[Steve Bentley]

A couple of things to keep in mind. If the scene is large it may not fit in the GPU ram and then it just won’t render, so back you go to CPU rendering which usually has more machine ram to use and disk swap space – in theory, it’s unlimited.

If you pole the guys on the web who use C4D professionally I think you will find a lot of threadrippers – so that means more and faster CPU cores. Now to be fair, those who can afford threadrippers generally have a few GPU’s chained together too. And while they are doing a huge amount of their rendering on GPU they are still relying on that threadripper a significant amount. And that can’t be a coincidence (or a waste of money). There are also limits to how many GPU’s you can tie together efficiently.

Depending on your render engine you may also find that due to sampling and customizing the render engine for the task, the CPU engine will run faster, even though it has fewer buckets to throw at it. And check the denoiser you might be tempted to use – it may engage the CPU even at the end of a GPU run. The denoiser can take longer than the render.

Some of the third party engines are both GPU and CPU but a few of them aren’t
really mature yet in the GPU category. In the trade GPU is often used
for look/dev and then the file is thrown against a bunch of CPUs for
final output.

The standard and Physical engines both use the CPU so again, seeing lots of buckets will depend on how many threads your CPU has. And there is nothing wrong with these engines. They have gotten a bad rep because they don’t have a progressive mode that is so helpful in look/dev. But they are excellent render engines and can render stuff realistically if you know what you are doing.

There are also some issues with flavors of GPU’s and depending on platform (like Mac) you may be stuck using CPU for what you need (and this also depends on what render engine you pick). There’s also compatibility issues with open GL versions and Open CL. A modern Nvidia should have no issues, but an older one or an AMD might run you up against a brick wall on the GPU side of things. Not saying they will, it’s just your chances are higher for an incompatibility issue. Also consider what your compositing engine uses – might as well have a machine that does both well.

Coding for iterative and recursive sims (like evolving particles) is really hard on a GPU at the moment so I could guess that Indsydium is CPU weighted – the render might not need to be but the simming part will be I’ll bet.

Ironically, the answer usually comes down to budget rather than what is going to be best. Threadrippers aren’t cheap and multiple GPUs can set you back as well. Building the “ideal”machine can break the bank so you often need to shoot for a cheaper “sweet spot”. Consider also smaller cheaper machines on a network that can work on the frames you are working on simultaneously to give you faster response times. More, cheaper CPU’s spread around can often be cheaper per core than one big monster.

Let us know what you end up with.

(I love the “faster single core GPU” from Maxon. By “single”, do you suppose they mean “thousands”? I think you are right in your assumption there)

[Steve Bentley]

Do you have the project to play with?

I’m going to take a stab at what it might be and it may help you along. (I couldn’t open the images for some reason so they remain small thumbnails in the post – too small to see what’s going on)

The first is that the node system is still in beta and those growing pains may include some of the more parametric material nodes. With that “excuse” out of the way….

Things like noises are no long colors in the true sense in the node editor, they are now considered Floats (16 or 32 bit data). So often you will have to covert a float to a color or visa versa to get anything meaningful. Some of the procedurals are vector based so even though they look like colors (the RGB pixel telling cinema which way to point the vector) they are, again, not a true color and while you can convert these to floats or colors they won’t really make any sense when you do so. Then in some cases you can use a float noise but you will only get the red channel say. This will still work as intended because the engine just takes the values out of that one channel and ignores what you perceive as a color.

So try a conversion node between your procedurals and your input and see if that helps. If not send the project along and we’ll take a look.

[Steve Bentley]

Hey Daz,

I’ve never really understood why we see the parametric instead of the actual spline shape until you deselect it. You would think that would be helpful.

In place of this I usually just throw a short length extrude on the spline and continue editing – the spline stays smoother than it really is but the extrude will update in real time and show you all the kinks and folds the way the spline will render or affect other objects.

