Distributed Bucket Rendering

[mskin]

To distribute Bucket Render, i only need 3d max installed on another machine. without invoking backburner, i can simply select "distribute bucket render" and identify the other machine and it will - if on - participate in the glory?

 

i am limited to 10 cpus (without purchasing additional liscenses) - what does that mean? 10 machines, or or 10 cores?

 

lastly, say i want to up the number of cores available to me - i need another liscense, but of what? 3d max or mental ray.

 

thanks all

[Matt McDonald]

The Mental Ray Satelite service needs to be running on the remote machine (with admin rights if it will need to access a network file server for maps and the like).

 

Regarding the number of machines. Through your license of Max you can have ten installs for remote renderings. However Mental Ray distributed rendering will only connect to six additional machines (might be 5). Its also worth pointing out that with distributed rendering you are going to see diminishing returns. The sixth machine is not going to make nearly the difference that the first machine did.

[mskin]

Thanks Matt. I have max installed on about 7 machines - for the sole purpose of distributed render. It seems to be working nicely, the FG still takes just as long, but its pretty great when a slew of buckets start working for me and people over in the interiors department start complaining about computer performance.

 

What is the best method to take advantage of 7 or so quad core machines? Dist. Bucket Rendering or Backburner? or are there other alternatives?

[Matt McDonald]

I think it kind of depends. If you are working on a still image then distributed is going to be your best bet but if you need multiple images you'll probably want to use backburner. Distributed rendering can be a little buggy so I wouldn't recommend it for animations.

 

The only other option that I can think of would be using backburner to split the render into strips (assigning each strip to a different machine). I've only vaguely heard about people doing this and don't have any idea how well it works.

[Justin Hunt]

Split scan line works quite well, as long as you dont need an alpha channel. For some reason it isn't saved.

 

If your using max2011 the administator right are very important so safes be is to run as administrator.

Also be sure that every machine has the same gamma settings and all can see the maps. Large files/models can take a while to load can slow things down when rendering starts.

 

Just be sure no-one switches off a slave machine during rendering as this will crash Max.

 

jhv

[mskin]

i tried that strip rendering early on and got some weird results. every strip was slightly off in tone from the others leaving a striped appearance. upon further reading it looks like saving a FG an GI maps to a networked share is a possible solution?

[Matt McDonald]

If splitting the render into strips works well you could assign 9 machines to rendering your beauty pass and the tenth could be assigned a job to render your alpha.

 

At work, I have crapo laptop and many times I've stuck around an hour or so late and used distributed rendering to tweak materials. If you cancel the rendering, distributed crashes max but with a bunch if additional cores things finish up naturally pretty quickly.

 

As Justin points out, network performance becomes a bigger deal with backburner or distributed rendering.

[mdviz]

Make sure you lock the fgm map file or it will be updated by each machine and overwritten. Hence the different look to each stripe. I would render the fg seperately on one PC (with DBR on) and then put a cloned version onto each machine to save the constant network hit you get from what can be a large file (often over 150Mbs) and can cause crashes if a network bottleneck occurs. Once again, ensure you lock the FG before submitting to render. This is what I do for animations when the render needs a 'multi frame' FGmap file. It might sound complicated but renders faster.

 

You can also submit to network render and let one node pick the job up and use DBR, utilising the rest of the PCs however if there are other renders in the queue the manager will not know that DBR is in use (it is not reported as it is a seperate app) and try to render the next queued job at the same time. Not desirable at all so be careful when trying this.

 

Having said the above. It is a great way to off load all the rendering to a manager pc that then uses other PCs (selected in the hosts file before submitting and unassign any jobs from those nodes in the manager) allowing you to continue working.

 

If you are going to render a single (large) render using DBR then I would not use machines that are being used by others unless you know how to, and can set the affinity for those particular PCs. In my experience fiddling with the affinity can cause random crashes anyway so I would try and stay away from doing this.

[Camby1298]

Has anyone come across the error when using DBR and BSP2, consitantly when I would render an image using DBR and had BSP2 set, it would render out the whole image, and then crash before saving. DBR seemed to work fine with BSP, but when a scene has thousands of proxies, ya gotta use BSP2. Has this been resolved in Max 2011????

[exxoo]

I've had some problems with rendering hitting time limit (600min) and restarts, i think its a bug because it happens even if I change the max time..

I usually do the rendering with stripes then.. I trust backburner more than DBR, if a frame crashes backburner will render it again, with DBR you dont have anything when you get to the office the next morning

Never tried render with backburner submitting a DBR job like mdviz said, all try that next time But there is a reason I save before ever rendering using DBR

[TSuess]

The max number of cores I have been able to get working is 24. Using 3 dual quad core machines. I also use backburner in order to keep my primary workstation free. IMO that's the way to go, well until GPU rendering is the mainstream.