GPU is more important if you use it more for rendering.
But today people still use cpu based renderings because it’s more robust and it won’t leave you hanging.
So it really depends on your workflow.
That said a solid gpu is still a must, because viewport performance.
I’m definitely getting 2080ti because viewport performance on eevee.
I see some recommendations for old 6 core cpu. I would advise against it.
Today 6-cores are obsolete for workstation. They are mainstream in gaming.
Far better choice would be ryzen 7/9 lineup or threadripper cpu.
Don’t choose hardware based on former experience.
It’s like buying Pentium today because 20 years ago they were bomb.
While you’re on the right track, don’t forget that multicore CPU are also playing their strong card during rendering. There is not as much multithreading going on in general Blender operation as people might think. Also the main point for CPU rendering still today is RAM limitations on GPU.
That is true. Multithread is only utilised in sculpting as far as i know.
But bear in mind that you are also buying hardware for couple of years to come.
Also, there is more to it than number of cores. Cache size and IPC also play a role.
I suspect that Cycles and E-Cycles will have out-of-core rendering too before long, which will remove vram woes the very day it is in Cycles. Octane has had it for a couple years now, so Cycles can’t be too far behind…
And to be clear; when I said above that one could go a gen or two older to put more cash into GPUs, I was talking about high-end stuff, like server/workstation mobo, CPU like i7-5960X, etc…
Oh, I just thought of something else… guys, for any of you who are new-ish to this monkey-business of GPU workstations, there is one thing that is very, very important that must be a factor when building a GPU beast. That, is PCI-E lanes. In the GPU render game, having your GPUs on 4x is often considered the very minimum on workstations, but folks are mostly shooting to have all GPUs running at 8x if they can, and the holy grail is to be 16x on all GPUs.
i7-5960x has 40 PCI-E lanes. i7-5820k has 28 lanes. I have 5820k. My GPUs are running at 8x each (8 lanes each), and my m.2 nvme is using the remaining four lanes… works out perfect for me. 40 lanes would be even sweeter. But I’m doing just fine with 28.
Although, to be clear (again), pci-e speed relates to how quickly GPUs are pumping data through them, not how fast they are rendering.
For gpu rendering 4x is more than enough. The only time bandwidth is needed is during the scene loading on the gpu, i.e. the first seconds before rendering starts. 4gb/s is good enough for scene loading.
I have the 5960X. I don’t think it’s a bargain to purchase one for 300$. Waste of money. Think that the new 3700X is at ~ 320$ right now and it’s much better in any scenario.
And I am telling you from experience that when you are rendering animations bro, especially animations that are asset-heavy, but render relatively quickly (like a couple seconds each frame), that extra 4x is an effing godsend… especially with deadlines.
CPU performance will always be behind the GPU, however the memory limits is what is killing large scene renders. I’m talking Pixar levels that would rather be sure their scenes render then get “out of memory” message.
For us regular folk, 8GB is “sufficient”, with 16GB being “future proof”
NVME or M.2 are considered the fastest of the drives.
SATA SSD’s (small 2.5 inch boxes) have max speeds of 550MB/s.
NVME SSD’s (small packs of gum size) have max speeds of nearly 5000MB/s (5TB/s) (though most boards will permit max 3.5TB/s. Unless you go with AMD’s Ryzen platform.
x4 lanes would be a factor if TS was willing to get up to 4 gpus for rendering. But that’s not the case here. His budget allows up to two gpus without serious compromises. For more than 3 gpus the difference in cost would be so great that changing the platform wouldn’t be an issue imo. And remember, the budget is in AUD$.
In case of demanding animations and tight deadlines a cloud based rendering wouldn’t be the best option? I’m just asking, since you have the experience on that.