V9 Renderama question

[supernovavfx]

first comments on my situation

#1- 95% of my work is rendering 30 - 900 frames 1280 x 720 on average resolution. I use an 8 core MacPro

#2 - I have tested a number of my v8 scenes with v9 I have seen a nice performance increase initially. All positive.

#3 - 90% of my FRAME renderings average 3 minutes to 15 minutes PER frame.. (depending on project obviously)

#4 - I have seen speed increases but NONE of them come close to 7 or 8 times faster than the V8 Camera.

QUESTION

Can I still set up renderama with 8 slaves just like I used too and have them each operate on a core just like in V8?

it still seems to be be my fastest option for rendering my animations. I have read the manual and it explains adding multiple MACHINES

to renderama...not multiple slaves per core on ONE machine... help please.

Scott

[Tomas Egger]

Ola Scott,

Camera 9 using all 8 cores will not make Camera 9 run 8 times faster, even Camera 8 with 8 slaves cant be 8 faster..

Answering: You need only to make your 8 slave folders like before, go to Camera 9 settings and type 1 thread only, so, each Camera will use only 1 core, you will have 8 Camera 9 using 1 core each.

Hope that helps.

Thanks

Tom

[supernovavfx]

thank you Tomas... i KNOW that i dont get 8 x the speed :) but I DO get a much better overall render time total..... I see.... I just set the Camera to one core....

its still REALLY GREAT to have my 'working' test renders FLY so much faster which leads to faster overall job performance.... I wonder if setting 2 slaves to 4 cores

or 4 slaves to 2 cores would make any difference.... the whole app just feels and reacts so much faster on all levels!! great job again... :)

[supernovavfx]

last question on this.... when Camera V9 uses just one core (such as renderama using 8 cameras on a machine with 8 cores) is Camera 9 faster than Camera V8 is?

[Tomas Egger]

Ola Scott,

Camera v9 using one core is faster than Camera v8.

Thanks

Tom

[David Sander]

I have a large scene (over 6 million polys, rendering out at 3000 x 1276), rendering just geometry (no textures or shaders) under GI skymap and one raytrace-shadowed parallel light (for sunlight). A full-res test render out of EIAS took 17 minutes. However running the animation through Renderama, my frames are now taking an hour each, and there's 825 of them. I have 11 rendering cores at my disposal (across 2 machines), so it's not desperately bad time-wise, but I'm curious as to why the difference from 17 minutes to an hour.

Camera settings are as follows:

Hardware info: Memory 32768

Processors: 16

Current Settings

Architecture 32bits

Use Memory, Mb: 2000

Run Threads: 16

Is this right? Or should I have it set to something different?

David

[Tomas Egger]

Ola David!

Your test was done in your local Camera with 16 threads in 17 minutes ok?

If you setup to render directly in Local Camera without Rama, How many time per frame?

In the local Camera, Are you using 32 or 64 bits?

Now, the second machine, the slave, How many cores? 32 bits or 64 bits?

Thanks

Tom

[verb]

I've also been doing some tests as I've got a large project to render with limited resources.

 

In my testing, it's quite a bit quicker to allocate 1 camera to each CPU thread - in my case an 8 core mac pro so 16 threads. (I also tested 1 camera per core but 1 per thread is significantly faster)

 

Here are my test results

 

 

- 16 Cameras given 1 CPU thread per camera

Allocated 1 camera to 1 single thread, Frame 78 took 2 hours 36 minutes to render 720p at 3/4 size. This config gives me 16 frames in 156 minutes.

 

- 1 camera given all 16 threads - multi core

Allocated 1 camera the full 16 threads, frame 78 took 18min 42 seconds to render 720p 3/4. This gives me 16 frames in 283 minutes

 

So if the multi core camera rendered 16 frames it would have taken 283 minutes. Whereas allocating 1 camera to each CPU thread x 16 threads give me 16 frames in 156 minutes - almost 50% less time.

 

Of cause this is based on one specific frame - 78, each frame will have a different render time. This was all done on a mac 10.8.2 in 32bit mode. Ram allocation didn't play much of a part - the scene only uses about 250mb. It might be different on a PC.

 

Anyway feel free to pick holes in my crappy test situation, but until I'm shown this is wrong, I'll be rendering 1 camera per thread!

[Tomas Egger]

Ola Verb,

 

This kind of tests are really specific, some will be faster like yours, others, slower.

Huge scenes are needing one Camera only + 64 bits on most of the cases.

 

Thanks

 

Tom