Render Farm

[skogskalle]

Rebus costs alot more when you use the distributed rendering option... if you dont have alot of time this might be the only option, but if youre rendering 5 stills and need them finished over night, you can let Rebus use only one computer/image. This ofcourse takes longer but the price is really low.

The upload time can be a pain if your scene has alot of proxys and stuff... the good thing is that the textures/proxies etc stay uploaded, so if you change your scene a little and want to render it on rebus again, it doesnt have to upload everything again, only the new max-file.

[voltron7]

Rebus costs alot more when you use the distributed rendering option... if you dont have alot of time this might be the only option, but if youre rendering 5 stills and need them finished over night, you can let Rebus use only one computer/image. This ofcourse takes longer but the price is really low.

The upload time can be a pain if your scene has alot of proxys and stuff... the good thing is that the textures/proxies etc stay uploaded, so if you change your scene a little and want to render it on rebus again, it doesnt have to upload everything again, only the new max-file.

Thanks for the info skogskalle...
That must be it, I chose distributed as I only needed one large still. I was actually considereing trying it again but the idea of re-uplaoding that scene made me cringe. How do you resubmit it without having to re-upload? Does the manager just know better?

Thanks again

[voltron7]

Well, I am trying again same scene and no Distributed. Funny thing...it took rebus manager almost over an hour to re-upload the scene, even though it was reporting skipping all of the maps/assets, and I even got rid of 2 maps.
Oh well, still better than the first time. Will cross fingers, but 25 minutes after hitting "start" and it still seems stuck at "rendering: 0%" Maybe it will pick up speed soon...

[Dariusz Makowski (Dadal)]

More of a reason to use EC2. With that solution u can just dump all ur assets to FTP and let it quickly resync ur scenes from FTP to amazon servers. This way you should be able to save a lot of time. Good think to do is also to use backburner and save jobs as full scenes with maps so you just copy zip to FTP which then Sync to Amazon servers for render. It should take 5 min to sync 1gb scene and you can get ur renders managed a lot better. Since you know exactly why your render hangs and dont render Using Backburner Distributed tile rendering you can also add extra nodes as you need them. U can also use deadline

Thanks, bye.

[voltron7]

@DADAL, Still definitely going to give EC2 a shot next...but all this testing out has really eaten up waaay to much time that could have just let the thing keep going for xx hours, lol.
I do use Backburner, but on this scene, it seems to get to a point and then it decides to restart the task! So it's looking good, and the next day I check and it has started all over again and now back to waiting like new.
I also tried to save as bat file (in BB) and tried to render from the bat, but nothing happens. Is this how you upload as a full scene?
Thanks for your replies.

Rebus informed me that on single frames, the progress won't show more than zero, until it is complete!

[Dariusz Makowski (Dadal)]

Heya

not sure what you mean by "but all this testing out has really eaten up waaay to much time that could have just let the thing keep going for xx hours, lol."

With back burner when you use max to export scene there is a ticker to save maps. One you do it in program files backburner or documents somewhere he creates a RAR file with your max scene and textures. I then take that to EC2 BB server and start it there. BB pick it up and render it to my local HDD... Now that was quite long time ago that I used this workflow. Nowdays I just mirror my project HDD to Amazon EC2. So basically

Locally I have D:/Projects/TestMaxAmazon/
what I do is I create folder on my FTP called Projects/TestMaxAmazon.

I put all the files to that FTP location and when I start amazon instances they all automatically download that folder to their own D:/ drive, this way they have mirror of my specific project.

Once that is done I just save files to my local X drive(cant be D as it would save on instances own D driver) using Deadline and all rest is automatic.

I send jobs from my home deadline, I let them sync and all get saved localy.

Piece of cake more or less once you set it up

[voltron7]

Heya

not sure what you mean by "but all this testing out has really eaten up waaay to much time that could have just let the thing keep going for xx hours, lol."

I mean I spent 30 minutes getting the scene ready for rebus, 8 hours uploading to rebus, and several more hours to realize it was going to cost too much for this particular job...now I know what to expect and can plan better if I do another one there...



Well thanks a lot for trying to help explain the process.

[Morne]

I was hoping I'm finished with java< but you need it for farmanizer from rebus

[skogskalle]

voltron - yup the manager dont show the progress when rendering stills. You can sort of get around it by clicking "frame info", then "refresh frame info" and the click "more". It will then download a logfile (like the vray message box when rendering locally) thats shows how far the render has come. It works fine, but they should really fix it so the progress shows directly in the manager... would save some time and annoyance..

[voltron7]

voltron - yup the manager dont show the progress when rendering stills. You can sort of get around it by clicking "frame info", then "refresh frame info" and the click "more". It will then download a logfile (like the vray message box when rendering locally) thats shows how far the render has come. It works fine, but they should really fix it so the progress shows directly in the manager... would save some time and annoyance..

Yes, you are right, IMO...Still, in fairness to them, I really had not optimized my scene enough...I even found an object without a material!
Thanks!

[QuakeMarine1]

Rebus here

nice threads but many wrong info
<Well with EC you pay around 0.48$ per 1 hour of 16 core rig... so if rebus farm takes 1h on 1 computer and cost u 30$ amount then amazon 1 pc will cost u 0.5$ <for 1h more or less...

the image was render using 100! pc and all in all the process tooks 1 hour
30$ for 100x12core nodes for 1 hour
can EC beat that ?
keep in minde you pay at Rebus only the CPU usage in seconds!
at EC you pay full hours and all pc!

a simple example about DR
for distributed render i use 100 pc - it means the image you send need to be load , prepared and started 100 times
if each pc need about 1 minute to do so it mean 100 minutes total plus the render time
its logic not to use DR in case your local PC render the image in 30min
DR is 130min to pay
no DR is 31min to pay
insteat if you render 20 hour on your local PC the "overhead" is not noticeable
DR is 20hour and 100min to pay
no DR is 20 hour and 1 min to pay

I hope everyone understand what i mean ?!....

the % progress is update the soon a pc finished the assigned work
for DR the progress is pretty accurate
no DR jobs stay 0% until all the image is done - we are aware about and try to find a workaround somewhen


just ask me all you like to know about Rebus

Ceo Andre

[Dariusz Makowski (Dadal)]

Heya

QuakeMarine1 - I sort off got lost... but its interesting to see that 100x12 core rig can cost 30$ for 1 hour... it sort off beats EC2 to be honest. I think I might do some comparison tests once I have some free time... end of july or so...

Thanks, bye.