing is ever perfect in software or pipelines. Dont worry about it, move on. For starters if I am lighting a shot I use super low settings and I layer up my shot with lights and features. I certainly dont turn on PBR , add a key and crank all the settings on and sit waiting for a render , then to make changes.

Set your mantra rop to ip in the output parameter > use the the film rol fly out menu to select the rop > mantra will kick off a render to mplay , you make a scene change then MMB on the film roll logo and your last mantra rop selection will get rendered again . Just leave mplay open and pin open a render scheduler window .

Rob

The IPR will only switch micropolygon renders to ray tracing if “preview” is on. Preview is impossible with micropolygon renders which is why this change is applied automatically. You can see the rendering engine used to produce the render in the top left corner of the render view or in mplay (eg. “/out/mantra1(raytracing)”) to make sure it matches what you expect. If preview is off the render should be the same as the ROP render, as long as the same camera is selected.

This doesn't explain the different render times you were seeing between the render view (with preview off) and the ROP, though.

Andrew

Well, now I'm confused.

It does look like the mPlay window is just a popped-out version of the “Render View” window. And I do see the render type in the upper left, which is why I assumed that was always the way it's going.

I'm going to take off all my object-level sampling settings, and see if that makes a difference. Otherwise, I'm not sure where to go. If I can't get out of hbtach what I can see in mplay or render view, I'm not sure how to work with renders at all.

-Mike

tk421storm
Hey Rob -
I'm shocked that I can't find anything about that in the documentation either.

if you need more reference:

http://www.sidefx.com/docs/houdini12.5/ref/views/ipr [sidefx.com]
'When this option is enabled, the view will use raytracing for primary samples even if micropolygon rendering is enabled in the Mantra Output Driver'

Edit: Though, it looks like PBR can also render, not just the micropolygon.

Yeah, I'm 100% sure I haven't been using it with the “preview” button checked. I watch the buckets fill from the center like normal, for the final render.

I'm rendering to the mplay window, and I'm seeing times and results much more like what I saw through the hbatch render. I'm having trouble getting that original “Render View” result, even with sampling settings deleted on the object level.

I'm not sure the exact differences you see are but could the gamma color correction be the major issue? Right-click on render view/ color correction, and, change the gamma to match MPlay