Speed up octane - suggestion

[[email protected]]

Hi, since octane 2 is much slower than previous versions, and Karba mentioned it's because new features were implemented like Motion blur, displacement and other stuff...

is it possible to make like "optimize" rollout, where we can uncheck features that we will not use in current project, for example you have Motion blur checkbox, displacement checkbox, hair-fur checkbox, and when we disable them - octane speeds up

by optimization rollout - I kinda think that when you uncheck displacement for example, octane will appear like a complete version without displacement - so theoretically it will speed up the whole thing

I guess this is not possible... but what do octane team think about it?

[mykola1985]

Buy newer or more GPUs. This should increase the speed

[enigmasi]

but he is right, it shouldn't be slower since you don't use that feature

[suvakas]

It is slower because the engine has extra code in it and it's not excluded even when you don't use the features.
It's the way GPU programming is atm. Vray builds it's GPU "code" dynamically according to what features are used. This makes the speed more adaptive.

Suv

[[email protected]]

@ suvakas - yeah something like that, when function is not used, code get's smaller

[bcravin]

IMO Octane kind of shot itself in the foot. They released a renderer that was far from full featured. It was fast as hell but didn't have the features needed to make it a studio worthy renderer like render passes, motion blur, Hair/Fur, atmospherics. But they do have a large client base that is totally built around SPEED SPEED SPEED. Now they overhauled the code to add all the features it should have had in the first place. From what I can tell, every new feature they add is going to slow it down more and more. Until, it doesn't really have any advantage over other GPU or maybe even CPU renderers. I see a lot of posts about the speed hit from updates. Sadly, I think its only going to get worse. So get used to it people

Overall, I'm fine with the speeds because I can just buy more GPU's and they are Cheap (gtx 780 6gb <$600) I just wish that they would give us 2 or 3 renderslave licenses per purchased license to make up for these speed hits that are bound to happen in the future.

What I hope OTOY is working on, is a simple one button click cloud rendering service. What I mean is, I want to be able to work on my local cluster 12 or more gpu's. And when its time to render the final sequence. I click render on cloud. and select how many gpu's I want, lets say 100. It automatically transfers my scene to my cloud account, renders it and downloads the final file sequence. I should be 1$ per gpu/hr. That will make it very simple and easy to calculate costs for productions. Right now the AMS cloud service is just far to complicated. If they want it to take off like it should things must be simplified. And it must be built into the plugins.

This is not a rant in any way, just some observation and suggestions. I bought into Octane because I believe GPU rendering is the future. And so far Octane has helped make it a reality.

[Goldorak]

IMO Octane kind of shot itself in the foot. They released a renderer that was far from full featured. It was fast as hell but didn't have the features needed to make it a studio worthy renderer like render passes, motion blur, Hair/Fur, atmospherics. But they do have a large client base that is totally built around SPEED SPEED SPEED. Now they overhauled the code to add all the features it should have had in the first place. From what I can tell, every new feature they add is going to slow it down more and more. Until, it doesn't really have any advantage over other GPU or maybe even CPU renderers. I see a lot of posts about the speed hit from updates. Sadly, I think its only going to get worse. So get used to it people

Overall, I'm fine with the speeds because I can just buy more GPU's and they are Cheap (gtx 780 6gb <$600) I just wish that they would give us 2 or 3 renderslave licenses per purchased license to make up for these speed hits that are bound to happen in the future.

What I hope OTOY is working on, is a simple one button click cloud rendering service. What I mean is, I want to be able to work on my local cluster 12 or more gpu's. And when its time to render the final sequence. I click render on cloud. and select how many gpu's I want, lets say 100. It automatically transfers my scene to my cloud account, renders it and downloads the final file sequence. I should be 1$ per gpu/hr. That will make it very simple and easy to calculate costs for productions. Right now the AMS cloud service is just far to complicated. If they want it to take off like it should things must be simplified. And it must be built into the plugins.

This is not a rant in any way, just some observation and suggestions. I bought into Octane because I believe GPU rendering is the future. And so far Octane has helped make it a reality.

The cloud service will work very much like what you are describing. It is being built deep into the core engine, so 2.0 plugins can use it, like they do now with network rendering,

Regarding speed, we can push more out of 2.x, we'll revisit that once render passes are done,

[muski]

If speed continues to be a problem in future versions then a free (or very affordable) network render licence would be a good way to compensate the already paying users for the unexpectedly lower speed we get on 2.xx.

[bcravin]

Very excited for Render Passes! Does the team expect another hit on render speeds with Render Passes. With my understanding of other renders certain passes such as Normals, Depth, World Position, would not effect render speeds very much because they are all calculated anyways during render. So most of the hit is just writing them out. Does Octane work the same way?

Also, will we be able to embed the channels inside of EXR's for compositing inside Nuke and other compositors?

[acc24ex]

- actually yes, the network render that you have to pay for is a bit of a smack in the face - come on, I was thinking I could network like 5 more computers with fairly weak GPUs - just to get that extra power - this way it's like you really have to have 1 other computer with more GPUs for you to justify buying it - I have another octane standalone licence and it's gonna stay unused .. at least give the option to use the 1.0 licence.. let's meet halfway

.. yeah, maybe the first oversimplified approach could be for us to choose a 1.3v, 1.5v or 2.0 kernel
- the new features are great, but still if you do not use them 80% of the time, why would you need the more advanced and slower kernel..
- and I did get more GPUs - fitted 4 590gtx and a 580 - interiors are still slow - so I'm guessing I have to also do a hell of a lot more optimisation on my part

- nvidia is also not helping - to be honest - octane made progress in the past two years - and their main and only plaform - CUDA hardware - didn't make almost any progress - what - measly 10% increase in speed - only with 4-6 GB of VRAM - with prices going above 5k - like nvidia titan Z - there is a lot of valid reason to trash on nvidia - they are still making hardware just to suite gamers - either CUDA cores became twice as slow than two years ago - or they are seriously faking it

- how can you justify having 1024 cuda cores from 590gtx perform in almost identical speed like on the lates gtx Titan with at least 2000 cuda cores - which took two years to develop ? I would really love to see some concrete explanation on that front.. looks like they gave up on developers, and instead went for the money - like any true corporation does every damn time

- if you look at it, the bitcoin hashing cards seem to perform 10 times faster than CUDA counterparts - maybe use that as the new platforms, cause people are a lot more enthusiastic about mining for virtual coins, than for developing code or actually working on the damn computers - it's games, and collecting coins .. go figure