24gb of vram just to render these splines? Please help with this, I'm doing something wrong with Houdini + Redshift, it can't take 22gb of VRAM just for these few items. I attach photos and the scene to understand the correct way to send this to render.

Redshift will reserve 90% of your gpu memory by default even if it not gonna need that much. you can control that behavior in the redshift node in the system tab.

Hello Jacquesf! Thanks for your explanation. I am a regular user of Octane/C4D and the speed of loading and preparing the render of Octane in Cinema4D is actually much faster than here with Houdini 19.5/Redshift 3.5. There is no point in comparison, on a day-to-day basis, working so slowly to preview is a real kick in the ass.

Any way to speed up all this? Maybe some setup that can load the geometries faster in Redshift/Houdini?

I'm afraid I stopped using redshift a while ago in profit of karma xpu.

but I don't remember redshift being that slow, I used to render character with multi million of strand of fur and the wait time wasn't that drastic.

I unfortunately cannot test your scene as it don't work out of the box, but check that you aren't generating billion of polygon by mistake by middle clicking on the node to display the geometry information.
whatever you show on your screenshot shouldn't take up more than a second to initialize.

RS performance is not measured by how much VRAM it uses.

If your system is validated within comparable performance range (by running RS benchmark), then there is some other factor that is affecting performance.

There are multiple tools to diagnose where time was spent for rendering a frame. The RS Logs is the best place to acquire that information.

If scene extraction (of your splines) is what’s taking up time, there are several ways to remove that cost, one of those is RS Proxy.

I would add here, that this (SideFX) forum is not monitored by RS Support - I would suggest making a post on the Maxon forums, where the dedicated RS Houdini plugin developer can assist.