OTOY Forums

Hey All,

I am mainly using octane for rendering animations.

And sometimes the majority of the time it takes to render something (escpecially when rendering tests with low sample limits) it takes around 30 seconds to process the geometry... and only 5 seconds until the rendering of the image is done.

I am using a rather old mainboard with 3 gpus (PCIe 3.0 8x / 8x and PCIe 2.0 8x) (z77, sandy bridge i7 - 3770k).

If the scene would consist of static geometry, disabling the "update animated geometry" feature makes it a lot faster. But I guess when the scenes mainly consist of moving objects, my 3d program (3dsmax 2011) has to upload the geometry into the GPUs for every single frame.

So I guess my bottleneck is the PCIe Speed?

Is there any way to speed up the process? I can't believe that uploading around 1 GB takes 20 - 30 seconds - even with the x8 speed.


I guess for some jobs it would actually be faster to take out the other 2 GPUs and use the fastest that I have with the full PCIe 3.0 16x speed. Instead of two times X8 speed.
Is it possible to just disable the other two GPUs in my system manager (win7) to make all Lanes available for the one GPU? Or would I actually have to take the other two GPUs out of my mainboard?

Is the speed of PCIe 3.0 x16 exactly twice as fast as x8? Has anyone ever benchmarked this?



Thank you in advance,
alex

I׳m using 2x Titan X with PCIe 3.0 x16 and still same as you having relatively very long translating time.
I׳m also working mostly on animations and its the weakest side of octane for me.

Same as you - I׳ll be really happy to know if there is some ways to speed that process up.

Use less polygons, lower subdivisions and partial updates as much as possible. Do you have lot of changes on meshes on each frames?

Yes unfortunately. It's a moving ocean with a couple of boats on it... so basically every single polygon in the scene is moving... :/

So use a few copies of program if you can. Give different part of animations.

do you mean starting octane a couple of times on the same machine? That is a good idea. Is that possible? I only have one license.

Or do you mean to start it on different machines? That would mean I'd have to buy more full 3dsmax licenses right?

a1x wrote:do you mean starting octane a couple of times on the same machine? That is a good idea. Is that possible? I only have one license.

Or do you mean to start it on different machines? That would mean I'd have to buy more full 3dsmax licenses right?

On single pc, I do it for c4d with single license, so max should do. Of course it can spend 2x vram and ram.

Thank you. That is a very cool suggestion. I guess it helps getting the most performance out of the pcie slots / cpu etc. all the parts which are at idle most of the time anyway.

Btw. 3dsmax creates a placeholder file for every frame it starts to render. That means I could let both copies render the exact same sequence by enabling "skip existing frames" and it would just render until everything is done. No juggling with estimating sequence lengths.

Thank you! It's a cool hack. But I also wish the whole process would be faster.

alright. Tried it with my 3000 frame animation yesterday.

And I was wrong. 3dsmax does not create placeholder files... so I had to set the sequence parts manually.

With one instance of 3dsmax it rendered one frame at a time in 20 seconds (I heavily optimized the scene)
With three instances of 3dsmax it rendered 3 frames at a time in 30 seconds... so effectively 10 seconds per frame. Twice as fast. And GPU usage was almost at 100% all the time. Instead of having pauses for around 10 - 15 seconds.

So as you said, it can be very effective to use more than one instance for rendering animations.

a1x wrote:alright. Tried it with my 3000 frame animation yesterday.

And I was wrong. 3dsmax does not create placeholder files... so I had to set the sequence parts manually.

With one instance of 3dsmax it rendered one frame at a time in 20 seconds (I heavily optimized the scene)
With three instances of 3dsmax it rendered 3 frames at a time in 30 seconds... so effectively 10 seconds per frame. Twice as fast. And GPU usage was almost at 100% all the time. Instead of having pauses for around 10 - 15 seconds.

So as you said, it can be very effective to use more than one instance for rendering animations.

Yes, when i had no time and similar scenarios i was using this system for my jobs. Nice to see helped you.