Thanks for the breakdown in time savings, really helps put things in perspective. In your case the 3090 really was a no brainer since the 1080 is already reasonably fast and it wouldn't have made sense to get something with less memory (which you were already exceeding in some of your renders). I'm actually surprised that the render time difference is THAT dramatic. That really is impressive even if you ignore the memory issue.I've just decided to make the jump to a 3090. Found a KFA2 version on Amazon for around 1,700 EUR (the cheapest I've seen for a while). It was only up for about an hour before it sold out. Ordered it Sunday evening with 1 week for delivery, but arrived this morning (Tuesday). This was an upgrade from my dedicated 2x1080 GTX (non SLI as I was using 1 for display, 1 for rendering). I'll also be using it for VR and I had a fair few amazon vouchers to use up so I thought why not. I already had a 850W power and 64GB RAM so I no need to upgrade them.
Hopefully, this should give you an idea of the types of benefits (and drawbacks) from this type of upgrade:
Overall, after 1 days usage, I would say it appears to be worth the upgrade, purely on a time saving basis. As I'm currently working on my first game where I expect to be rendering virtually constantly for the next 6 months, the time savings alone (both on rendering and how it's making the scenes easier to setup) should make up for the cost, even at the inflated price. Of course, only time will tell.
- Performance is brilliant compared to my previous. As an example, this render:
- 4 vehicles, 4 people, 1 ground plane, hdri light and a emissive plane (+ 1 axe). Previously, I had to remove the Fire Engine in the back as it would constantly cause the render to hit CPU rather than GPU. I also had to split the image in 2 (2 girls on the left, front fire engine + police car then 2 girls on the right, ambulance and police car) and then merge into one. Each render took over an hour each time so probably overall time for the render was 3 hours. With the new card, I could include the engine in the back and all other characters and vehicles and it complete fully in 10 minutes. And a quick post work, and overall time was now 15 mins. So already a saving of around 2:45.
- Using the iRay viewport is now so much smoother. Previously, I would have to setup in texture view, switch to iRay (or a quick render), back to texture for adjustments etc. Now, I can do most directly with the iRay view and I get virtually instant feedback on my changes (the exception is if you have a seen with a lot of geometry edits (like using Mesh Smoother) as these rely on CPU for the calculation)
- It's bloody hot. The card itself had good heat dissipation (was not too hot to touch) however, by doing this, it dissipated the heat to the box. This meant the temperatures of the box as a whole started to increase to noticeable levels. This was after a few hours of nearly constant rendering however (was doing a lot of tests). Simply opening the cover of the box allowed it to cool down, but looks like I will need to get some better box cooling.
- It's big, pretty much taking up the same amount of room as the two GTXs, so make sure your box is big enough to get enough free air circulating around it. I would not want to put it too close to any key components as the dissipation will heat them too.
- Sound wise, it's pretty quite, only really making noticeable noise when I had the issue with the box cooling. Here, I could start hearing the fans spinning up (with the closed box and sitting about 1.5 metres away). As soon as I opened the box and let the system cool for a bit, I no longer had the problem. Now, when I render, I can barely here the difference (just the usual quite hum of the pc that you can hear because the box is open)
I would imagine rendering the same image on a 3060 would take more than twice as long, maybe even 3-4x as long and that's assuming it would even fit inside that 12GB of VRAM, which it might very well not. It would still be a lot faster than what I have now though. I imagine for me to render that image I would have to break it into 3 or 4 pieces rendered separately since I only have 4GB to work with.
For an entry level render box I'm actually more concerned that 12GB might still not quite be enough to avoid having to optimize. I could deal with it being a bit slower but not if I also have to waste a bunch of time on optimizations. I think the main reason to get the 3060 is if I know for sure I just want to use it as a placeholder and grab the top end 40xx card as soon as it comes out but it might still make more sense to keep waiting until either a 16GB 3070 or 20GB 3080 comes out since I think those will both be more likely to have enough VRAM that optimizations are no longer likely to be needed. I think 24GB is overkill for all but the craziest of scenes but I know some of the scenes I've done would exceed 12GB (and they took 8+ hours per image to render on my rig).
