Hi Dan, you're welcome :)

Best of luck with the new purchase, lots of exciting new hardware coming onto the market right now too, so interested to hear what you buy and how it helps :)

Please feel free to come back to the forum anytime and post more questions, hopefully it will help everyone in the same situation or who has the same question as you :)

Thanks for the very detailed answer James, you're a legend. I hadn't even considered the GPU VRAM as a factor!

I've checked and my most complex scene so far is only 5GB, so an 8GB card should give me some headroom, I also suspect that there is a lot of optimisation that I should be doing on my scenes, especially with materials - I need to read up on that at some point, right now I'm just trying to get work delivered to my clients - I'm just about managing to blag everything right now :-)

Hi Dan,

Certainly adding a GPU will help for sure :) the only drawback of GPUs at the moment is their fixed VRAM. VRAM is important since if your scene is larger than the amount of VRAM you have, the GPU will not render, regardless of how much system RAM you have.

Having said that RTX 2070s, I believe, now have about 8GB of RAM, which is probably alright unless you're doing complex scenes. The higher your poly count and texture resolution/number of textures, the more memory your scene will consume. Also things like micro displacement are the nemesis of your RAM. I once had a scene on disk that was about a few MB. But I got cocky and decided to check micro displacement on a landscape mesh. When it finally finished being 'loaded' into RAM, it was 55GB. Luckily my system can use the SSD to supplement main RAM, but it was incredible slow. FYI, ray tracers like cycles apparently spend most of their time shuffling data around between the compute units and RAM. So access times for your RAM are critical to getting good render times.

So once you go from using RAM modules to accessing your SSD, the time required for these operations explodes, check out this resource if you're interested -> https://www.prowesscorp.com/computer-latency-at-a-human-scale/

As you can see once you go from accessing RAM to an NVMe SSD, the slow down is massive. If we put it in human scales like the article above does, RAM access is 4 minutes (if we consider one tick of your CPU to be 1 second), SSD is 17 hours!

So when it comes to your upgrade, RAM vs how large your scene is should be something you look into. You can survey your projects quite easily by rendering a single frame of each and looking at the data of how much RAM you need. Then you can more accurately buy hardware with confidence you won't get stuck. Both for GPU and CPU.

Pros and Cons for CPU vs GPU? GPUs have for a long time held a speed advantage for their cost, also if you have the right motherboard you can usually upgrade with a couple of cards. To build an equivalently powerful CPU only machine starts to get really expensive, so bang for buck wise, it's clear GPUs are a great solution, however, just be careful about the VRAM limitation, if you go over, your GPU is useless.

Hope this helps :)

James