[Steve Bentley]

There’s nothing here that can’t be done inside of AE – the only tricky bits are the fully 3D objects (like the Bradley and the Phone for instance) but even those can be rendered inside of AE with Cineware (that comes with C4D) or a plug in like Video Copilot’s Element3D. And then you aren’t hopping back and forth between C4D and AE when changes are made, it’s all there in one place.

It will also generally render faster in AE (especially if you do the models in C4D and the text in AE as Zeyd suggests)

The only gotcha to watch out for is that not everything in AE is fully 3D aware, so sometimes you have to do tricky mattes or special layering to get things that were in front of an object to pass behind that object (when you rotate the camera for instance). The layer stack still rules in AE as far as what renders on top of what and anytime that rule can be broken it’s usually a setting in the plugin that gets around the problem with a trick. (Trapcode has some settings for this)

[Steve Bentley]

While it’s dangerous to say “never” with C4D, I very much doubt you can assign a polygon selection to an ID tag. How a poly renders (with a phong tag) is very much dependent on the poly normal next door so that selection would need more information to render correctly than an ID tag could supply.

But you can certainly get there with a full white luma texture (no specular) assigned to a polygon selection tag on an object, that also has a black texture (all channels off) covering everything else.

[Steve Bentley]

Xpresso could do it and you wouldn’t need a tag on every object (not sure why the poster thought you would). You could use a link list and just iterate through the objects in the list and set each object’s visibility or object ID that way. You could do it based on frame# or just a count. So many ways. And you wouldn’t need to involve Octane in the process – it would just render what you give it.

[Steve Bentley]

The booles work well when the geometry of both objects has roughly (or even better, exactly) the same size of subdivisions, but that’s a rarity. I’m not sure why it’s so bad in C4D. Most other packages use booling as a normal part of the process and no issues.

[Steve Bentley]

What about using an effector and using the visibility button? Not sure what the test is for when to hide, but one of the effectors (or effector with field) will do the trick.

Any more info?

[Steve Bentley]

I see what you are doing now (thanks for the pics, I was scratching my head to figure out how a radial clone could have all the hemisphere’s centers all in the same spot).

That’s an interesting little problem. There are lots of ways to do this and I would probably normally approach this with xpresso or python (when you have a hammer…. or a driven/setDriven combo), but if you want to use clones, have a look at the linked file.

If you you use a plane effector with a linearField in the falloff and then a stepfield affecting that linear field you can get there.

The trick is to set the number of “steps” you want this to happen in within the steps field (the number of shells), then set the Max value of the remapping tab in the linear field to 200%. Animate the linear field across the clones and you get a stepped shell closing.

I have changed a few things in the file – I have used a linear cloner because it’s easier to control the angle the shells rotate on – when in a radial clone they are all spread around a circle but they are also rotated to face inward so rotating in the “X” will rotate them about the axis of the tangent to the circle where ever they happen to be on the circumference. Now thinking about that, you might be able to turn off “align clone” and you can keep your radial clone the way it is. Yep, just checked; this does work.

You can look through the mechanics of the file and just apply it to your file if I’ve changed too much.

One down side to this approach is I don’t think there is any way to ease in or ease out the motion. Animations that start or stop instantly are usually the sign of a beginner since nothing does that in real life (that’s not to say you can’t do that if instant start and stop are desired, or, in the case of the spheres appearing from out of the plane, it might not matter). A delay field on top of the other two might do the trick though to smooth the start and stop of the rotation.

Also, Bools in C4D aren’t the best. They are clunky and don’t always produce the best geometry and can often shimmer in animations. If you don’t need to keep the hemisphere parametric, you can use an actual hemisphere type of the standard sphere and make it editable and then add a cloth object as it’s parent. Then set a thickness in the cloth object. This way you only have two objects under the clone (hemisphere and cloth). I haven’t done that in the file I sent but let me know if you need to me to show you.

https://drive.google.com/file/d/11FavJXyj4pzRdjJhz2pjlp2B1nSv5fZK/view?usp=sharing

[Steve Bentley]

I wont be able to look again for a few hours but in the mean time try a step effector. That might do the trick.