Multi GPU / One Frame for one GPU

[NVN]

open 7 copies of Maxon C4D. Select one gpu for each one. Divide sequence to 7 and render. This is most effective way to distribute sources.

thx for ur solution
but effective way....no.

Why is there no option to choose which way i want to go?
Which solution will Otoy will have for our "fram"??
500 GPUs for 1 Frame????

I dont need 16k Samples for every Frame......

[acc24ex]

OK, if you have enough VRAM turn all objects to moveable proxy - now the whole scene is in the memory and if you only move objects, it shouldn't "export" or voxelize each time and dump entire scene to GPUs.. it should be quick

- try it and see

[NVN]

OK, if you have enough VRAM turn all objects to moveable proxy - now the whole scene is in the memory and if you only move objects, it shouldn't "export" or voxelize each time and dump entire scene to GPUs.. it should be quick

- try it and see

lol...we have automatic detected Export....
what ever he export only the first frame, thats why the first frame tooks longer.
And there is only Camera movement.....all works fine and seems ok!

Its not only a Export Question!
If u have to mutch GPUs its better to render each Frame with 1 GPU.
If u need more Samples (because Causics, transparency...) and you have a big scene u need more GPU for your Picture.
If u have a small scene with low-mid samples its better to render each Frame/1-3 GPU

[aoktar]

open 7 copies of Maxon C4D. Select one gpu for each one. Divide sequence to 7 and render. This is most effective way to distribute sources.

thx for ur solution
but effective way....no.

Why is there no option to choose which way i want to go?
Which solution will Otoy will have for our "fram"??
500 GPUs for 1 Frame????

I dont need 16k Samples for every Frame......

Of course i'm not serious partially. Which way you mean? That's impossible for plugin. We have a straight workflow to detect/arrange data so thank you. Export/prep times will be very small delta in whole rendering time when you have 16k maxsamples. Otoy solution will not be available. But it has some handicaps. You need to export scene and upload. This can be longer then rendering in some cases.

[NVN]

open 7 copies of Maxon C4D. Select one gpu for each one. Divide sequence to 7 and render. This is most effective way to distribute sources.

thx for ur solution
but effective way....no.

Why is there no option to choose which way i want to go?
Which solution will Otoy will have for our "fram"??
500 GPUs for 1 Frame????

I dont need 16k Samples for every Frame......

Of course i'm not serious partially. Which way you mean? That's impossible for plugin. We have a straight workflow to detect/arrange data so thank you. Export/prep times will be very small delta in whole rendering time when you have 16k maxsamples. Otoy solution will not be available. But it has some handicaps. You need to export scene and upload. This can be longer then rendering in some cases.

?????
I said nothing about ur plugin and low Export speed.

ALL what i want is:
If i have 10 Gpu's i will rendering 10 Frames simultaneously (each Frame by one GPU) because in some scenes it will be better! Nothing else...
And i wish me from Otoy that i can chosse between two options:

a) Rendering with all GPU's
b) Rendering each Frame with one GPU.

As i heard Royal Render works on this way.......so why not Otoy?

[karanis]

When it comes to distributed calculations on a network of heterogeneous computing nodes, there are many many things to consider. You can not expect a software or a plugin to do all of the task for you. Focus on the bottleneck.

Why is there no option to choose which way i want to go?
Which solution will Otoy will have for our "fram"??
500 GPUs for 1 Frame????

AOktar`s solution of "Running 7 individual C4D ..." was a pretty nice trick but on the other hand, there is already a solution already from OTOY.
If you are wealthy enough to finance a farm of 500 GPUs, then you are already wise enough to write down some lines of code.

Use lua scripting.
octane.render.setRenderDeviceActive(deviceIndex, setActive)
( If C4D plugin does not allow you to script, export the scene for standalone )

[glimpse]

Here is a Screenshot (its a extreme Example) but one Case....
Dont forget my 2 GPU's are 780, not 980TI like my Renderslave.

