Everything related to Maxwell network rendering systems.
By DavidLemelin
#395160
I have been encountering a problem using the denoiser on a network render. The denoiser itself works fine if I render on a single machine, but with multiple nodes it appears to crash.

I've been directed to the documentation for help, but there is no mention of this issue so I'm still looking for a solution. Are there any settings to account for other than enabling the denoiser in the render set up?

Has anyone else experienced a similar issue?
By bitnaut
#395580
Using the latest v4.1.1.1, the denoiser on my OS X network/render farm doesn't crash but it is unable to write the denoised result to the specified output path. Only the render node on the machine running Maxwell Network Manager is able to write it's denoised result. This tells me that the trickery that they do for the other scene/output paths isn't being done for the denoiser.
By DavidLemelin
#396843
Sadly, I have no resolution yet for what on the surface should be a simple fix... but what do I know?

What I can report is that if your computer is powerful enough, the denoiser can actually reign in production time enough to make it feseable to render work directly and quickly (relatively speaking).

It’s certainly no selling point as there are many other faster options out there, but if you’ve invested in Maxwell and have a string of archived projects full of MXM’s, it’s been my experience that rather than build a network for Maxwell the better solution is a more powerful workstation that can drag Maxwell through brute strength.

I had a workstation built around AMD’s Threadripper processor and can report that with the denoiser, I can easily render out a dozen 3500px x 2400px images in an hour or less. I’ve even been able to walk away from the render farms I’ve had to resort to. The number of projects I’ve been able to “not”send to a render farm is likely to offset the investment in the Processor.
#397144
Denoiser in network didn't work for me as well until I changed some settings.
Got some of the info from support.

Settings for cooperative job

First Denoise Tag has to be in %scenename% instead of just 'denoised' otherwise it won't merge or write the denoised output and overwrite files in animation jobs. (changed this in my plugin)
From my experience only short filenames for mxi and image output with underscores work. (changed the filenames in render options pop up)
Also I selected all output channels for denoising manually (shadow, normals_camera, position_world and reflectance) as EXR , send dependencies checked, denoiser checked.

settings worked for CPU denoising , image resolution 3200x1800 and SL 12
had some issues with higher SLs and resolutions on a project two days ago because GPU ran out of memory so I'm using CPU denoising on big images
hope I could help
#397145
Hello,

I'll try to help you as much as I can with the problems with Denoiser + Network.

There are some things you have to take into account until we fix or remove these hurdles:

-Thiemo is right about naming the denoised image as %scenename%_denoised.png (for example) when launching an animation job; otherwise, one frame will overwrite the previous one.
-When you launch a single job, the denoised image is calculated by the node computer and what is more important, the resulting image is stored by it (instead of sending it to the manager to save it), so it should be able to understand the output path. In the case of Macs, this is sometimes hard to achieve.
-On the other hand, when you launch a cooperative job with denoiser, the nodes do not compute the denoised image; they just launch a normal render (with the required passes) and send the mxi files to the manager; this merges the different mxi files down to two and generates the denoised image from those; then it saves the denoised image in the specified path. So in this case, the output path has to be accessible from the manager computer.
-Additionally, there's a bug when using three nodes in a cooperative job with denoiser; one of the renders gets black and the denoised image gets blurred; this doesn't happen with two, four, or any other amount of nodes.
-The amount of memory is important, so if the GPU cannot host the whole scene, it is recommended to use the CPU to calculate the denoised image.

I hope this information helps.

Best wishes,
Fernando
Will there be a Maxwell Render 6 ?

Let's be realistic. What's left of NL is only milk[…]