Sry to start this again but I would like to know your tricks.

I'm using Redshift for a while now and must say its rediculously fast compared to anything I experienced with Mantra. 50 km displaced ocean surface it renders me in a total of 13 seconds, combined with 15 million instanced polygon spheres for whitewater it takes 7 minutes or one full HD frame. In the beginning of the next year we will get direct particle rendering what will raise the particle limits considerably and reduce rendertimes even more.

But as much as people trying to push GPUs a lot of people permit on their CPU workflow. Just a comparison:

https://vimeo.com/195146295 [vimeo.com] this took 7 minutes per frame with 1920*1080 - Redshift
https://vimeo.com/157951443 [vimeo.com] this took 45 minutes per frame with 1280*720 - Mantra

While Mantra is still preparing the displacement of a few km ocean RS is crushing through a lot, lot,lot of frames. With 12 GB of VRAM I already feel quite limitless. With some tricks such as “screen based adaptive” tesselation I can render large scale displacements without any problem at all. Things that would fill up a whole lot more system ram.

I was wondering: what are the tricks to speed up Mantra. A lot of guys are defending it, kind of “you just don't know how” So how? I mean, when I don't have access to a huge renderfarm while working as a freelancer.

Some alembic tricks, a bit of packing?

Mantra can not be fast as Redshift, unless it gets GPU support, simple as that. Now there are many other things in production that are also important, it is not just about speed.

I know but when I hear “nonono, Mantra can be superduperfast, you just have to know how” I would like to know what makes people think so. In a production environment it just doesn't matter if you have access to a lot of CPUs but as a freelancer no one can tell me that Mantra is worth it.

Depend is from what pov you value it. If you value it from speed point comparing it with GPU renders, than yeah it is not worth it, but than that comparison is not fair either. I find it ‘normal’ speed for CPU Path Tracer. As I said I do not think that you can do anything to make it fast as Redshift, whatever you do. It is same as I would say well I do not think Redhift worth it as I can fit 32 GB scene in it … Also maybe some people did not try Redshift and do not know what is speed

Well, you can pput much bigger scenes in 12 GB VRAM with Redshift than in 24 GB System RAM Like I said, “screen based adaptive tesselation” foe example lets you render huge, huge displacement scenes with very little impact on memory. The usage of availyble ressources is just way more efficiant. Also you can go out of core if your VRAM actually is full - what really rarely happens. There are already 24 GB GPU cards available. So I can't imagine GPU acceleration won't be the future. Since I'm rendering on GPU I just can't go back and wait forever anymore

I think its absolutely fair to compare those two. I don't care for “its fast for a CPU solution”. Its like saying “well, a photoreal painting is quite fast for being done by hand”. If I want a photoreal photo I want it fast, don't care if a hand drawn picture is done fast for what it is

You've used so many caveats that your argument directly leads to your conclusion.