Sooo...one GPU needs ~8 Seconds.
With 10 GPU i have 10 Frames in 8 Seconds.
As u can see, my Renderslave needs with 9 GPU's (~2 Seconds/Frame) 20 Seconds......

sorry, what lan solution You use??? That's the biggest bottle neck as far as I understand..
work on 7GPU machine with one animation & other machine with other part of animation.

what OTOY use on their ORC service will probably be nodes with at least 10Gigabit network adapters..
- & it's quite logical why..

question what are You using? maybe You focusing on the software when You should be focusing on hardware first..

[NVN]

Here is a Screenshot (its a extreme Example) but one Case....
Dont forget my 2 GPU's are 780, not 980TI like my Renderslave.

Sooo...one GPU needs ~8 Seconds.
With 10 GPU i have 10 Frames in 8 Seconds.
As u can see, my Renderslave needs with 9 GPU's (~2 Seconds/Frame) 20 Seconds......

sorry, what lan solution You use??? That's the biggest bottle neck as far as I understand..
work on 7GPU machine with one animation & other machine with other part of animation.

what OTOY use on their ORC service will probably be nodes with at least 10Gigabit network adapters..
- & it's quite logical why..

question what are You using? maybe You focusing on the software when You should be focusing on hardware first..

Good morning glimpse,

the problem is not the lan connection i think. We have 1 Gig?
The main problem is if u have mutch and big cards somtime is your Rig to big for your Scene.

Another little example:

You have a scene which takes 5 Seconds with 1 GPU.
When u have 5 GPU the rendering takes 1 Seconds.

IF you have 20 GPU the rendering takes also 1 Seconds......because it could not be faster.
In this example it would be better to splitt the animationen because the the last 15 GPU can't bring their full potential.


Another Question is for Otoys Renderfarm:

I have a scene that takes 10 min with my RIg (2GPU's)
So i would upload my scene to their farm and they rendering it with 500 GPU's in 1 Second.

But what if 200 GPUS can render it also in one Second? The other 300 GPU's bring nothing....

Could you understand what i mean?

[karanis]

This thread seems to be so theroetical and useless,

1. If you have a scene that renders in 2 seconds on a single GPU, you don`t need to use multi GPU`s / renderslaves in real life.
(if you are rendering a movie of 2 hours = 120 mins = 7200 seconds = 180,000 frames ,
then you`ll spend
2 seconds / frame => 50 seconds rendering for 1 second of movie => 50 x 7200 seconds => 50 x 2 hours = 100 hours = 4 days of rendering. Let it render mate, no worrys)

2. If you use Octane, the output that you`ll get from a single GPU in 2 seconds will not unleash the power of Octane. Use some other software / rendering algorithm.

3. It is so ridicilous to talk about distributing a frame of 2 seconds to render in a network or a multi GPU system in this forum.


Learn TCP/IP, try to understand SYN/ACK thing... Read about bus speeds, think of disk access timings...

But what if 200 GPUS can render it also in one Second? The other 300 GPU's bring nothing....

They are waiting for the next job.

[NVN]

This thread seems to be so theroetical and useless,

1. If you have a scene that renders in 2 seconds on a single GPU, you don`t need to use multi GPU`s / renderslaves in real life.
(if you are rendering a movie of 2 hours = 120 mins = 7200 seconds = 180,000 frames ,
then you`ll spend
2 seconds / frame => 50 seconds rendering for 1 second of movie => 50 x 7200 seconds => 50 x 2 hours = 100 hours = 4 days of rendering. Let it render mate, no worrys)

2. If you use Octane, the output that you`ll get from a single GPU in 2 seconds will not unleash the power of Octane. Use some other software / rendering algorithm.

3. It is so ridicilous to talk about distributing a frame of 2 seconds to render in a network or a multi GPU system in this forum.


Learn TCP/IP, try to understand SYN/ACK thing... Read about bus speeds, think of disk access timings...

Did you ever look at my Picture on page one .....?!!?!?!?
I think you dont understand the whole thread.......
SryI'm tired to repeat me all the time....


Its a simple example....but fuck up---> it is hopeless